"Uncounted Votes in Cuyahoga County" by R. H. Phillips. How many??

I'm not sure I agree with their numbers. In another thread, Chaos in Cuyahoga, a lessor number seems the consensus. At the beginning of that thread, this issue was new to me. I must acknowledge those who made me aware of it. Also, Joe Knapp did a lot of work putting the spreadsheet together. Here's the other thread this summary is extracted from:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=166913&mesg_id=166913&page=

<snip> from RH Phillips article on non-votes:

"Brian Julin provides this analysis: 'Without considering ballot order (rotation) problems, which are obvious in several precincts as there are aberrant third party votes, and which may be less than obvious in collocated precincts where Bush and Kerry votes were swapped, if all undervotes were allocated in proportion to votes that were validly counted, Kerry would pick up approximately 7935 votes and Bush would pick up approximately 2675.' " </snip>

There are some difficult to analyze aspects to the non-vote issue. I hope this summary is useful.

jmknapp post: ."some precincts are so heavily Kerry that any Bush votes switched to Badnarik/Peroutka would not be noticeable. For example, if 100 random votes were switched in a 95% Kerry (inner city) precinct, 95 Kerry votes might go to Bush, 5 Bush votes to Badnarik--no red flag."

AirAmFan post: "Since polling places in pro-Kerry areas tended to be overcrowded much more often than polling places in pro-Bush areas, the vast majority of miscounted votes would have been intended for Kerry, though some intended for Bush doubtless wound up being miscounted too."

L. Coyote post:

Statewide, punch card non-votes ... is near 2 percent. If E-Voting is a true baseline..."

"These two cases illustrate the winners and losers..."

"In the first precinct Kerry loses 93 or less to Bush's loss of 9 or less. In the second case Kerry losses up to 24 to "disqualified" while Scrubya loses up to 4 to Badnarik."

"The numbers of each ballot type allocated to locations.
Sorted by the number of ballot types per location.

	5	4	3	2	1	Total

bBKdp	9	47	85	115	31	287

BKdpb	10	46	99	102	31	288

KdpbB	11	49	90	101	35	286

dpbBK	10	47	94	87	49	287

pbBKd	9	49	97	98	31	284

They are not evenly distributed!"

jmknapp post: "there should be no association with cluster size or any other factor. I would think that the distribution in Coyote's table should be even. It's hard to argue I think that a particular order should be favored in any particular grouping."

L. Coyote posts: "if Kerry and Bush are in the same position on two orders at one location, cross-precinct voting will switch their votes. So, basically you want cross-precinct voting to occur where your opponent has the highest support. If 90 percent are voting Kerry, for every 9 votes switched from Kerry to Bush, 1 vote switches from Bush to Kerry. (Anyway, that's when both precincts cross over equally. If you can get just one precinct to cross over, say by shorting just that precinct of equipment for example, the equations change dramatically.)"

"With only two ballot types at a location, if Bush and Kerry swap votes in one position, from their other positions Kerry and Bush votes go to Badnarik and Peroutka.

b B d K	p --- b	B d K p	
d K p b B --- K p b B d "

I found this to be the county wide pattern:

#	Ballots	Count
10	5	1
9	4	1
7	5	1
7	3	1
6	5	2
6	4	3
6	3	1
5	5	4
5	4	22
5	3	6
4	4	28
4	3	26
4	2	6
3	3	106
3	2	43
3	1	2
2	2	176
2	1	18
1	1	134

"Here is the precinct count and average number of non-votes. The higher the Kerry percentage, the more non-votes there are for those precincts.:

		> 90	80-90	80-60	<60

Count 		106	59	183	155

Non-Vote	12.11	10.03	7.69	5.91

"The mean number of non-votes per precinct sorted by ballot numbers is:

Ballots		5	4	3	2
	
Non-Vote	4.47	5.50	7.14	8.33

"The sequence of the candidates is constant. So I reduced the possible combinations to just four, the number that the second ballot shifts in relation to the first ... I'm applying the term "crawl" to this.

Irrespective of which candidate is in front of the line, this is a "4 crawl"

b	B	K	d	p
B	K	d	p	b

The lower row has crawled four space to the right. It doesn't matter which one is top or bottom either. ... Likewise with "crawl 1." Crawl one is the same as crawl four if you switch upper and lower and sort columns to ascending order!! The same can be done with crawls 2 and 3."

"The actual vote numbers give a good idea of the relative weight of the categories: 'the number of ballot types at a location.'

	Kerry	Badnarik  Bush	    Non-Vote	Peroutla

5	10,163	   38	   8,682        219	 19

4	58,382	  256	  43,684      1,308	 230

3	125,631	  505	  63,090      3,329	 422

2	145,473	  719	  58,922      4,200	 798

1	 55,263	  165	  18,108      1,633	 132

All	394,912	 1,683	 192,486     10,689	1,601 

The totals are off because I lack the ballot order at several precincts."

Iceburg post: "Only a hand recount of the punch cards combined with comparison to poll signature books is likely to (but not with absolute certainty) tell the truth. What we have now, is mounting circumstantial evidence ...not absolute proof."

jmknapp post: "Ran a simulation of random vote-switching to see what happens in Cuyahoga County

* pick one of the 591,348 votes cast at polling places
* pick another precinct at the same polling place to switch it to
* if the order is the same, ignore
* if the order is different, tally result

When done a large number of times in the simulation, it turns out that Kerry loses 0.55 vote on the average for each vote switched. Bush loses 0.17 vote for each vote switched. .....

Since Badnarik got 1,886 votes countywide, if all were due to random switches, an estimate of the total switched would be 1886/.24 = 7,858 ballots."

L. Coyote post: "...there are only Two Patterns of Switching. ... The "crawl 1 and 4" pattern swaps Kerry and Bush votes both ways, and benefits Badnarik one way and disqualified the other way.

The "crawl 2 and 3" pattern has Kerry and Bush votes going to the other 3 positions.

The big payoff for Bush is the "crawl 1 and 4" combination in 95% Kerry precincts...."

Enough for an intro to this issue I hope. A lot of discussion went into distilling these understandings. Praise the hive mind!

