I think those of us who have followed a lot of Australian elections have some instinctive sense of how election night progresses – when we should expect results to come in, and how quickly they do so.
But I thought I would dive into the data to actually map out the rate at which results flow in. The Australian Electoral Commission publishes a media feed on each election night, with a new XML file uploaded roughly once a minute throughout the night. This is the source of data for media outlets like the ABC and Channel 9 who run their own automated results analysis, but it has the benefit of giving us a snapshot of what the election results looked like at any point in time on election night (or the following days) at the last four elections.
While these trends may be affected by a big shift away from election day voting towards early voting in 2022, it’s still a useful starting point and an interesting way to look at how results flow.
For most of this post, I will be looking at how much of the total House primary vote has reported at any time, as a proportion of the final vote totals.
This first chart shows the rate at which House primary vote results came in for each state and territory at the 2019 election, based on Australian Eastern Standard Time (AEST).
Timezones will be the same for this election as in 2019, so this firstly gives you a sense of which states will come in when.
Very little had been reported by 7pm. Tasmania led the charge, with 9.2% at 7pm, and 55.6% by 8pm.
The three big east coast states are clustered closely together, but take longer than Tasmania. About 3-5% was reported by 7pm, which had reached 32-37% by 8pm. The remaining east coast jurisdiction, the ACT, took much longer. Barely 16% of the ACT vote had been reported by 8pm, putting them at least an hour behind their bigger neighbours.
South Australia has a half-hour delay compared to the east coast, but caught up to the big east coast states by about 8:45pm. 20% had been reported by 8pm, with 55.6% by 9pm.
The Northern Territory shares a timezone with South Australia, but it’s results come in much later. Almost 40% of the South Australian vote had been reported by 8:30pm, but but just 11.6% in the Northern Territory. The NT vote is much jumpier, likely due to the smaller number of booths and seats leading to less granular counting.
Western Australia is two hours behind the east coast, and an hour and a half behind the central timezone, so it comes in much later. By 9pm, just 3.4% of the WA vote has been reported. This is almost 40% as of 10pm, and has started to level off at about 63% as of 11pm.
The figures won’t be exactly the same this time, but some of the underlying factors should remain the same.
I’ve also made a chart showing the same information but adjusting for local timezones. This makes the results much more closely clustered.
Tasmania still has the lead, with the five mainland states following similar curves until about 8pm local time, at which point Western Australia and South Australia keep up the pace for longer while the count slows down in the big east coast states. The ACT trails behind until about midnight, while the NT starts slow but catches up around 9pm-10pm.
All of this data is just looking at the 2019 election, but now I wanted to do two things: look at the overall national picture, and compare it to the previous three elections.
The curves are very similar in shape. Very little is reported by 7pm, there is a big surge until 8pm, it slows down until 10pm, and not that much comes in after that.
But there is a gradual shift over the decade, with the pace of that early counting slowing down, and thus having less of a dramatic deceleration later in the night.
In 2010, 60% of the vote had been reported by 8:30pm. In 2019, only 40% had been reported by then, and it took until 10pm to get to 60%.
This also means the later hours have increased in importance. Only about 4% of the vote was reported between 10pm and 11pm in 2010. In 2019, it was more like 9.5% of the vote.
The charts are scaled to the final vote count, and they all end up roughly around 80-84% by 2am, although there has been a slight decline in this number.
All of this is consistent with the shift towards pre-poll voting. Pre-poll votes tend to come in later in the night, so an increased pre-poll vote should lead to a slower election night count.
In by-elections and smaller state elections we’ve seen a pattern of the election day votes coming in a rush, followed by a decent length break during which there is no new data, before the pre-poll votes come in. I don’t think federal elections have that sudden step change, because there is enough diversity of timezones and types of seats that means that the first pre-poll votes come in not long after the election day vote starts to peter out.
I suspect in 2022 we will see this curve flatten out even more, with more pre-poll voting, but also the final count at the end of the night should be lower, since postal votes won’t be counted on the Saturday.
Finally, I thought I would compare the 2019 election night primary vote curve to the two other sets of results we get on the night: two-candidate-preferred and Senate:
The other two types of data do come in later, and never reach the levels of completeness we see with primary votes.
The two-candidate-preferred count is about half an hour behind the primary vote early in the night, and that gap widens to about an hour by around 10pm.
Meanwhile we only start to see substantial counts for the Senate around 9pm-10pm, and it flattens out around midnight.
Hopefully all this will be useful in understanding the timing of election results on election night. I’m going to come back later this week with another post looking at how the votes for each of the parties change through the night, and how long they take to stabilise.
One thing I think will be interesting to see is how the ability for the AEC to open + sort pre poll ordinary votes from 4PM on election day will impact these curves. Will we see those curves go a bit more vertical in the 8-10pm slot.
It seems Lyons maybe the first seat to watch to what the night will be. Having a lot of tiny rural booth they usually count pretty quickly
@Nimalam funny thing on election night with divisions like Lyons though is that every little booth tells a different story.
