Friday, November 07, 2008

What's different - more ...

Yes -- the Longmont City special election in January used folded ballots. They could have been folded so as not to interfere with the boxes, but they were folded so that they did. I didn't hear of any issues concerning where the fold was, but I wasn't near the machines. Neal may know better.

Interestingly, the security sleeve for both the November 07 and January 08 elections would not fit back into the mailer once the ballot was completed and inserted. Voters could either tear the glued seam on the sleeve to make it fit (which is what was done in my household) or toss the security sleeve, which is what most voters did.Longmont elections officials (our clerk) and many voters that I've talked to complained about this in November. However the exact same problem was repeated in January of 08.Since I voted on the first day of early voting (my mail ballot never arrived), I have no idea what the security sleeve was sized for this time.

Perhaps the security sleeve was the source of the paper dust?

The best way to overcome the Ballot Now systems' issues with tiny particles is NOT TO USE IT. Every November this system has some kind of issue that causes terrible problems. Every year we here the same four words, "How could we know?"We can know because it does the same thing every time. Neal's tiny dot test proved that Ballot Now cannot discern a vote from an accidental pen swipe."How could we know?" makes me want a prescription for the same drugs the clerk's office is taking. Maybe I can forget the last 8 years.

Paul Tiger -- Longmont

Thursday, November 06, 2008

Looking forward

I generally concur with Al Kolwicz on his deeper analysis below. It is true that there is no adequate mechanism for information flow about voting systems, and to the extent that such a mechanism is in place it is too often scuttled by those in positions of power and influence. Their actions represented in the points below are taken at a real and tangible risk of loss of the accuracy of our elections. The problem of spurious marks is not unexpected by those of us watching closely, and has been encountered before. It is soluble by simple but time consuming human interception of any machine’s limited capability at vote interpretation.

In the future we must redesign our system (laws, rules, common practice, and equipment) with improved checks and remedies and perhaps more direct human involvement. But for now we must allow our existing checks in place to go forward, even if it takes days to interpret and count the marks on the paper ballots. We have always recognized that the benefit of having a countable record of voter intent depends on the ability and willingness to actually count it. While there are those who see benefits in simplicity and efficiency in losing the human countable record of voter intent, and there are also those who demand rapid results reporting, I see only loss of accuracy in those places.

Al says the spurious marks issue is not forgivable, but I see it only as foreseeable and some aspencts of it actually foreseen. This very real issue remains difficult to address under the currently popular presumption of the fallibility of the human counting of votes by many election officials. Naturally we should look for remedies involving an increase of human involvement in the vote interpretation process.

Harvie Branscomb

Answers to Audit questions

On Thu, Nov 06, 2008 at 10:36:46AM -0700, Al Kolwicz wrote:
>
> Neal,
>
> Does the group's audit procedure provide the facts needed to answer"

A good audit according to the Principles from the auditing experts would give good answers to those questions.

> 1. How accurate is "mis-interpreted vote detection" by the machine audit procedure that "the group" has developed?

First, the procedure is not a "machine audit", but rather an audit of voter-verfied paper records and how various parts of the whole system interpreted them.

The benefits of audits include: http://electionaudits.org/whyaudit

* Revealing when recounts are necessary to verify election outcomes
* Finding error whether accidental or intentional
* Deterring fraud
* Providing for continuous improvement in the conduct of elections
* Promoting public confidence in elections

The details of even defining a mis-interpreted vote depend on state laws, so as always in the US, local conditions will be important.

> 2. Does the percentage of mis-interpreted votes, in one or more audited contests, extend to the non-audited contests?

If you don't audit a particular contest, your information will of course be limited and it would be hard to say. But of course you'd have a lot more information than if you don't do a good audit of at least some contests, and very few jurisdictions do good audits now.

> 3. How about the non-audited machines?

In a proper audit, all ballots would be eligible for selection, and thus all machines and other elements of the election tallying process would in some sense be "audited".

A proper audit will give good statistics about ballots that weren't chosen for hand-counting and the performance of associated machines for the contest in question. The level of confidence chosen is of course important here. And note that it is often best to focus most on detecting mistakes in tight races, which would mean that smaller levels of misinterpretation in races with a wide margin wouldn't be picked up.