I'll place more info in a separate post following, then return to discussing the number of cross-votes, the important question.

Read the Phillips article.

L Coyote posts: "We know that the pattern of machine distribution is not even. And the different ballot combinations do not produce equal switching either. Did their placement in key precincts skew the pattern?"

jmknapp post re his simulation: "Assumptions. Since I selected "ballots" at random from the 500,000-odd universe, the switched ballots would automatically reflect the Kerry percent.
And the precincts that had larger number of votes would thereby suffer more cross-voting."

L. Coyote post: "How many votes would Badnarik have gotten in Cuyahoga?
Certainly, less than his state-wide percentage. The state-wide percentage incorporates cross-voting, and Cuyahoga is different than the rest of Ohio, esp. given 66% Kerry support. Perhaps one-ballot precincts areour best clue.

This Cuyahoga statistical summary might help resolve the problem.

	County percent: Cast   Kerry	Bush	Badn.   Per.   Non-vote

count		1435	1435	1435	1435	1435	1435
mean		60.0	66.6	31.0	0.320	0.274	1.804
median		62.2	64.4	33.3	0.182	0.000	1.464
skew		-1.082	0.113	-0.117	22.43	18.79	3.231
.5 trmean	61.7	65.5	32.2	0.136	0.084	1.527
stdev		9.34	17.46	18.19	1.70	1.38	1.42

One Ballot subset

count		177	177	177	177	177	177
mean		56.6	74.3	23.1	0.212	0.211	2.216
median		57.9	75.2	21.1	0.178	0.000	1.920
skew		-0.562	-0.487	0.487	1.291	11.210	1.583
.5 trmean	57.8	76.4	20.8	0.143	0.078	1.989
stdev		9.80	17.56	18.22	0.248	0.712	1.561

And now a summary of actual numbers.

One Ballot	Kerry	b	Bush	non	p				
mean		312.2	0.93	102.3	9.226	0.746
median		298.0	1.00	76.0	8.000	0.000
skew		0.212	1.241	1.598	1.394	1.918
.5 trmean	304.1	0.640	81.6	8.157	0.371
stdevp		114.7	1.055	97.6	6.706	1.104
quart 1		232	0.000	18.0	5.000	0.000
quart 2		298	1.000	76.0	8.000	0.000
quart 3		401	2.000	169	13.000	1.000

Countywide	
				
mean		275.6	1.174	134.2	7.457	1.116
stdevp		103.19	4.950	96.06	5.964	6.752

Using the one ballot group, the 0.5 trimmed mean is 0.640 votes. Given the skew, this is my best choice and may be high. Assuming honest ballots for these 171 precincts with no possible cross-voting, this multiplier yields 920 actual Badnarik votes in the precinct voting.

This is 0.153% compared to the inflated statwide 0.257%. That means about 765 cross-precinct votes for Badnarik were recorded.

This does not seem like a lot, but it has to be multiplied for its effect on the Kerry/Bush margin.

Using 0.150% Badnarik votes for all punch card votes in the state would then produce over 4100 cross-votes for Badnarik. Albeit Cuyahoga may be a skewed sample to apply to the state.

Badnarik's State Op-Scan percentage of 0.176 would mean a 3100 excess of Badnarik punch card votes statewide. So these two multipliers/estimates produce a 1000 ballot range, very rough estimating. Their mean times an adjusted multiplier (excess Badnarik/Joe's number roughly adjusted for Ohio) yields about 10,000 cross-precinct votes in Ohio. That's a really really rough estimate. Any refinements??

What's 10,000 cross-precinct votes in light of the fact that the Washington state governor's race was being decided by whose cars didn't start!! That before the hidden votes came to light and changes everything.

jmknapp post: "Well, in Cuyahoga I figured 0.4 votes loss of Kerry margin for each vote switched (randomly). So if 10,000 votes were switched statewide that would be about 4,000 votes of margin. Since Kerry is short by 118,000, not about to change anything."

L. Coyote posts: "Not all things being random impacts simulation models/complexities ...I can think of several off hand ...the correlation between high percent Kerry precincts and non-votes ... percent of Kerry vote shifts with the number of ballot orders at locations ... the uneveness of the ballot type distributions ... the number of non-votes also changes by ballot order numbers per location ... a correlation between number of non-votes and Kerry support ..."

"When the votes are converted to percentages yet another image emerges. First with the columns converted to total 100:

	Votes		Kerry	 b	Bush	Non      p
						
5	3.180		2.57	2.26	4.51	2.05	1.19
						
4	17.271		14.78	15.21	22.69	12.24	14.37
						
3	32.090		31.81	30.01	32.78	31.14	26.36
						
2	34.939		36.84	42.72	30.61	39.29	49.84
						
1	12.522		13.99	9.80	9.41	15.28	8.24
						
All	100		100	100	100	100	100

Now compare that to the rows converted;

	
	Total		Kerry	 b	Bush   Non-Vote	  p	
							
5	19,121		53.15	0.20	45.41	1.15	0.10	100
							
4	103,860		56.21	0.25	42.06	1.26	0.22	100
							
3	192,977		65.10	0.26	32.69	1.73	0.22	100
							
2	210,112		69.24	0.34	28.04	2.00	0.38	100
							
1	75,301		73.39	0.22	24.05	2.17	0.18	100
							
	601,371		65.67	0.28	32.01	1.78	0.27    100 

We are still assuming that ballots are marked by precinct and cannot be switched after they are marked. If that were not so, it would be easy to move ballots to the precinct where the punch matches your candidate and swap back an equal number with the same effect. That would explain ballots sorted by candidates seen in the recounts.

