Electronic counting

Electronic counting is being used at a number of counts at this week’s elections across Scotland, England and Wales.

London Elects 2012 (the independent team that organises teh Mayor London and London Assembly elections) produced a detailed – ie, eight minutes long – overview of the electronic counting process. While awaiting confirmation of Boris’ margin of victory, I know some Slugger readers will be keen to increase their political trivia … particularly since MLAs, Electoral Commission and EONI staff are observing electronic counts as part of an investigation into the suitability of e-counting for Northern Ireland elections.

, ,

  • Reader

    The system looks fine for voting with an X. I guess that it will struggle far more with PR voting (I have some experience with batch scanning and OCR). In that sort of count, almost every paper will have to be verified by a data entry operator. However, once that is done, the final count will run like lightning.
    Therefore it will be no fun watching a count unless the system keeps a running, recalculated, result as each batch of papers is released to the count. I’m not sure a returning officer would be willing to put on a show like that.

  • Reader

    Oh – and I suppose the obvious check is to let candidates look at all of the scanned images and challenge any data entries that look dubious, up to a certain time limit. Particularly unreliable data entry operators can be dislodged if too many of their entries are overturned by the returning officer.

  • Comrade Stalin

    Electronic counting of votes is wrong.

    There is no way that it can be done without compromising the transparency of the poll. Vote counting is a unique kind of problem in that you have the anonymity of the poll and therefore have no way to properly verify that the voter’s original vote was recorded correctly.

  • Reader

    Comrade Stalin: Vote counting is a unique kind of problem in that you have the anonymity of the poll and therefore have no way to properly verify that the voter’s original vote was recorded correctly.
    How’s that? – the electronic record of each vote can be checked against the original ballot paper over and over again by every interested or disinterested party.

  • Comrade Stalin

    How’s that? – the electronic record of each vote can be checked against the original ballot paper over and over again by every interested or disinterested party.

    If you’re going to do that then what’s the point in spending $millions doing it electronically ?

  • Reader @ 6:17 pm:

    My experience is the London one.

    Three colour-coded ballots.

    No problems on OCR for the Mayorality — two separate columns for first and second preferences. Elsewhere a simple X-marks-the-spot. No folding of ballots (wow! does that save time at the count!). Insert face-down into slot of ballot box (looks like a five-ream pack of A4). I was in and out of the polling station before the W7 bus due in 3 minutes arrived.

    On the other hand … I gather that Alexandra Palace was not the only power cut. Could that be because all the media types plugged in simultaneously? And,of course, for the second election in succession the scanning machines went down serially. Mid-afternoon, my local polling-station was clocking 40-50 voters an hour (in a middle-class ward of over 7,000 adults, 74% of whom turned out in 2010).

    The most remarkable cock-up of the count was courtesy of BoJo’s “highly-professional” team. They simply didn’t understand the second-transfer rules. Apparently, armed with the raw data, they informed Tory Central that they expected the final result to be as close as 300 votes. This was dutifully fed to the reptiles of TV and the press, who took it on face value (like every other utterance from the Sublime Porte), despite Livingstone and Labour saying, no: BoJo’s home and dry by 60-80,000.

  • Reader

    Comrade Stalin: If you’re going to do that then what’s the point in spending $millions doing it electronically ?
    Swings and roundabouts. The count will be faster, the count can be more accurate. Challenges to the count can be dealt with faster, and the result can be checked over and over again by the parties without any interested party getting their mitts on the original ballots.
    Though the video made the process look long winded and laborious, it’s probably a lot less labour intensive than the manual system, so there would be massive savings across the country for the cost of an investment in a single design.

  • Hold on, I’m worried.

    I’m actually in the same territory as Reader. Uncomfortable. Very.

    As I understand it, the London Elects system is not electronic in (say) the Ohio or Florida sense (neither of happy memory). There are hard-copy ballots. There is a definite paper-trail. In the last resort (as happened on Friday evening) it can and does all work on human-eyeball OS1.0.

    As far back as the 1970s returning officers were using counting machines (borrowed, as I recall) from the local Barclays. Not too much has changed. We’re a small-c conservative people, after all.

  • Comrade Stalin

    Swings and roundabouts. The count will be faster,

    The count is only faster if you do not need to manually verify the ballots.

    Which then leads to the question – what are the criteria when deciding whether to accept the electronic count and therefore whether or not to perform a manual count ? If a candidate raises his hands and says “I object” what do you do ?

    the count can be more accurate.

    This is solving a problem that doesn’t exist. There are no serious challenges to the accuracy of manual counts at the moment.

    Challenges to the count can be dealt with faster

    How ? A computer can’t do the same thing twice and get different results, so if the electronic count is challenged, by definition a manual count has to be performed.