> 4. Does "the group's" machine audit procedure have the ability to project the total number of misinterpreted votes?

Note we have "principles", not yet detailed, specific recommended procedures.

The procedures designed by Philip Stark do a great job of determining when problems found indicate need for an escallation by doing that sort of projection.

He has a nice page of good auditing references:

http://statistics.berkeley.edu/~stark/Vote/index.htm

> 5. How about the number of mis-interpreted votes that occurred in undervoted contests?

Sure - all categories of vote interpretation must be audited.

> Thanks for work that you are doing.

My pleasure! A good canvass with tracking of ballots is also of course critical to a good audit, so your work there is also important.

Neal McBurnett

Speed is not the issue

Hi Harvie,
what is the stated capacity of the Hart scanner in ballots per min? Even the ES&S will do >200 ballots / min (with stops for ballots that go to resolution committee. So 12,000 per hour. now that is high speed. With the Hart system they might as well do hand counting...... that's your point!! ??
Joel

-------------------------------------

Joel
The Hart system is simply software which attaches to a variety of COTS (off the shelf) scanners (which is a good idea by the way).
The speed of scanning at full speed depends on the scanner. The Kodak scanners used by Eagle and I am sure the ones used by Boulder are plenty fast. This isn’t the issue. If anything they are too fast and don’t reliably handle forms which have been on people’s living room tables…
scan problems are of the following nature:
  1. Part of ballot is folded under in the stack
  2. Ballot upside down (supposedly can be handled, and hart prints ballots with a curved corner to avoid this, but more memory is required)
  3. Ballot folds help drag the adjacent ballot into the scanner
  4. Ripped ballot gets caught on the next ballot or on the scanner
  5. Bar code on ballot is slightly damaged by a mark (this rejection seems to happen unnecessarily, considering that bar codes are very resilient)
  6. Foreign material build up on various components in the scanner

The scanner we are using Kodak i610 I think is about 1/sec- plenty fast if the software didn’t screw up in between scanning the batches. (I mean the frequent need to reboot systems, struggle with user interface, etc.)

Also the Hart system allows all scanning to be done before resolving… the ballot images are all stored (inaccessibly as far as I can tell in a big database) so scanning can be done independently of either resolving or of uploading votes to memory cards. This allows a lot of flexibility in use of the equipment.

I think the point is that voting system software can actually aid in hand counting ballots accurately! If you have a scanned image of a ballot you can show a particular contest magnified from one ballot to the next (Hart does this well) and flip through the ballots at 10/second If you want looking for an anomaly. This is an excellent service to those who would check ballots for voter intent. For example, the voting system could assemble all the ballots in which it interpreted the vote to be for McCain and the human operator could actually sequentially visit all those portions of the ballots very quickly to look for anomalies. Such a voting system does not exist today. The system could (Hart doesn’t really do this very well) keep a record of the human interpretation and in principle could then machine mark the vote interpretations in a harmless location on the ballot image or conceivably even on the paper ballot itself (in a different design of system) for later auditing.

I would say that it makes sense that the ballots are being examined one by one, as I think they are in Boulder, on the screen. It is true that very light marks, almost invisible to the naked eye are presented in full black on the image, and so certain stray marks and dust are emphasized on the electronic version of the visible image of the ballot. Certainly with a Hart like system you could do very good machine assisted hand counting (machine suggested human confirmed interpretation and machine tabulation) in the same amount of time that Boulder is spending with the ballots and get probably even better accuracy. With an improved design machine, you could do it much faster, and also do it in precincts.

If the press would stop bad mouthing the speed of producing results from Boulder, we would have a better chance to make a more sensible system in the future.

What happened in Florida with the Republicans and the press bad mouthing the punch cards is an example of what we do NOT want to do again.

All of the above discussion relates for the long term only to mail-in ballots or what happens if we go to all an all mail-in election.

At precinct polls we could hand count (or machine assist hand count) our ballots much sooner because of the larger number of people involved.