Statistical Summaries for the "Number of Ballot Orders" Subsets

		% Cast	% K	% B     % b	% p    % non    
1 order								
count		177						
mean		56.61	74.31	23.05	0.212	0.211	2.22	
median		57.85	75.16	21.13	0.178	0.000	1.92	
skew		-0.56	-0.49	0.49	1.291	11.210	1.58	
.5 trmean	57.80	76.44	20.75	0.143	0.078	1.99	
stdevp		9.77	17.51	18.17	0.247	0.710	1.56	
min		28.83	27.21	0.96	0.000	0.000	0.000	
max		86.77	96.23	71.48	1.068	9.091	11.00	
range		57.94	69.02	70.51	1.068	9.091	11.00	
								
2 orders								
count		503						
mean		58.59	70.16	27.07	0.350	0.380	2.04	
median		60.07	70.03	26.95	0.166	0.000	1.67	
skew		-0.83	-0.12	0.14	15.64	13.68	3.63	
.5 trmean	59.76	70.49	26.69	0.133	0.085	1.75	
stdevp		9.73	17.46	18.18	1.577	2.145	1.63	
min		14.59	31.37	0	0	0	0	
max		82.32	97.73	66.51	30.827	38.188	19.27	
range		67.74	66.36	66.51	30.827	38.188	19.27	
								
3 orders								
count		466						
mean		60.97	65.84	31.93	0.271	0.230	1.74	
median		63.28	64.36	33.57	0.189	0.000	1.46	
skew		33.38	27.05	27.19	37.18	37.27	37.34	
.5 trmean	62.87	64.89	33.04	0.131	0.090	1.51	
stdevp		9.15	16.97	17.64	0.901	0.730	1.30	
min		7.85	27.41	0	0	0	0	
max		77.55	100	70.18	16.294	9.113	12.50	
range		69.70	72.59	70.18	16.294	9.113	12.50	
								
4 orders								
count		238						
mean		63.60	57.28	40.79	0.462	0.219	1.25	
median		65.03	56.81	41.05	0.196	0.000	1.11	
skew		-1.59	0.55	-0.55	15.126	8.697	2.12	
.5 trmean	65.04	56.29	41.85	0.164	0.094	1.13	
stdevp		7.17	13.51	14.08	3.239	0.591	0.83	
min		28.57	30.05	0	0	0	0	
max		74.82	100	69.51	50.000	7.252	6.49	
range		46.25	69.95	69.51	50.000	7.252	6.49	
								
5 orders								
count		49						
mean		61.34	53.03	45.59	0.189	0.090	1.103	
median		63.47	52.46	46.55	0.000	0.000	1.014	
skew		-1.56	0.28	-0.27	2.007	1.753	0.481	
.5 trmean	63.03	52.69	45.95	0.095	0.008	1.067	
stdevp		6.74	8.87	9.12	0.284	0.165	0.551	
min		41.63	29.85	16.78	0.000	0.000	0.000	
max		69.69	80.82	70.15	1.208	0.648	2.653	
range		28.06	50.97	53.37	1.208	0.648	2.653	

It's late. Time to rest.

How come nationwide almost every error (outside of the accepted Margin of error)
goes to bush?

These next 2 weeks OURS.

KERRY WON get over it.

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

Just the discovery recount of a few places, as far as I know.

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

Non-votes, Badnarik votes, and Peroutka votes are indicators of cross-precinct voting. The represent a proportion of the Kerry and Bush votes that were switched.

This issue goes to the validity of the election. Recount 1,000 times and you won't fix cross-voting. The voter punched the wrong chad out.

The real questions posed by this thread are:

"How many votes were switched from Kerry to Bush, or from Bush to Kerry?"

"In what proportions did this happen, and in what proportion where?"

"What caused it to happen where it happened?"

"Is the pattern of the degree to which it occurred random of the result of premeditated actions by partsans?"

There are more questions, but this is a good start.

For example, "Why does the Kerry vote decline when there are more ballot orders at a location?" This was the question originally posed by Iceburg in his "Chaos in Cuyahoga" thread. He pointed out this correlation, and I got interested in finding the answer:



And RH Phillips points out, "In Cleveland there are 12 precincts with more than 6% of the ballots uncounted. These account for 380 uncounted votes, 8.07% of the city wide total, and 9.19% of the total ballots cast in these 12 precincts. John Kerry won them all overwhelmingly, by a margin of 12 to 1 in the aggregate."

In precincts with 95% Kerry support, if cross-voting happens in equal partisan proportions, for those voting on ballots with Kerry and Bush in the same ballot position, for every 1 Bush voter whose vote gets counted for Kerry, 19 Kerry votes get counted as Bush votes! Once I realized this, I really got interested!!

So the big, overarching question is, "Can we figure out the true ratio of the candidate's gains and losses, and what would the outcome have been in a clean election without these irregularities and/or frauds?"

Cuyahoga
http://bellaciao.org/en/article.php3?id_article=4815
http://24hour.startribune.com/login/?goto=http://www.startribune.com/stories/484/5131411.html

The first linked article reports:

"Anomalies were found. Almost all of the witnesses that I spoke with felt that the ballots were not in random order, that they had been previously sorted. There would be long spurts of votes for only one candidate and then long spurts for another, which seemed statistically improbable to most."

This is disturbing. The only rea-- moving ballots from one precinct to the next in a manner that avoids adding third party votes or non-votes. Can you think of any other reason?

Code could have been tied to ballot order in some say.

Is there a correlation with electronic voting? (And does that matter, if all votes, paper absentee ballot, punchcard, electronic, et al, end up in a central electronic vote tabulator?)

I looked at the statewide picture today re: non-votes. The pattern in Cuyahoga is found statewide for all counties. I divided the 88 counties by the median value of percent of non-votes. In the 44 counties with higher non-vote percents, Bush's percent of votes go up with the non-votes and Kerry's votes go down. The very strong correlations with Peroutka and Badnarik indicate that non-votes have a definite cross-vote component statewide. Number of switched voted remains a question.



I need to do the same for the punch-card-only counties and compare.

I too wonder if the recount workers were aware of these problems.

available online. A summary of statistical analysis of the Ohio vote. 67 KB.

The entire Ohio spreadsheet has been updated with more analysis and new charts. Now 757 KB.

Also, a brief explanation of cross-precinct voting.

First let me say, many thanks to DU minds for taking on this long-overdue analysis. Determining whether the catapillar ballot was shuffled randomly or intentionally is a key point IMO.

