How scanners and PCs will choose London's mayor
Can e-counting technology get it right this time?
Very few politicians are recognisable by their first names only, but next week, two such larger than life characters will face each other in the closest battle for the office of London Mayor since it was re-established in 2000.
The polls have the Labour incumbent Ken Livingstone running neck and neck with the Tory contender, Boris Johnson. So the count, on May 2nd, will be the most closely watched in recent times. But how much can we trust the technology that will tally up the votes?
As with the previous two Mayoral races, the votes will be counted electronically, a decision which might cause some eyebrows to raise after the disastrous electronic counting in last year's Scottish elections.
While every vote matters in every election, in a close race where the outcome could swing on a tiny percentage, each vote matters even more.
Indeed, after the Scottish debacle - in which scanners discarded some 70,000 ballot papers with no human oversight, and a further 70,000 ballot papers were judged to have been spoiled - the Electoral Commission recommended (pdf, and we paraphrase) that electronic counting not be used in major elections until the kinks have all been ironed out.
Have you seen the size of my bandwagon?
Deputy Greater London returning officer John Bennett says he expects that, unlike previous years, all candidates will have a full staff of assistants and representatives at the three count centres across London, to make sure things are running properly.
A spokesman for the Electoral Commission told us that its position is still that electronic counting should not be made available at any further UK statutory elections until the Government has "undertaken a further programme of support and testing of the e-counting system".
But it specifically excluded the London Assembly and Mayoral elections from its ruling, because "we recognised that planning for the 2008 Greater London Assembly (GLA) elections is at an advanced stage".
By this, the Electoral Commission doesn't mean it was simply too late to switch to a manual count; rather, that London Elects had done a massive amount of testing and research to ensure the integrity of their chosen system.
Tick box government
"We recognise that the Greater London returning officer has taken steps to ensure the integrity of the count but the GLRO is ultimately responsible for the organisation and conduct of the election and ensuring the successful use of e-counting," the spokesman concluded.
Deputy GLRO John Bennet explains that there are some major differences between the Scottish set-up and the system that will be used in May, from the technology that will do the counting right down to the design of the ballot papers.
"In Scotland, as in 2000 and 2004 in London, the scanners scan and interpret the vote in one pass," he says. "Now we're scanning in one pass and sending data to [an] interpreter PC which matches up with templates."
If the analysis identifies a problem with a ballot paper, then it is sent for adjudication rather than being declared spoiled. This means that a scanner cannot bin a ballot paper - only a returning officer will have that authority.
Technical support was also a major issue in the Scottish elections. With the count spread over more than 30 centres, when things went wrong, it was almost impossible for the engineers to respond quickly enough. In the London elections the counting will take place in just three locations, all of which will have a staff of engineers on hand to deal with any problems that arise.
A third major difference is the design of the ballot papers. In previous GLA elections the candidates for both the Assembly and the position of Mayor were listed on the same page, meaning that voters had to cast multiple votes on one ballot. Scottish voters faced a similar set-up, which meant lots of people simply failed to vote in the election that was listed second on the sheet.
"We did market testing of the ballot paper and found that having two votes on the same paper was a major – if not the only - cause of the confusion," says Bennet. "So this time each assembly contest is on a separate paper."
So London voters will be issued with two ballots: one for the Assembly elections, and the other for the Mayoral race itself. Bennet says he expects this will reduce the number of missing votes considerably, and the number of ballots that will be sent to adjudication.
My world is your Oyster
"Sensitivity [to unintentional or confusing marks] is set very high, because of our experience in 2000," Bennet says, referring to the election where roughly one in five ballots needed adjudication.
"The processing software can identify a continuous line as a fold in the paper, but if even a speck of dust makes a mark where a vote might be cast, the ballot paper will be put in the pile for adjudication."
Keeping STVs confidential
Despite this, he says he expects the voter error rate to be of the order of one per cent for the Assembly vote. The Mayoral race will be closer to ten per cent, he says, because the voter needs to indicate a first and second preference.
"With a single transferable vote system, you have to have two votes on the same paper. Market research indicates that 80 per cent of people look straight at the names and voting boxes. Only after that, if at all, do they read the instructions at the top of the page. Additionally, in London you have the significant issue of English not being everyone's first language.