Harvie Branscomb

Hold these officials accountable

Harvie,

I have a different take on your question, “Why hasn’t Boulder County Clerk Hillary Hall heard about dust problems on her Hart Scanners? I would suggest that the reason is that there is no mechanism for this kind of problem sharing to take place.“
  1. The problem of spurious marks was detected by the State’s Voting System Certification Team. A CONDITION OF USE was set up that would require special manual pre-scanning treatment of all HART paper ballots. At a hearing, we attempted to protect this condition. However, the powerful, secret, Colorado County Clerks Association lobbied, using incomplete and erroneous arguments, to protect the vendor and themselves from this condition of use. They persuaded the SOS to remove the condition of use.
  2. The architecture of all paper ballot vote counting systems certified in Colorado is fatally flawed. None of these systems provide exports of the ballot images and the interpretation of each vote on each individual ballot. Consequently, independent verification of vote interpretation is not possible – not even by the Canvass Boards.
  3. The LAT is contrived and inadequate. The procedures and materials are idealized, and not at all representative of the real world. And, in the case of Boulder County, the Clerk contrived to exclude the people most knowledgeable of the HART system from serving on the LAT team. (Not to sound like sour grapes, I was not permitted to serve.)
  4. Despite repeated requests that officials follow Colorado Statutes and Rules, the County and the State have neglected to perform “acceptance Tests” of the voting system. The system test would include ballot printers/vendors as well as every other component/subsystem.

From my perspective, the problem of spurious marks on HART optical scanners is not a surprise and is not forgivable.

Election officials should be held individually accountable for this problem.


Al Kolwicz
Colorado Voter Group

Objection to Daily Camera Editorial Staff

TO: Erika (Stulzman) of the Daily Camera Editorial Staff
(copy of a letter to Erika written through the Daily Camera web site)

I would hope you get a chance to educate yourself on the election reform topic more before writing again about how bad it might be that Boulder is slow reporting vote counts. Speed of reporting is just about the lowest criterion for quality in the election, and there is much which should be prioritized over speed. I hope you come to realize how destructive the pressure for quick reporting can be. This is an enormously complex topic and some of us are basically devoting all of our free time to trying to understand it, and testify about it on behalf of the voting public at the Legislature. I have written about what is wrong with Colorado election law, as well has testified on numerous occasions. I do hope that you will contact members of the Boulder community of election quality activists such as Neal McBurnett 494-6493 (Green) or Al Kolwicz 494-1540 (Republican) or Mary Eberle 442-2164 (Unaffiliated) or Margit Johansson 442-1668 (Democratic). All of us have been working together to improve elections.

One of the greatest challenges is to spread a better understanding of the reality of voting through the media. I just don't understand why you wrote the following:

"But the county needs to get it together. If Jefferson and Denver counties can have their votes counted in a timely manner (and they did), why can't we? A serious dialogue about scrapping our system is step No. 1. Had this been a contested election with major problems, Boulder County's mysterious "dust" would be this year's hanging chads. That is unacceptable.
And a note to anyone wanting to challenge Hall for her clerk's position because of the slow election results: That's what she said."

I don’t disagree that a discussion about scrapping the system is in order. In fact a discussion about scrapping all the systems in Colorado is in order. The technical test board of the Secretary of State actually recommended not certifying any of the voting systems currently in use in Colorado (except hand counting). Only through the catch-22 of Colorado law and with the help of our 2008 Legislature could our Secretary of State (now Congressman to be) Mike Coffman personally decide that all voting systems were ok to use. There are better voting systems which could be certified and there are certainly better systems which could be designed if state laws would only accommodate these better ideas.

It could be that Jefferson and Denver counties are not as careful as Boulder County about election quality. Have you checked? Why not investigate the truth about "hanging chads"? These were foolishly presented as the worst thing that ever happened to an American election when in reality they were an indication of a reasonably well designed voting system valiantly recording evidence of voter intent when rank mismanagement of the voting system made it very difficult for voters to vote. Rank mismanagemnet on a totally electronic voting system leads to no record of voter intent whatsoever. Think about it. That is a very good reason to use paper ballots. Paper ballots allow voters in difficult situations (and eligible voters represent a spectacularly wide variety of human characteristics not the least of which includes old age) to express themselves. Machines do have difficulty interpreting some of these marks on paper ballots, although they are quite good at adding them up, once the marks have been understood.

