Online Polling Integrity

I thought in the interests of transparency I should publish an update about this. As you might know, we’re using our online voting platform more often to guide my decision-making about a range of local issues, from footpath upgrade budget allocations, to the best location for local facilities like dog off-leash areas, to which locations we should support for proposed new footbridges.

Our online voting system has to strike a tricky balance.

Some online surveys and voting systems (including most of the ‘official’ surveys that Brisbane City Council runs) have very little effective protection against duplicate voting, meaning that survey results can easily be stacked and distorted by a few people.

Other systems go too far in the opposite direction, requiring multiple two-step authentication processes and making online voting so inconvenient that you end up with such low participation rates that the results aren’t worth relying on.

We need enough checks and balances to guard against people gaming the system and fraudulently casting multiple votes from duplicate accounts, but we also don’t want to create too many participation barriers. People should be able to jump online and vote without having to provide heaps of personal information and wading through endless forms.

In the past week or so, we noticed someone was trying to create dozens of fake accounts to cast multiple votes (apparently in the St Lucia footbridge poll). The names and email addresses were made up and the phone numbers provided were fake.

We’re now deleting all those accounts on mass (at this stage it seems like roughly 60 accounts all up), including the fraudulent votes those accounts cast, which will lead to a bit of a shift in the running tallies for the bridge polls.

Our process has been:

  • System operators use basic pattern recognition to identify a list of accounts that appear suspect
  • We send an email to each of those accounts advising that we suspect they are fake and giving the creator a chance to reply to confirm that they are actually a real individual
  • We call phone numbers to double-check whether the numbers are real and match the names provided

Once we’ve gone through this process, we suspend the accounts and hide their votes. If any of those accounts eventually get back to us and say “Actually I’m a real person, please don’t delete my vote!” we can restore the account once we’ve asked a few more questions to confirm their identity.

This helps us guard the integrity of our polls and helps us feel more comfortable about relying on the results.

If we see multiple attempts to create fraudulent duplicate accounts going forward, we might look at reintroducing two-step email authentication as an extra protection, but we’re mindful that people who feel they have high stakes in the result of a poll will always try to find new ways to skew the results, so we’re using a wide range of checks to protect integrity.

 

How much weight do we give to an online vote?

I should add that generally speaking, the more legitimate votes a poll attracts, the more weight we tend to give the result. This is proportionate to the decision in question. For example, if 200 local residents cast a vote on which park a dog off-leash area should be built in, we consider that a reasonably representative sample size for a decision of that magnitude, and will be heavily guided by the result, whereas if only 200 people participate in our online poll about the best location for a $150 million footbridge project, we will pay some attention to the result, but certainly won’t treat it as binding.

 

There are still a range of legitimate arguments against relying too heavily on online voting, including the fact that poorer people, older people, and those with less education might be significantly less likely to participate. But no form of consultation is perfect, so ultimately our approach is to rely on a wide range of engagement and feedback processes, including face-to-face forums, doorknocking, small group facilitated discussions and the more traditional phone calls and letters (in addition to online engagement) so we can get as balanced a picture as possible of public sentiment, given our very limited resources.