The body overseeing elections in the Australian state of New South Wales (NSW) has acknowledged researchers' claims of a bug in the software it uses to count votes.
The NSW Electoral Commission (NSWEC) has corrected an error detected and described by researchers Andrew Conway and Vanessa Teague, and verified by computer science academics from the University of Melbourne and the Australian National University.
The bug relates to extrapolation of voting patterns, a technique used in some Australian jurisdictions where a Single Transferable Vote (STV) system is used. Voters' second preference candidate can secure a vote if the first preference has already been elected to a chamber using proportional representation.
Counting votes under STV can be laborious, so some jurisdictions decide to just grab a random sample of votes and then use software to extrapolate results based on that sample.
That's the case in NSW local government elections (like local authorities in the UK, city or county governments in the USA). But the researchers found an error in the NSWEC's software for counting randomly-selected votes.
The researchers aren't saying the software got the election wrong, rather that it mis-counted votes and therefore reduced candidates' chances to be elected based on the random samples of votes chosen in elections for the council in the town of Griffith. But it reduced the likelihood a long way: from 91 per cent to 10 per cent.
The NSWEC has acknowledged the bug with its statement saying that "The academics' research paper discovered an error in the counting software used by the NSWEC that negatively affected the probability of the final candidate being elected in the 2012 Griffith councillor election."
"It is important to understand that even if the error had not occurred, the unsuccessful candidate may still not have been elected due to the effect of randomisation."
The bug's since been fixed, and the NSWEC says it has analysed all other 2012 local government elections and found no others in which the bug may have impacted the result.
Perhaps it should have acted sooner: the potential for randomness to impact vote-counting had already been identified by Australian psephologist Antony Green earlier this year. Green points out that counting votes using software and random samples means candidates have a probability of being elected, rather than an actual count determining the result. Green ran similar software and produced different outcomes with the same set of votes.
The researchers say the incident highlights the need for vote-counting software to be open-source so that bugs become easier to find, a suggestion that Australian courts have already rejected for national elections. They also consider a public ceremony used to generate the seed used to fire up random number generators, as a measure to boost public confidence in the process of random vote selection.
One small ray of sunshine: the NSWEC may have made an error, but records of the votes were available to make the researchers' analysis possible. ®