To speak specifically to the issue of the "disqualified" votes reported as non-votes, relying on numbers in this category can be dicey, as some legitimate non-votes do occur and are lumped in with "disqualified". Unless numbers can be extracted for "disqualified" votes via legal avenues (discovery in Moss v Bush) any argument based on these numbers are going to get hit with counter-arguments like the following:

http://jsekhon.fas.harvard.edu/papers/BlackCandidates_HerronSekhon.pdf

This is why the work on Badnarick and Peroutka votes is better.

However, in order to shore up the "disqualified" issue, let me just say that if you examine the very detailed canvas of King County, WA for a comparison, you would find the following:

In this election, at least in King County, if there was any bias in who undervoted, it was Republicans who were more likely to undervote.

Overall, there is a weak negative correlation between the proportion of undervotes and democratic support per-precinct. Overall turnout correlated positively with Kerry support.

Among Absentee/early voting (which in WA was 2/3rds of the vote) where turnout was negatively correlated with Kerry support the negative correlation between undervoting and Kerry support is stronger.

Among booth voting, where turnout was positively correlated with Kerry support, Republican undervoting was present but very weakly represented.

Although overall Democrats dominated absentee/early voting in most precincts (contrary to common wisdom) in King county, with their candidates receiving a greater proportion of their support from absentee/early ballots than Republicans did, in those few precincts where Republicans did "turn out in numbers" they tended to do so with absentee/early ballots. On the face this may seem inconsistant with the above, but that's the way the numbers work.

Do note all the correlations are relatively weak. In Massachusetts, the data is less detailed, and the correlation is weakly in the opposite direction, even in paper ballot towns (though those tend to be whiter than a loaf of wonder bread and the sample is small). Of course, in Cuyahoga, the correlation is much stronger and in fact visually evident on scattergrams without the aid of a regression line.

I'm not posting my charts and stuff now, as I'm still working on them for a KoS diary, but those are the results: King County is a real-world example of Republicans undervoting in this election, and while it only has 5% black population, compared to Cuyahoga's 27%, Republican undervote was enough to outweigh any race-based undervoting that the above link would suggest. I'm not going to comment on the merits of the above link, but even that paper puts the cap on the race-based undervote at around 10%.

RE: the issue of the "disqualified" votes reported as non-votes.

You are correct, it is a more difficult statistic to work with, to disassemble into actual and cross-voting components. We do have some useful data and facts.

With the 5 position presidential ballot, when cross-precinct votes go to a wrong candidate in one diretion, they can go to disqualified in the other direction. We can use the Badnarik and Peroutka votes to detect locations where cross-precinct voting occurs. The disqualified vote in those locations is suspect.

We have statewide statistics for the amount of non-voting in counties without punch card voting, both e-voting and op-scan.

We have statistics in punch card counties for precincts without Badnarik or Peroutka cross-voting.

So, the best we can do is inference in determining the rate of disqualified that belong to other candidates. But a reasonable estimate should be possible. This might require some complex programming and precinct level analysis. Before computers, this would have been nearly impossible. I think the formulas are already worked out. They were in the "Chaos" parallel thread.

RE: cross-state comparison. You state: "In this election, at least in King County, if there was any bias in who undervoted, it was Republicans who were more likely to undervote." I assume you mean that the rate of non-votes correlated with voting for Republicans. This does not indicate which partisans are non-voters. It does provide a probability that had they voted, they were Republicans in more cases than not. Even that is based on a false assumption. So we should avoid calling the non-votes either Democrat, Republican, or black. That's the precinct trend, not the vote. Perhaps Democrats in Republican neighborhoods are less likely to cast Democrat votes, but still won't cross thee line and cast a Repug vote???

Good point on the strenght of correlations as a useful indicator. That's a valid stat to focus on. Opposite correlations for different candidates also are useful. Note the Bush and Kerry differences in the graph above.

In a few spots, I've found "non-vote" correlating with 2004 primary Kerry performance but not as strongly with Kerry support in the general election. There's no logical reason for this correlation, at least in heavily democratic precincts, so finding out where it exists would be interesting.

But that's far and aside from the issue of the caterpillar ballot.

where and why? We know it happened. Who lost how many votes to whom? Was it fraudulent?

This is a complex problem to analyze. It takes a while just to understand the mechanics of how it occurred. The complexity of numbers of ballots at locations, possible combinations, etc. makes it difficult. Is there a programming whiz reading this thread??

Focus is everything. There are so many issues and threads, and lots of chasing rainbows or spouting of useless preposterous probabilities.

And this issue is tough enough without complicating it with comparisons to other races.

Caterpillar crawl, or "cross-precinct voting" to the neophite, is one issue. There the votes are mis-cast without the voter knowing it. Taking ballots from one precinct and moving them to another precinct, by accident or by design, should have its own name, like "cross-precinct tallying."

...but that does not make me a statistician.

What is needed? It seems you all have everything well in hand.

Imagine the complexity (IF .. THENs) for number of precincts at the location, ballot orders, possible cross-combinations, etc. 1400+ precincts in Cuyahoga, I don't know how many statewide.

I don't know of anyone working on the other counties, but perhaps these thread could model where to start looking. RH Phillips has written on other counties, but I don't have spreadsheet links. Go to freepress to find his articles.

As in any statistical analysis, simple data summaries provide a good start and point to where to further analyze.

I just sorted Cuyahoga precincts by ballot orders, then number of ballot orders at each precinct's location, then number of precincts at the location. I will compare locations with one order to with all others. I will also compare a grouping with same number of orders as precincts to the grouping with more precincts than ballot order. Comparison results soon.

If you are an Access user, you might be able to make fast progress on another county using queries.

Is fully understanding what the "customer" wants.

Are you looking for an answer to "if the distribution was
random, what would the impact be, within a margin of error?"

Or, related to that, are you looking for an answer to "are the
real results probable given the known ballot orders, or was there
a bias in which ballots cross-voted to which others?"

Complexity isn't a problem for me, I have many more
constructs at my disposal than if-then statements.

Where can you get ballot order for counties other than Cuyahoga?
(For that matter, where is the cuyahoga data coming from?)

How many votes were switched from one candidate to another or to a non-vote?

