The great NSW iVote crash

The NSW Electoral Commission is currently in court seeking to have three council elections thrown out – not because of any failures on the part of any of the electoral participants, but because of the failure of the Commission’s own voting systems, which resulted in a number of voters failing to cast a ballot.

iVote was first used for the 2011 NSW state election as a method of casting a ballot online, but it was limited to a small number of voters with a particular need for an alternative method of voting. It was widened in 2015, with the numbers of voting increasing sixfold, but was still a relatively small share of the total vote. I remember conversations with NSWEC staff who pitched it as a replacement for long-distance postal voting, pointing out how many ballots cast by overseas voters never made it back in time to be counted.

Over 6% of votes cast in 2015 were via iVote, and this dropped to about 5% in 2019. Nothing to sneeze at, and a bigger share than postal voting.

iVote was first introduced for NSW council elections in 2021, and the eligibility was significantly expanded. Importantly, it was available for use by any voter finding themselves outside of their council area.

At state and federal elections, you can cast an absent ballot anywhere in the state. That is not the case for council elections, so iVote created the ability to cast an absent ballot online for the first time.

The iVote procedure requires voters to register for the service. They then receive credentials that allow them to cast a ballot remotely. It’s not necessary to cast this vote on the day, but on election day demand spiked. I was on a polling booth in Parramatta and saw more than a few voters realising they couldn’t vote at the booth and that iVote was their only practical option.

Registration for iVote theoretically closed at 1pm on election day, but the system struggled badly on the day. Many voters attempted to register before the deadline but failed. They either failed to register, or they registered but did not receive credentials in time to vote. I suspect some also had their credentials but didn’t manage to vote.

According to this piece from immediately after the election, the number of people voting via iVote had increased from 234,401 at the 2019 state election up to 652,983 by 1pm on election day. This number appears to have reached 671,657 votes, or 15.6% of the total vote, in the final figures.

It’s not surprising in hindsight that demand for iVote shot up, and that it was concentrated on election day. The concept of voting on the internet is easy and convenient and will be attractive to a large proportion of the population, so the main thing limiting its spread is limiting who is eligible. Giving access to anyone outside of their local council was going to massively increase the potential pool of online voters. Apparently the NSWEC tested their system for 500,000 votes, but that seems naively low.

And with low awareness of council elections, and certainly little awareness that there are no other ways to cast a vote outside your council area except via iVote, so it’s not surprising that registrations surged in the final hours.

The NSWEC quickly came out and said that anyone who had tried to vote using iVote and did not manage to vote would not be fined, but that didn’t resolve the issue, since those were still legitimate attempts at votes that were not counted. They have also stopped using iVote until further notice.

But the more interesting problem is how you deal with close election results where the number of failed iVoters may have been more than the margin of victory.

Margins of victory tend to be small in NSW council elections, particularly for a final seat. Most candidates may win decisively, but if the last seat is close then there is a potential issue.

Close elections tend to bring greater scrutiny on every part of the electoral process. I suspect a federal electorate decided by less than, say, five votes, would likely end up with a new election, because it’s hard to see how you avoid there being enough problems with voting to potentially change such a close result.

The NSWEC came up with its own analysis which identified three council races where the risk of a change in result was particularly high. These races were Kempsey, Singleton and the A ward of Shellharbour. They weren’t necessarily the races with the smallest number of votes deciding the final seat, although they had a very small number, but the NSWEC’s modelling found they had the highest probability that the missing votes would have changed the outcome.

The NSWEC found that there was a 61% chance of a change in Kempsey, where there was a number of close margins in the distribution of preferences. It also suggested a 43% chance in Singleton, where the final margin was three votes, and a 7% chance in Shellharbour A, where the final margin was four votes. They also found odds of a change under 1% in Hay, Kiama and Rosehill ward of Parramatta.

Thus the NSWEC decided to bring those three council results to the NSW Supreme Court and ask for the election to be voided.

This approach has drawn criticism from mathematician Vanessa Teague, who has been a long-term critic of iVote more generally, but also had alternative theories about which council results were compromised by the iVote failure. Teague has provided evidence to the court but also published analysis online.

Firstly, Teague points out that the pool of potential lost votes is greater than what has been suggested. The NSWEC appears to have only considered those who successfully registered with iVote but did not receive the necessary credentials to cast a ballot.

But there are also two other categories: those who failed to register, and those who received their credentials but were not able to vote. It is obviously harder to quantify these people, but they exist, and they make the pool bigger.

Secondly, the NSWEC analysis assumes that the pool of lost iVoters cast their ballots in a similar way to the remainder of the electorate, but there’s no reason to assume this. There is a lot of history to suggest that voters who vote using different methods tend to cast their votes differently. It also seems likely that there are demographic differences between people who were able to go out and vote in person after their iVote attempt failed, compared to those who were not able to do so.

None of us know exactly how these unsuccessful voters would have voted, but they could well have favoured particular candidates.

When Teague applied a more extensive definition, she identified 39 races where the number of identified unsuccessful voters exceeded the margin of victory.

So how do we deal with this situation?

While Teague is technically correct about the potential scale of the problem, I think it is reasonable to limit the remedy to the most serious cases – the ones where the ratio of unsuccessful votes to margin of victory is the most severe. It is not practical or sensible to void dozens of council elections.

The Supreme Court is currently considering options after hearing the case on Tuesday and Wednesday.

The NSWEC is asking for three council results to be voided. A number of the councillors from those councils have argued that the NSWEC should reimburse them for campaign costs if an election needs to be re-contested, with others saying they plan to retire if the elections are voided.

The point of this blog post is not to assess the likely outcome of the court case, but whatever happens we need to think about what happens next with iVote.

I don’t necessarily think that it’s not possible for iVote to work within our system, but I think the scale needs to be limited. It can’t just be a tool of convenience. It should be limited to people with disabilities that make voting via paper ballot difficult, and people who do not have a reasonable prospect of getting to a polling place, such as those outside of the state or living in remote locations. If you had two weeks to cast a pre-poll ballot and just happen to be outside your local area on election day, that shouldn’t be a reason to cast an iVote.

I don’t think we should have an electoral system where most people vote online. I think it’s good for our democracy if voting remains embedded in a local place (whether electronic or paper-based) and that online voting should only be used as an option when that isn’t possible. This requires that online voting is seriously limited to those who not only can’t vote at a booth, but did not have a realistic chance of doing so (either on the day or at pre-poll). I am not convinced that the problems with security and verification are serious enough to make it impossible to solve, but I do think it should be limited to a small share of the total vote.

The failure of the iVote on election day seems to have been primarily an issue of scale – they couldn’t handle the numbers of voters who attempted to use the service in a short time period. They do need to increase capacity to deal with these surges, but I think they should also bring forward the registration deadline to give time to resolve any problems that occur. A deadline one or two days before election day would be very sensible.

As for these potentially voided council elections: I think there could be a case for voiding elections where the probability of a change in the result was high, but I don’t think it can be justified to do so in a wide range of councils.

Kempsey and Singleton councils also made the argument that only the last-elected councillors should have their election voided, and special elections be held for that single seat. Sadly I don’t think such an approach is viable. The councillors are elected as a package deal, and a single-seat re-election would be a very different contest. In this scenario, the last-elected candidate may have only won a few less votes than the candidate before them, yet will face another election where they need to win a majority of votes across the council, not just one quota.

Whatever happens with the current court case, and whatever form iVote takes in 2023, I doubt we will see it expanded further, but rather limited to the most necessary cases. That’s not a bad outcome.

Liked it? Take a second to support the Tally Room on Patreon!