I would be glad to send you more information about the defects of Colorado elections and election law, and would invite you to attend the Election Reform Committee hearings on Nov 12, although these should be taken with a grain of salt considering that the Commission consists of mostly election officials who are not the ones who most need a commission to improve the election... they have considerable power in their own hands. Perhaps you could widen the scope of your writing on this topic.

FYI Eagle County votes with a similar system to that of Boulder County. I am the Canvass Board member for the Democratic Party. I have first hand experience, through testing, of the various defects and benefits of that system, although I am now sure I have not seen all the defects yet.

Please send me your email address and I will reply with more information.


Harvie Branscomb
Eagle County Canvass Board
Colorado Voter Group

Eagle County has encountered a number of similar problems with HART

Excerpt from article reproduced in full at bottom of email (referring to Boulder County vote counting delays)

To ensure accuracy, election workers are visually checking a picture of
each ballot on computers, searching for the troublesome dust line. Though
officials hope a couple of technological fixes will speed up the counting, the
rate of votes counted Wednesday hung around 1,200 ballots an hour.
“Accuracy
is more important than speed,” said Boulder County Clerk and Recorder Hillary
Hall. “People expect accuracy — this is the job I was elected to do.”

The molasses speed of the counting process is frustrating Hall, but
perhaps more confounding is that there doesn’t appear to be any obvious
explanation for why the paper-dust problem showed up now.

“I just don’t know,” she said. “If this is a common problem, I’ve
frankly never heard about it.”
The Laura Snider article is quite reasonable
in its coverage of the Boulder vote counting process.


Why hasn’t Boulder County Clerk Hillary Hall heard about dust problems on her Hart Scanners? I would suggest that the reason is that there is no mechanism for this kind of problem sharing to take place. The manufacturer has little motivation to encourage the sharing of election problems with their equipment… the CCCI (clerks association) doesn’t either, apparently, other than what happens through serendipity. The state has no mechanism for this and the election quality advocates among the public are largely shut out of the process so have difficulty learning about these problems, and also have problems having their criticisms heard once they do have information.

Eagle County has encountered a number of similar problems with Hart and Kodak systems, and has been very frustrated by the operation of this system during the past few days. Will the reports of this frustration reach other counties and other states? Probably not.

“The county needs to talk to their printer,” said Peter Lichtenheld, director of
marketing for Hart InterCivic. “Hart did not print the ballots ... and the
printer did not use Hart secure ballot stock.”
Unlike Boulder, Eagle County has its ballots manufactured by Hart. These ballots are also problematic but in a different way. They do not seem to be emitting dust particles, but they are showing evidence shadow imprinting of portions of the ballot onto other portions. In our case, fortunately, this is occurring only on the removable stub… but in past elections we have experienced transfer of ballot toner onto the active part of the ballot (on Diebold printed ballots in the past).

In addition, Hart created (I do not think these are actually, technically, “printed”) ballots are folded at the factory. These multiple folds are very difficult to straighten out and have caused immense difficulty in feeding the ballots into the scanners. On many batches there are one or more ballots which fail to scan and therefore have to be deleted from the digital records and rescanned in an adjacent batch. (Alternatively the batch must be sequentially rescanned until successful, but this proves to be too slow). In another area, the score at the top of each page which allows the stub to be removed is too weak and hence the stub is extremely difficult to remove, also delaying the scanning process and often leaving a rough edge at the top of the ballot.

Eagle County is stopping to dismantle parts from the scanners and vacuum the interior on a very frequent basis. During the tests it was found that the scanner needed extra white metal shields to be added to successfully scan 17” ballots. This information had not been provided in advance by Hart.

It was also discovered that inserting ballots bottom first caused an extra load on the ballot processing software, actually slowing the process and causing it to run out of memory, frequently. When it ran out of memory the program would not recover gracefully. This actually caused certain batches of ballots to consistently fail to scan. When all ballots were loaded head first, this problem was relieved (apparently for technical reasons related to the design of the software, but probably never encountered during State testing and in any case unknown to Hart system users). However it was then discovered that since the first fold was about ¾ of an inch from the top, the ballots were bent in a direction which, if inserted top up, would not allow the scanner roller to pick up the pages consistently. It was deemed necessary to load the ballots bottom up. None of this was advised by Hart, the voting system supplier, in advance and was all discovered in the initial scanning process of this particular ballot in this election.