    Though the video made the process look long winded and laborious, it’s probably a lot less labour intensive than the manual system,

    probably ?

    so there would be massive savings across the country for the cost of an investment in a single design.

    None of the countries that have examined electronic counting so far have been able to accomplish this feat. All they’ve done so far is spent a lot of money on equipment, found that it hasn’t worked and abandoned it. In worse cases the equipment was found to have had serious flaws. Which is the other big problem – not only do you need to be a computer scientist to understand how it works, you need to be the vendor in order to understand how the implementation has been done. None of this is good for democracy.

    Malcolm:

    There is a definite paper-trail.

    The paper trail is a red herring. Think about it.

    Why do you need a paper trail unless you do not trust the electronic count ? By extension – under what circumstances do you justify not verifying the electronic count by doing a manual count, given that everyone is agreed on the need for a paper trail ?

  • Reader

    Comrade Stalin: How ? A computer can’t do the same thing twice and get different results, so if the electronic count is challenged, by definition a manual count has to be performed.
    There are two tasks to be performed – the registration of votes into a database; and the analysis of those votes into a count and a result.
    Take the first part – the ballots are scanned onto a system, then, using a combination of OCR and human assessment, those ballots are converted into data and stored in a database. The bar code connects each database entry to an original piece of paper, and the scanned image allows each database entry to be checked by candidates and their hangers on. There are few counting clerks faster or more alert than a party tallyman. The registration of votes onto a system will probably be faster than the *first* count of a PR election, and it only has to be done once; there is no re-sorting for each round of counting. If there is a recount, then it will probably arise from some complaint that can be put right by adjusting a few ballots and hitting the big green button again.
    The second part of the count is converting the database entries into an STV result. Alliance could run their STV program against the Returning officer’s STV program, and hopefully get the same result.
    There’s nothing mystical or regressive about the need to check what has been going on. Counting clerks make mistakes already, and if they work on a computer screen, they will still make mistakes. Other people check their work now, and that will still happen in an electronic count. When an error is detected now, stacks of paper have to be resorted, quotas and surpluses and fractions and transfers recalculated, and it can take ages. But with electronic counting, a few database entries are fixed, and the counting program runs again in 10 seconds.
    If you don’t like depending on a single counting program, use half a dozen by different authors. Test them against historical manual counts, keep the winners.

  • New Blue

    I was at the Council count at Edinburgh, the software developed by Derry based ‘Opt2Vote’ ran very smoothly, indeed a number of individuals from the local electoral commission were in attendance with two of our MLA’s.

    The ballots were fed through the scanning system with no issues, questionable ballots were displayed on 50″ plasma screens giving a onlookers great entertainment as we read some very interesting reasons why individual voters had chosen not to cast their vote.

    I am looking forward to planned presentations in Belfast where this software will be shown to local political parties.

    Council elections in 2014 would be an excellent launch for using this local product to move us politically one step closer to the 20th century.

  • Comrade Stalin

    New Blue,

    What made you confident that the result of the vote was correct and not completely wrong ? How did you determine there were “no issues” ?

  • New Blue

    Comrade, the scanning system required a clerk to watch as each vote was processed, the system flagged any ballots it could not read, where this error was due to unusual handwriting etc, they were manually amended.

    The process was very transparent, and after the votes were processed the results were almost instantaneous.

    Having attended a number of counts (including the farce at the Kings Hall last year) I must say that I am, as both an anorak and a candidate, very happy with this system.

  • Comrade Stalin

    Reader:

    The bar code connects each database entry to an original piece of paper, and the scanned image allows each database entry to be checked by candidates and their hangers on. There are few counting clerks faster or more alert than a party tallyman.

    It sounds awfully like you’re saying that the votes are input to the computer at approximately the same speed as they are sorted by hand. Which is taking me back to my “what is the gain?” question.

    I accept that there is a marginal gain for an STV count as the ballot papers only have to be sorted once. But I’m not convinced that this is much of a gain, and there are other problems as I’ve described below.

    If there is a recount, then it will probably arise from some complaint that can be put right by adjusting a few ballots and hitting the big green button again.

    What sort of complaint ? The computer will not be ambiguous. It will tell you what it thinks the result is, there will be no margin of error. On what basis do you decide that a recount should be invoked ?

    You could say “the computer must be correct” but since you’re keeping a paper trail, and thereby accepting the possibility of an error, you’re exposed to legal action. In court, the judge will call upon expert witnesses, and every single computer scientist called up will repeat what he was taught in CSC101 which is that no non-trivial computer program can be certified error-free for any given input data.

    The second part of the count is converting the database entries into an STV result.