I suppose this would be best constructed with some flexible variables, some "what if" statements like, "What if Badnarik received one vote per precinct?" that could be adjusted of 0.9, or 0.8 etc.

The idea is to do it for individual precincts instead of on a county basis. We know that things were not the same in all precincts.

The number of Badnarik and peroutka votes would be a starting point to determine cross-voting rates. Assumptions might include equal cross-voting in the other direction so Bush-Kerry swaps could be estimated. There is only a 3rd party trail in one direction.

I've just updated the Ohio page. It has the link to the Cuyahoga spreadsheet. There is also an explaination of the cross-precinct impact.

I just added a breakdown of change in percent of vote by location classes.

After sorting the Cuyahoga precincts by ballot orders and by the number of ballot orders at locations, I can present the following statistical summary.

The first three rows below are a county-wide summary, with all precincts except several where ballot order is unknown. The rows are:

all is all precincts
x\x is all locations with one ballot order or the same number of precincts as ballot orders.
x\x+ is all locations with more precincts than ballot orders.

2\all is all locations with 2 ballor orders.
2\2 is all locations with 2 ballot orders and 2 precincts
2\3+ is all locations with 2 ballot orders and 3 or more precincts.

Etc., etc. with same logic.

	% K	% B	% Badn.	% Per.	% N-V	count
all	65.63	32.05	0.282	0.271	1.809	1427
x\x	67.52	29.96	0.297	0.314	1.956	970
x\x+	61.59	36.51	0.251	0.178	1.500	457
						
2\all	69.24	28.04	0.350	0.388	2.041	502
2\2	70.94	26.03	0.385	0.490	2.223	351
2\3+	65.34	32.65	0.268	0.156	1.626	151
						
3\all	65.05	32.74	0.267	0.223	1.754	465
3\3	64.60	33.12	0.279	0.251	1.794	314
3\4+	66.00	31.96	0.242	0.164	1.671	151
						
4\all	55.94	42.36	0.231	0.220	1.269	234
4\4	58.70	39.58	0.209	0.202	1.333	108
4\4+	53.55	44.77	0.250	0.235	1.213	126
						
5\all	53.15	45.41	0.201	0.101	1.159	49
1\all	73.39	24.05	0.224	0.179	2.217	177

1\all includes locations with two or three precincts, a total of 177 precincts. What this group has in common is one ballot order, no cross-ballot voting can occur at these locations. Even when cross-precinct voting occurs, the votes count the same as cast.



In locations with 2 ballot orders, Kerry's vote is 5.6% lower when three or more precincts are at the location (2\3+) than when only two precincts are at the location (2\2).

In the three ballot group, Kerry's vote is 1.4% higher when there are more precincts than ballot orders. This is the only exception to the trend of Kerry votes being lower when the number of precincts exceeds the number of ballot orders.

In the four ballot group, Kerry's vote is 5.2% lower when there are more precincts than ballot orders. Because the number of locations with 5 precincts is small, I did not subset that group.

Kerry is the overall loser when there are more precincts than ballot orders.

I wondered if the ballot orders were apportioned such that cross-votes would not favor a particular candidate. For example, are there 2 ways to turn a Kerry vote into a Bush vote at one location but only one way to cross vote for Bush, or vice-versa. This analysis seemed easier than checking 1400+ precincts individually. Now, given these results, it seems we should!

I also sorted by ballot orders. Here are those results:

bBKdp: Kerry % (x\x) less (x\x+) = 4.7%
BKdpb: Kerry % (x\x) less (x\x+) = 5.3%
KdpbB: Kerry % (x\x) less (x\x+) = 7.6%
dpbBK: Kerry % (x\x) less (x\x+) = 5.0%
pbBKd: Kerry % (x\x) less (x\x+) = 7.6%

In every comparison, Kerry vote drops when the number of precincts exceeds the number of ballot orders.

The next grouping compares locations where cross-voting can occur to locations where cross-voting cannot occur, again sorted by ballot orders. Kerry vote is lower by 6-12 percent where cross-voting can occur.
2-5\all is all locations with more than one ballot order. Precinct cross-voting can alter votes in this subset.
1\all is all locations where the precincts have a single ballot order. Cross-precinct voting cannot alter votes.

bBKdp: Kerry % (1\all) less (2-5\all) = 8.9%
BKdpb: Kerry % (1\all) less (2-5\all) = 6.3%
KdpbB: Kerry % (1\all) less (2-5\all) = 8.7%
dpbBK: Kerry % (1\all) less (2-5\all) = 8.7%
pbBKd: Kerry % (1\all) less (2-5\all) = 12.3%

For more detailed tables of these last groups, go to the online table. There the tables include non-vote percentages.

This table uses the same notation as in the parent post. The first three rows have only one ballot order with 1, 2, or 3 precincts, etc. The columns are sorts by ballot order

	bBKdp	BKdpb	KdpbB	dpbBK	pbBKd	Totals
1\1	22	23	35	31	24	135
1\2	6	8	0	15	7	36
1\3	3	0	0	3	0	6
2\2	68	74	73	66	70	351
2\3	41	26	28	18	18	131
2\4	6	2	0	3	9	20
3\3	57	70	55	64	68	314
3\4	23	19	26	17	23	108
3\5	4	5	6	11	4	30
3\6	1	3	0	0	2	6
3\7	0	2	3	2	0	7
4\4	19	19	22	25	23	108
4\5	19	20	20	18	23	100
4\6	3	4	4	3	3	17
4\9	2	3	3	1	0	9
5\5	4	4	4	4	4	20
5\6	2	2	3	3	2	12
5\7	1	2	2	1	1	7
5\10	2	2	2	2	2	10
all	283	288	286	287	283	1427
x=x	179	198	189	208	196	970
x=x+	104	90	97	79	87	457
2-5\all	252	257	251	238	252	1250
1\all	31	31	35	49	31	177

In the aggregate it may appear random, in detail there are discrepancies. Why are there so many locations with more precincts than ballot orders? How does this non-randomize cross-voting?
151 precincts at locations with 3 ballot orders and more than 3 precincts.
151 precincts at locations with 2 ballot orders and more than 2 precincts.
457 total precinct at locations with more precincts than ballot orders.
Cross-voting possible at 1250 precincts.