Will reports of these problems reach Hart and other Hart county users? Will someone make a huge manual of gotchas which will benefit local election officials?

Doubtful.

The remaining references to attention to detail in this article are admirable.. I note that John Gideon’s 2005 report of the white line is probably not entirely what Boulder County is experiencing although it may be part of it. I suspect that there are a variety of single pixel or larger black marks on the ballots as well. It would be better if the press were able to get sufficient access to be able to report accurately on this topic.

Harvie Branscomb
Eagle County Canvass Board
(Eagle is a user of a similar system to that used in Boulder County)

How precise are audits?

Does the Audit Elections group's audit procedure provide the facts needed to answer"

  1. How accurate is "mis-interpreted vote detection" by the machine audit procedure that "the group" has developed?
  2. Does the percentage of mis-interpreted votes, in one or more audited contests, extend to the non-audited contests?
  3. How about the non-audited machines?
  4. Does "the group's" machine audit procedure have the ability to project the total number of misinterpreted votes?
  5. How about the number of mis-interpreted votes that occurred in undervoted contests?


Thanks for work that you are doing at https://launchpad.net/electionaudits


Al Kolwicz
Colorado Voter Group

How closely are other counties looking at their ballots?

The best story I've seen so far on this is from Laura Snider at the Boulder Daily Camera. Or maybe I'm biased since she called me and put in a quote or two 8-)

http://www.dailycamera.com/news/2008/nov/05/boulder-countys-counting-crawls/

From the limited information I have, it seems to me that the magnitude of the problem is relatively small and might not be noticed by counties that don't look closely at their ballots via a good audit or the sort of screen images that the BallotNow system provides (despite its many flaws, that is an interesting aspect of the Hart system).

So how closely are other counties looking at their ballots, from systems known to have problems?

Snider writes:

Last winter, Secretary of State Mike Coffman de-certified all Hart scanning equipment because the scanners "failed to count votes accurately when there are extraneous marks on the ballot."
...
Coffman's testing board recommended that he institute regulations requiring county officials to review every ballot, looking for the stray marks. Instead, Coffman chose to re-certify the machines, which are used in 47 Colorado counties, without the extra regulations.

That creates the possibility that some counties using the Hart scanners may be having dust problems and not know it.

Improved audits were required by the conditions of use imposed by the Secretary of State. But who besides Boulder in Colorado is really doing good audits of mail-in ballots?

What is your county doing?
Neal McBurnett http://neal.mcburnett.org/

More on print through

As concerns print through or transfer of one part of the ballot to another, either at the original manufacturing process or while being crushed in folded form in postal sorting machines, or stacked after receipt at the county... it is important to know that many ballot manufacturing processes are similar to laser printing... ie toner is used, and toner is subject to transfer from one place to another when heat is involved if not in other circumstances. Hart manufactured ballots seem to sometimes have toner transfer onto the removable stub on the back side, in this election.

It is not the case that ballots are necessarily printed with ink. However, it is useful to know if this is the case. I understand Boulder County Colorado ballots may have been manufactured or printed by some company other than Hart.

Harvie Branscomb

Ghost votes - print through

I've worked as a resolution judge today. The problem is called ghost
votes by the other judges with whom I worked. Changes in threshold
setting on the mark detector software seem to be making a dramatic
improvement in ths processing. But, to my untrained eye, there is
something very strange going on. The details of the images of marks
that I see don't fit well with any explanation that has been suggested
to me. I feel sorry for the people working on this problem. The environment
in which they are gathering data is certainly not a scientifit/engineering
laboratory. There a lot of constraints on the experiments they can
perform, and there is a lot of pressure to perform very quickly. But
when you discover that there is a problem with the actual count you
have to announce to the press and try to deal with their questions.
I'm sure a lot of time is being wasted dealing with crazy theories
generated by the press and public.