"Short of separating the vote into two rounds, we are pretty much stuck with this as a problem."
But despite all the changes, there are still those who are concerned that introducing technology into the voting process risks compromising the integrity of the vote. One such campaigner is Rebecca Mercuri, an American expert on electronic voting and counting systems.
She is worried that equipment provided by companies with little or no experience in elections could fail to provide an accurate count because of an unintentional design flaw. She cites the example of a US election where the counting system reached a certain number, and then started counting backwards.
"It's not that easy to design [a well planned counting system]," she told us.
But Bennet says he's confident that the machines will return a result that accurately reflects the will of the people, simply because of the huge volumes of test papers that have been scanned. Over the last 18 months, almost one million ballot papers have been counted on test runs, and Bennet is so sure of his system that he is prepared to claim the machines are more accurate than people.
"When we've had to do recounts because of a discrepancy between the machine and manual counts, it has always turned out that the machines are right and the people have made a mistake," he told us.
But Mercuri questions Bennet's confidence in the voting machines. Testing, she says, is only good at spotting the problems that can be forseen. Optical scanners in the US were rejecting ballots that had been marked with gel-ink pens, for example, but this wasn't picked up in testing because no one was looking for it. Only low vote totals alerted officials to the problem.
Pre-election testing is also no good at spotting machines that develop a fault, or have been compromised on the day. It might be possible, Mercuri contends, for a hidden piece of code to be activated, or for a machine to be subverted by scanning a particular bitmap image, or even by an engineer pressing a particular sequence of keys. These so-called Easter eggs are common in electronic equipment, she says, as manufacturers commonly install them to allow engineers access to configuration or diagnostic settings.
A Fujitsu spokesman told us that the scanner that will be used on the counting day, the fi-5900 "can NOT be programmed/altered based on the image 'content' of a specific page".
Despite the assurances of those involved, Mercuri remains unconvinced. The only thing to do is to run a manual sample count to check the machines as the votes are totalled.
"Basically, without audits, you have no way of knowing if there were problems. Your deputy GRO would like to believe that the pre-election test provides confidence in the election results, but actually it is just a false sense of confidence," she counters.
London Elects, the body organising the poll, has taken plenty of the steps needed to avoid a repeat of the disaster in Scotland - but it has stopped short of ordering a sample manual recount to audit the machines.
A recount would involve a statistically significant sample of the ballot papers being counted by hand, to check for any systematic counting errors in the electronic total. It would seem the ideal solution: a simple way of checking the integrity of the count that would silence all critics. So why not do it?
Bennet says that such an audit would be "meaningless" and bad for voter confidence. This is because the rules that govern the counting procedure do not allow for both a manual and electronic count.
Confident the voters will do what's right
The rules for the counting procedure for Greater London Authority elections differ from the rest of England and Wales, in that they are expected to be counted electronically. A manual count can be done, but only in exceptional circumstances. The rest of the country runs things the other way around: manual is preferred and electronic is possible in an emergency. In neither case do the rules allow for both methods to be used simultaneously.
"We could do a sample manual recount, but if it turned up a problem, we wouldn't be able to do anything about it, which would be the quickest way to collapse voter confidence in the result," Bennet told us.
This is an anathema to campaigners like Mercuri. "The law should always include some percentage of manual audit and there always must be a way that a problem with the check should trigger an investigation, possibly resulting in the discarding of the electronic totals."
And she is not the only one who thinks the electronic count should be audited. Becky Hogge, executive director of the Open Rights Group, says that ORG is campaigning for the law to be changed to make a manual recount of a statistically significant sample to be mandatory in all electronically counted elections.
Despite his confidence in the scanners, Bennet concedes there might be room for improvement in the way the elections are run. He notes that the GLRO can't actually order a recount; that can only be done on a local level, borough by borough. "It is possible that this might not be the most desirable situation," he says.
So, fingers crossed it'll all be alright on the night then.®
Polling stations open tomorrow at 7am and close at 10pm. London Elects says it will post results "as they happen" on May 2.