The same matrix as the Count matrix, this one with percentage of non-votes.

% N-V	Count	bBKdp	BKdpb	KdpbB	dpbBK	pbBKd	Average
1\1	135	2.6	3.1	2.3	2.0	2.1	2.41
1\2	36	1.7	1.8		1.5	1.7	1.67
1\3	6	2.8			1.5		2.15
2\2	351	1.9	2.8	2.0	2.0	2.3	2.20
2\3	131	1.5	2.1	1.6	2.1	1.3	1.72
2\4	20	1.6	0.9		1.0	1.5	1.26
3\3	314	1.9	2.0	1.5	1.8	1.6	1.76
3\4	108	1.4	1.9	1.4	1.5	2.0	1.63
3\5	30	1.7	2.3	2.0	1.2	1.5	1.71
3\6	6	2.6	3.4			2.8	2.93
3\7	7		1.7	1.3	0.6		1.20
4\4	108	1.2	1.6	1.3	1.2	1.4	1.33
4\5	100	1.0	1.1	1.3	1.2	1.1	1.12
4\6	17	1.3	0.8	0.7	1.4	2.6	1.37
4\9	9	1.1	2.7	1.6	1.5		1.72
5\5	20	1.3	0.9	0.7	1.5	0.9	1.08
5\6	12	1.2	1.4	1.2	1.5	0.3	1.14
5\7	7	0.9	0.5	0.8	0.8	1.4	0.87
5\10	10	1.2	1.6	1.6	0.4	1.6	1.26
all	1427	1.6	1.8	1.4	1.4	1.6	1.57
x=x	970	1.8	2.0	1.6	1.7	1.7	1.73
x=x+	457	1.4	1.7	1.3	1.2	1.6	1.45
2-5\all	1250	1.5	1.7	1.3	1.3	1.6	1.49
1\all	177	2.4	2.7	2.3	1.8	2.0	2.26

Note that the category 1\all, locations with only one ballot order-- hence no cross-votes--has the highest rate of non-votes (and the highest Kerry support).

Again, the same matrix with same shorthand as the parent post.
The Averages are for the table data, not for the aggregate precincts.

% Kerry	Count	bBKdp	BKdpb	KdpbB	dpbBK	pbBKd	Ave.
1\1	135	69.4	75.0	74.2	77.4	78.0	74.8
1\2	36	76.1	61.5		69.6	79.6	71.7
1\3	6	94.9			66.5		80.7
2\2	351	72.1	72.3	72.1	71.2	72.0	72.0
2\3	131	65.4	62.8	66.7	64.4	65.9	65.1
2\4	20	85.0	48.3		75.2	66.9	68.8
3\3	314	66.6	65.6	62.5	66.2	65.4	65.2
3\4	108	65.3	68.7	67.7	59.5	63.9	65.0
3\5	30	61.7	79.3	66.8	71.8	70.7	70.1
3\6	6	91.1	91.5			91.0	91.2
3\7	7		56.0	55.4	48.8		53.4
4\4	108	56.2	59.2	60.9	61.9	62.2	60.1
4\5	100	50.4	50.8	53.5	53.7	50.8	51.8
4\6	17	66.6	66.6	66.8	61.8	49.0	62.2
4\9	9	64.1	59.3	65.1	64.7		63.3
5\5	20	51.0	51.6	49.0	51.7	54.7	51.6
5\6	12	54.9	51.6	46.2	57.0	51.3	52.2
5\7	7	46.4	34.3	47.5	42.4	46.2	43.4
5\10	10	65.3	68.1	62.1	62.4	63.4	64.3
all	1427	66.8	62.4	61.1	62.6	64.4	63.5
x=x	970	63.8	64.0	63.7	65.1	68.6	65.1
x=x+	457	65.1	61.5	59.8	60.1	61.9	61.7
2-5\all	1250	64.1	61.6	60.2	60.8	62.4	61.8
1\all	177	73.1	71.5	74.2	74.4	78.4	74.3

The 2\2 precincts (locations with 2 ballot orders and 2 precincts) have a higher rate of non-votes (cross-voting). This large group is in 72% (after cross-voting!!) Kerry territory. the 2/2 locations give more Kerry cross-votes to Bush than 2\3 or 2\4 locations, significantly more.
The 2\3 and 2\4 groups have lower cross-vote ratios because of extra cross-over options (more precincts). They are in 65% and 68% Kerry territory. Somehow, the ballot/precinct combinations that yields more cross-votes are in higher support Kerry areas.

The pattern is pronouned. The correlation between the average columns for Kerry percent in this matrix and Non-Vote percent (in the above posted matrix) is a whopping 0.909!!!

This is not a random pattern. Who was in charge of this?

Again, the same notation as introduced in post 22. Count is number of precincts.

This is the percent of the total votes, with subtotals on the bottom and on the right.

 	Count	bBKdp	BKdpb	KdpbB	dpbBK	pbBKd	subtotal
1\1	0	1.52	1.59	2.50	2.25	1.78	9.64
1\2	36	0.42	0.51		1.00	0.47	2.40
1\3	6	0.26			0.25		0.52
2\2	351	4.43	5.27	5.26	4.64	4.78	24.37
2\3	131	2.88	1.97	1.83	1.28	1.34	9.29
2\4	20	0.39	0.15		0.21	0.59	1.33
3\3	314	3.97	4.70	3.95	4.27	4.82	21.69
3\4	108	1.61	1.39	1.78	1.19	1.57	7.53
3\5	30	0.29	0.32	0.42	0.74	0.22	1.99
3\6	6	0.06	0.21			0.16	0.43
3\7	7		0.17	0.18	0.13		0.49
4\4	108	1.33	1.46	1.67	1.85	1.63	7.94
4\5	100	1.37	1.39	1.49	1.43	1.56	7.24
4\6	17	0.23	0.27	0.26	0.22	0.25	1.23
4\9	9	0.16	0.24	0.22	0.10		0.71
5\5	20	0.30	0.28	0.31	0.30	0.28	1.46
5\6	12	0.11	0.09	0.24	0.20	0.10	0.74
5\7	7	0.06	0.08	0.12	0.06	0.05	0.36
5\10	10	0.16	0.11	0.15	0.06	0.14	0.62
all	1,427	19.54	20.19	20.38	20.16	19.73	100.00

Obviously, the number of voters in relation to ballot orders and number of precincts at locations is not uniformly distributed. Closer analysis is needed.