Part of the problem that I did see was a kind of print through of text on
the ballot onto other parts of the ballot. This was attributed to a wrong
ink used by the printer. This wrong ink remelted in the Post Office sorting
machine when the mail passed between hot rollers. (I don't think the
rollers are deliberately heated. They just get hot during normal operation
of the sorter. But I'm not sure.)

I had no problem fitting my mail-in ballot into the return-mail envelope
for this election. The envelopes were correctly sized, IMHO.

Paul Condon

Do not use “auto-resolve damaged contest”

The problem mentioned in the Gideon report is completely soluble or at least identifiable if “auto-resolve damaged contest” is not used.

In Eagle County we “auto-resolve undervotes” only. It seems that “auto resolve undervotes” is relatively harmless, if not completely so. The experience we have is that if any mark however light is found within the target the BNIP software detects either a vote or an overvote, and if there is any damage to the existing ballot target (occlusion of the target rectangle), manual resolution of a “damaged contest” is required. It is almost impossible for a foreign particle or something causing a white line to produce an undervote, unless “autoresolve damaged contest” is selected. One would have to imagine a particle of precise size and placement to occlude the entire voter mark without touching the surrounding rectangle. I would therefore recommend to operate with auto-resolve undervote selected (only). Having turned off auto-resolution of undervotes, the manual resolution of undervotes would be extremely time consuming and would not produce any differences due to human recognized voter intent. We have actually tried this in Eagle County in the early scanning of our mail-in ballots and have come to that conclusion.

On the other hand, if foreign particles, smudges or other spurious marks are causing extraneous marks on the scanned images of the ballots which are being recognized as votes, the solution would be a real hand eye scan of each contest which is actually resolved as a vote- i.e. in Hart, a contest which is not marked with a colored overlay at all- and this of course is generally the majority of all contests on all the ballot images. There is no way to select for slightly marked, or less than fully marked targets within the Hart system as far as I know. This is the function that would be required to help find the effect of spurious marks. Also of course, the “none of the above” option as a choice on the ballot would greatly help in finding spurious marks- without this the system is depending on the likelihood that an overvote will be generated to find that such a problem exists. If you know the problem already exists and is significant, then effectively, each contest must be hand counted… although it may be convenient to use the optically scanned version of the ballot to conduct this hand count and to use the hart resolution software interface to reflect and record the observed voter intent.

Having now experienced the resolution process first hand, I can say that there are some very interesting patterns of mis-voting which arise, and spurious marks certainly do create overvote situations. No doubt then, the same effects will cause voter undervoted contests to mistakenly become recognized as voted.

Harvie Branscomb
Eagle County Canvass Board

What's different?

So why is Boulder County having problems three years later?

Margit

==============================

These [November 4, 2008] ballots were huge so were folded in quarters to mail. There were tons of mailed ballots this time.

Neal, do you know whether the Longmont City Council special election used folded ballots? The Daily Camera article mentions that the Longmont ballots were created by the same vendor as was used for yesterday's election.

Mary

All it takes is a can of pressurized air to fix the problem.

Boulder Co is suffering from paper dust and dirt from the ballots. This is something that we reported in July 2005 from problems in Yakima Co WA.

http://www.votersunite.org/info/yakimaproblemreport.asp

All it takes is a can of pressurized air to fix the problem. --

John Gideon
Co-Executive Director
VotersUnite.Org

HART OpScan fails again

Daily Camera
Explanation for ‘paper dust’ still elusive
http://www.dailycamera.com/news/2008/nov/05/boulder-countys-counting-crawls/

Daily Camera
Dust slows Boulder County’s ballot count
http://www.dailycamera.com/news/2008/nov/04/election-day-begins-early-voting-may-result-shorte/

Longmont Times-Call
Boulder County - ‘Phantom’ marks frustrate election workers [Hart eScan problems]
http://www.timescall.com/news_story.asp?ID=11914

7 NEWS Denver
Boulder County - Dust, Creases Slow Ballot Count In Boulder Workers Have To Inspect Ballots [Hart eScan]
http://www.thedenverchannel.com/news/17895498/detail.html#-

VotersUnite.ORG
Hart InterCivic Optical-Scan Has A Weak Spot
http://www.votersunite.org/info/yakimaproblemreport.asp