@ SEQ Observer, good point this is a patch work seat so may not be able to be an early bellwether. Having that it seems both parties are campaigning hard in Lyons.
@Nimalan Sivakumar & @SEQ Observer you might find Antony Green’s speech to the 2022 Linux conference interesting. In it he explains the statistical basis for the ABC’s election night modelling. This basically analyses the booth-matched swings from the previous election. History has shown that by the time about 25% of the vote has been counted in an electorate the random (or locally influenced) swings at different polling booths cancel each other out and the ultimate result will not be very different. The actual algorithm is a bit more sophisticated than just considering the percentage counted, and it includes a statistical analysis of the probability of the final result being different to the current swing. Speech video is here: https://antonygreen.com.au/predicting-australian-elections/
Norfolk Island in Bean should be the first polling place to report, given it is an hour ahead of the East Coast, although (from memory) it wasn`t last time (even though it was only 335 ballot papers). However, as it was/is also as prepolling place, so it might need some slack.
Lord Howe Island should theoretically be ahead of all non-Norfolk Island Polling places, as its standard timezone is halfway between Norfolk Island`s and the East Coast`s and only had 215 votes last time. However again it was not and it wasn`t a prepolling place.
@canberra boy, I’ve watched his Linux conference lecture, it inspired me to make my own application with the same modelling so I could visualise the election count progression curves for all of the divisions.
@ Canberra boy agree, it in interesting lecture which explains the science behind the modelling including how booth size impact.
A truly astonishing set of graphs Ben.
Media Feed is a public site and anyone can access it.
I use Media Feed and generally accumulate about 20 of its (roughly 4MByte) XMLs throughout Election Night) on the night and then mainly a daily download. After all is wrapped up, I throw all except the latest away. Ben seems to have captured every XML!
The “Matched Polling Place” method was developed by Rod Medew, the AEC’s chief analyst honcho after the 1980 election, in time for the 1983 election and I saw it in operation from inside the National Tally Room on that rather famous night. Rod was VERY happy.
I am not sure where the AEC’s computers are now but, a couple of decades ago they were in the Parliamentary Annex. If you knew the Manager (I did), you could even sneak in there for a peek.
I haven’t saved anything Geoff! The AEC keeps all of the old media feeds up on their FTP site. Indeed before this election I haven’t really had a way to pull the data into a useable format but I’ve now developed some techniques in R so I will be able to pull what I need late on election night for analysis, although I’ll mostly just rely on the ABC website.
Sad that Media Feed still uses XML feeds on a FTP server. I hope that over the next few years, AEC intend to modernise their media feed and have a REST API available. Ideally a Webhook system for observers and media that want to consume live data directly and instantly. They can keep the XML+FTP process for legacy systems dependent upon it. But a JSON REST API, particularly a Webhook Service would be a gamechanger – this format is far more widespread in use and better supported today.
Another method I use is a cURL-scripted download for FPs, TPPs, TCPs, Senate Primaries. The downloads are small CSVs, which I drop straight into my Media Feed app for comparative purposes.
What you see on the ABC is, I think, still driven by the Media Feed.
I might say that the Media Feed is Australia’s own tweaking of EML – “election Markup Language. Wiki says this of it:
The OASIS Election and Voter Services Technical Committee, which met for the first time in May 2001, was chartered “To develop a standard for the structured interchange of data among hardware, software, and service providers who engage in any aspect of providing election or voter services to public or private organizations. The services performed for such elections include but are not limited to voter roll membership maintenance (new voter registration, membership and dues collection, change of address tracking, etc.), citizen/membership credentialing, redistricting, requests for absentee/expatriate ballots, election calendaring, logistics management (polling place management), election notification, ballot delivery and tabulation, election results reporting and demographics.
EML is just one of about a dozen separate feeds in Media Feed. The Media Feed Manual is at https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwjm5NO77tT3AhW-Z2wGHSUPDSMQFnoECAoQAQ&url=https%3A%2F%2Fwww.aec.gov.au%2Fmedia%2Fmediafeed%2Ffiles%2Fmedia-feed-user-guide-v4.pdf&usg=AOvVaw0Q8AUFLRnRLk_CyzIHtDlS
A few of us were test guinea pigs in the early years – both for AEC and a couple of the State Electoral Commissions. All pretty much use it now.
An interesting aspect of the Media Feed is the Thursday Night rehearsal which tests the system and subjects it to simulated minor or major outages …front end failure; total back-end failure. In Tally Room days, I used to go to those.
One of the deathly secrets of the rehearsal is the revelation of who the two candidates that are destined for the TCP/TPP are. One has to sign a deed poll in blood in front of a JP promising to keep the secret. It’s not worth the bother, really. In the booths, the pairing is revealed to the RO at 8PM, when she opens a sealed envelope. If later counting reveals a “bad pick”, the who TCP/TPP process starts again. This happened in Indi in 2013and caused a bit of a freak-out down there.
Comments are closed.