These matrices will paste directly into Excel.
Warning: The other matrices should to be converted to this same percentage scheme (each precinct as a percentage of the whole) before running comparative analysis. First make them apples and apples.

Here is the next dataset:

# b	Count	bBKdp	BKdpb	KdpbB	dpbBK	pbBKd	Sum
1\1	135	26	18	25	25	31	125
1\2	36	11	6		8	7	32
1\3	6	3			5		8
2\2	351	253	56	108	63	71	551
2\3	131	82	17	19	17	23	158
2\4	20	1	1		3	5	10
3\3	314	61	119	59	55	62	356
3\4	108	34	14	26	14	12	100
3\5	30	4	1	6	10	7	28
3\6	6	1	0	4		13	18
3\7	7		2		1		3
4\4	108	18	14	16	34	16	98
4\5	100	25	17	10	20	23	95
4\6	17	4	1	5	6	16	32
4\9	9	3	4	2	0		9
5\5	20	2	5	1	8	1	17
5\6	12	0	0	2	0	4	6
5\7	7	1	1	1	0	1	4
5\10	10	5	2	0	1	3	11
all	1,427	534	278	284	270	295	1661
x=x	970	374	218	209	198	188	1187
x=x+	457	160	60	75	72	107	474
2-5\all	1,250	494	254	259	232	257	1496
1\all	177	40	24	25	38	38	165

One more dataset:

# p	Count	bBKdp	BKdpb	KdpbB	dpbBK	pbBKd	sum
1\1	135	9	22	28	18	25	93
1\2	36	0	5		11	9	25
1\3	6	1			4		4
2\2	351	49	67	290	66	228	651
2\3	131	23	16	24	15	11	66
2\4	20	1	1		3	4	8
3\3	314	42	49	36	89	105	279
3\4	108	16	15	19	14	13	61
3\5	30	4	2	1	4	1	8
3\6	6	0	3			3	6
3\7	7		4	1	1		6
4\4	108	11	16	35	19	14	84
4\5	100	21	11	12	11	16	50
4\6	17	0	3	3	3	39	48
4\9	9	2	3	3	1		7
5\5	20	3	0	2	4	0	6
5\6	12	1	0	1	3	0	4
5\7	7	0	0	0	0	0	0
5\10	10	2	2	0	0	1	3
all	1,427	185	219	455	266	469	1409
x=x	970	115	159	391	211	381	1142
x=x+	457	70	60	64	55	88	267
2-5\all	1,250	175	192	427	233	435	1287
1\all	177	10	27	28	33	34	122

I almost forgot one candidate!!

	Count	bBKdp	BKdpb	KdpbB	dpbBK	pbBKd	sum
1\1	135	2562	2103	3644	2782	2233	13324
1\2	36	542	1118		2096	529	4285
1\3	6	35			464		499
2\2	351	7091	8060	7923	7765	7187	38026
2\3	131	5732	4197	3471	2526	2738	18664
2\4	20	324	443		281	1079	2127
3\3	314	7667	9111	8563	8310	9415	43066
3\4	108	3279	2453	3150	2974	3240	15096
3\5	30	635	366	836	1328	326	3491
3\6	6	23	59			43	125
3\7	7		436	476	388		1300
4\4	108	3330	3529	3902	4273	3801	18835
4\5	100	3954	3982	4162	3886	4458	20442
4\6	17	461	522	526	541	670	2720
4\9	9	321	532	432	197		1482
5\5	20	853	784	944	827	742	4150
5\6	12	278	256	766	495	283	2078
5\7	7	182	295	361	213	144	1195
5\10	10	322	199	317	131	290	1259
all	1,427	37591	38445	39473	39477	37178	192164
x=x	970	22080	24705	24976	26517	23907	122185
x=x+	457	15511	13740	14497	12960	13271	69979
2-5\all	1,250	34452	35224	35829	34135	34416	174056
1\all	177	3139	3221	3644	5342	2762	18108

In midstream I made a change in method, modifying which precincts to include in the x=x category. Those numbers change in post 27 as follows.

x=x	148	167	154	159	165	793

Before starting the candidate analyses, I separated out 1\all because it is a no cross-vote category. In all the candidate stats, the results stated in posts 29, 30, 34, 35, and 36 used the criteria x=x is the total vote numbers for 2\2 plus 3\3, 4\4, and 5\5. This subset is composed of all precincts with equal numbers of ballot orders and precincts at the locations.

In the following posts, 29, 30, 34, 35, and 36, one number changes, the count for x=x is 793. That is the count on which the data was determined. The number 970 misrepresents how many precincts are in the adjacent analysis.

I pasted the count column from before making the method change.
I have also made this correction in the summary spreadsheet.

I just got Fairfield county data, and wouldn't you know it, they listed Nader votes in their own column. And there was a little
caterpillar in the works. Now thoertically, this being Bush country, Bush should have lost votes to this... did he? Need
ballot orders to tell.

Polling location Violet Township Administrative Bldg., 12970 Rustic Dr. NW, Pickerington -- 2 precincts.

Prec                    reg     bal     L       R       D       Nader    C   Other
0510 224 VIOLET 0	733	540	0	326	174	29	10	1
0525 251 VIOLET R	606	473	29	241	156	0	43	4

Polling location American Legion Post #283, 7725 Refugee Rd. NW, Pickerington -- 4 precincts


0520 250 VIOLET Q	962	807	2	505	270	23	0	7
0500 222 VIOLET M	482	385	2	244	135	1	1	2
0485 219 VIOLET J	617	509	2	341	160	0	1	5
0465 215 VIOLET F	839	658	0	452	201	0	2	3

(EDIT: typo)

the amount of cross-vote from other candidates. In Cuyahoga County, the actual votes in the disqualified column are purposely obfuscated by recording that column as zero no matter how many votes were cast for "disqualified." Until there is a count of the disqualified votes in Cuyahoga, we remain uncertain about the true non-vote percentage.