    You missed the part where we verify that the data accumulated in the database is still correct and has not been corrupted or mistakenly altered post-entry by a software bug (this stuff happens in the real world). How are we accomplishing that ?

    Alliance could run their STV program against the Returning officer’s STV program, and hopefully get the same result.

    I accept that once we have figured out a way to confirm that the database is sound, this part of the idea is fine (this provides independent verification central to a working democracy), but it falls down on practicalities. The cash-strapped political parties will now have to go out and purchase or develop software, and hire computer scientists who understand how they work. IT professionals are expensive, and the parties will have to place their faith in them, which is not easy as for any non-trivial problem computer scientists tend to disagree on what the best solution is. And they’ll have to do it every time the government changes some aspect of how the polling system works, for example if there is ever a change in the way surpluses are redistributed.

    It’s a barrier to democracy if you have to hire an expert and/or pay money in order to validate the process. The current process can be monitored by a layman.

    There’s nothing mystical or regressive about the need to check what has been going on.

    .. as long as you’re a computer scientist.

    Counting clerks make mistakes already, and if they work on a computer screen, they will still make mistakes.

    The mistakes they make tend to have a relatively predictable margin of error, which can be seen when a recount happens. We are talking about fractions of a percent.

    Other people check their work now, and that will still happen in an electronic count.

    As stated above, in the count we can (slowly) verify that the computer displays back what is scanned and we can click a “yes” or “no” button. But we’re still not confirming that the computer recorded all the votes correctly.

    When an error is detected now, stacks of paper have to be resorted, quotas and surpluses and fractions and transfers recalculated, and it can take ages.

    It very seldom ever does. Aside from the farce here last May, STV elections are generally substantially complete within 24 hours of the count beginning. In exchange for a potential gain (a gain we, for some reason, did not see in London) a number of compromises have to be made with the transparency of the count. I’ve yet to see a compelling business or political case justifying these compromises.

    But with electronic counting, a few database entries are fixed, and the counting program runs again in 10 seconds.
    If you don’t like depending on a single counting program, use half a dozen by different authors. Test them against historical manual counts, keep the winners.

    That’s an awful lot of complicated extra stuff to have to be doing in aid of solving the not-very-serious problems around the length of time required to complete a count and the accuracy of the count thereof.

    This is up with that old urban myth where the Americans spent millions of dollars inventing a pen that would work in space, where the Russians just used a pencil. What you’re doing here is working out an elaborate scheme, involving lots of technology and expense, to solve something that isn’t really a serious problem.

  • Comrade Stalin

    New Blue:

    Comrade, the scanning system required a clerk to watch as each vote was processed, the system flagged any ballots it could not read, where this error was due to unusual handwriting etc, they were manually amended.

    How do you know it was telling the truth when it said that it could read all the other ones ?

    The process was very transparent, and after the votes were processed the results were almost instantaneous.

    How do you know the results were correct ?

    Having attended a number of counts (including the farce at the Kings Hall last year) I must say that I am, as both an anorak and a candidate, very happy with this system.

    Did you ever see the episode of Bagpuss where the mice made chocolate biscuits from butter beans and breadcrumbs ?

    The votes go into a big black box with nice flashing lights on, someone presses a button, the result comes out and you accept it without even considering there to be a possibility that a software error or other bug might have yielded an incorrect result. I’m not asking people to accept a conspiracy theory, merely to accept that throughout the history of computing and IT programs have had bugs and those bugs have often had serious consequences.

  • Sorry, New Blue and Comrade Stalin, to intrude on a personal spat.

    The essential points are:
    ¶ Get the job done with all due speed (in London, this time and last, that didn’t happen);
    ¶ Have verifiable confirmation for any re-count (which means, in my terms, a paper trail).

    Beyond that, 100.00% accuracy is not guaranteed and not necessary.

    And, yes, I’ve been (not with this pseudonym) the name in the frame.

    That said, one of the few pleasures of being a (repeatedly, but not always) defeated candidate are those discounted ballots. The obscenities and other comments are a vital commentary on the whole democratic sausage-making business.

  • New Blue

    Comrade, The scanner scans the ballot, the scan appears on the screen with the voters selections suggested, the human operator has 3 options;

    1: accept the suggestions as correct
    2: amend the suggestions if not correct
    3: refer the ballot to the candidates for adjudication

    This system has been tested under ‘election conditions’ a number of times, with a range of individuals watching from a number of electoral bodies.

    For the software to be as bad as you suggest it could be would require a level of incompetence in the development and assessment that would make Bagpuss and his friends look like Particle physicists (of course, Professor Yaffle probably was).

  • New Blue

    On a side note, the reporting system is amazing if you are a psephologist. within moments of the count completed you can receive a report that shows transfer trends, number of votes that stopped after each preference and a range of other fun anorak stuff.