Did someone in Cuyahoga decide that our ignorance is their bliss!! Why? This should not be covered up. Nader, speak out!!

This is useful imformation. It would be good to have this data for as many punch card couties as possible. Dear readers -- speak up if you know? And add links to spreadsheets if known.

... and fairfield is the first IIRC that actually lists nader votes.

Polling locations are listed for all the counties that use electionohio.com:

http://www.electionohio.com/boe.asp (the ones in red).

Ballot order seems harder to come by.

Here are all the possible ballot order combinations with an indicator of which precincts lose Kerry votes to Bush by cross-voting or cross-tabulating.

It gives some idea of the complexity, albeit the number of actual scenarios is less. In two ballot order precincts, there are only two scenarios for switching. But analysis of precinct data would have to detect these combinations after finding location names that match.

...and it shows why numerology isn't going to be any substitute for a precinct-by-precinct, by-hand analysis.

Note having the ballot orders, it is obvious that the first set of two precincts, VIOLET O and R, are a perfect example of 2,3crawl. The precincts were both voting at each other's booths, though only a fraction of voters crossed over. (Or the precinct marker cards were misplaced later before tabulation.) Bush would pick up some votes here if there were a way to read the ballots correctly.

Here's where things get strange. The second set of 4 precincts, VIOLET F,J,Q,M with precinct Q showing the only anomaly, votes in Nader's column.

First it is obvious that this was a one-way crawl. Voters from precinct Q were sent to the wrong booth, but voters from other precincts were not sent to precinct Q's booth. If they had been, votes would have turned up in greater numbers in Nader/Badnarick/Peroutka slots in the other precincts.

Now there are two paths by which votes can get through Nader. The first path via 2,3crawl is when Bush votes go to Nader. But if this happened, and both Democrats and Republicans were crawling in proportion to their share of the electorate, that would result in Kerry votes going to Peroutka, which did not happen.

The second way this could occur is through 1,4crawl where Kerry's votes go to Nader and Bush's votes go to Kerry. Let's look at a comparison of the 2000 to 2004 numbers:

             2000                          2004
  --------------------------    -----------------------------
  Reg   Bal     Bush    Gore    Reg     Bal     Bush    Kerry
F 674	458	334	108	839	658	452	201
J 649	478	338	130	617	509	341	160
Q 478	350	228	111	962	807	505	270
M 497	352	222	118	482	385	244	135

Bush gets a little under twice what Kerry does in Q... If 40 Bush votes did go to Kerry, the real numbers would be Bush 545 Kerry 253

Doing some math gives us some indicators as to whether this really
is what happened.

  Reg+  Turnout+     Bush+      Kerry+  KerryGain+
F 24%	15%	      35%	86%	38%
J -5%	12%	       1%	23%	22%
Q 101%	15%	     121%	143%	10%
M -3%	13%	      10%	14%	 4%

Does this jive with the rest of the county? It is hard to say. Almost everywhere else in the county, increased registration helps the KerryGain+ column -- the formula is (kerry/bush)/(gore/bush2k), while increased turnout is sporadic. However, Q is among the 7 or so precincts with extremely high registration, and for some reason that group of outlier precincts do not help Kerry quite as much. This could be the result of a very successful, if narrowly focused, Republican registration drive, or it could be some other error or fraud confounding us, or it could be that voters were hand-picked to be sent to the wrong machine, and the result is not detectable by 3rd party votes.

There appears to be one caterpillar I missed before because it is borderline:

Culumbus City:

A 0	294	268	0	0	8
C 13	235	240	0	1	4
D 2	261	406	0	3	7
F 7	358	375	0	2	9

Precinct C was split into Precincts C and F since 2000 FWIW.

You write, "...it is obvious that this was a one-way crawl. Voters from precinct Q were sent to the wrong booth, but voters from other precincts were not sent to precinct Q's booth. If they had been, votes would have turned up in greater numbers in Nader/Badnarick/Peroutka slots ..."

In one precinct, I had to wonder if someone got confused when resorting the ballots!! I was surprised by some of the one-way cross-voting. I expected a random pattern. Some other factors are at work. Precinct population numbers may be it. They are very skewed, and could move voters one-way due to booth demand. Machine number manipulation could do the same thing. And you are right to suggest people could have been led to the wrong booth. They could have been led to the right booth and given the wrong precinct's ballot too! Or the ballots can be swapped later.

Given one-way cross-voting, it's more difficult to get a reasonable estimate of the switched vote quantity.

I am more surprised by the way the ballot-order/precinct combinations are allocated. In Cuyahoga County, the precise and clear pattern of placing the combinations that yield a higher return from cross-voting in the Kerry strongholds where the proportion of yield is greater is remarkable. This is what to look for in other counties, to determine if the conspiracy was statewide.

If someone can think of some other reasonable explanation for the pattern, I'd like to hear it. I suspected this from the general statistics, before sudividing the ballot orders by number of precincts. With the two matrices of %Kerry and %non-votes, the pattern has come into focus (and their 0.90 correlation). This has to be considered in relation to how the numbers of ballot orders and precincts at a location have different yield values from cross-voting. If you understand the yield proportions, you see the allocation parallels with Kerry support pecentages. The yield proportions are now discussed in the Web page about ballot orders.

Now I'm curious about several other things. Was voter challenging targeted at specific precincts? And which Cuyahoga precincts were moved, split, recombined, whatever? Was that part of how this pattern was ceated?

He says that he was directed to the O booth and was told it didn't make any difference.

See my thread at:

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x177045

All the data in the series of posts #27 ... with distribution data is available in an Excel file, set up for easy statistical analysis.

Each data set is shown in multiple formats, with raw numbers, percent of the whole vote, percent for each location subset, and percent for each ballot order. The summary columns and rows have totals for the groupings.

The arrangment is a uniform grid pattern. Set up a statistical analysis of one group, it will copy and paste to another, etc.

The spreadsheet link. 141 KB.