Last week I wrote about the dilemma of how to estimate a two-party-preferred vote in Bradfield. An actual two-party-preferred count has not been conducted between Liberal and Labor in that seat, and may not be conducted for some time. At the time I attempted to assess the likely two-party-preferred vote, but I expected that a more accurate calculation could be done once the distribution of preferences (including the three-candidate-preferred count) was published.
Since then the AEC has given us a lot more information about the situation. The AEC has produced their own estimate of the two-party-preferred vote, but they’ve also shared with myself and some others the three-candidate-preferred data for Bradfield, making it possible to make my own estimates.
The main issue is that the AEC does not feel like they can make certain assumptions about voting trends from other elections or from other seats, and this has led them to using a simpler formula that seems to significantly understate the Labor 2PP. The AEC estimate has the Liberal 2PP as 54.95%, while my estimate is more like 50.6%.
So the first thing to note is that the AEC does not plan to conduct an actual two-party-preferred count while a court case may be pending.
It’s also worth clarifying that some of the AEC’s IT systems are quite old and need to be operated in the way they were intended to be used, which sometimes requires the following of procedures that aren’t logical. This issue also arose when the AEC estimated 2CP margins for redistributed electorates – the AEC’s systems didn’t seem to be able to handle multiple neighbouring seats with Labor vs Greens margins, even though it was logically possible to just add those figures together.
Once the vote-counting process has finished, the AEC’s results system transitions to a post-election archive. Some of the results data can only be published in that post-election archive version, including distribution of preferences data, and presumably preference flow data too. But they can’t make that transition until the two-party-preferred count is finished.
So it is necessary to make an estimate of two-party-preferred votes for Bradfield before that process can be concluded (and thus provide us with quite a lot of other data that we currently don’t have).
On Monday, the AEC held a briefing for a number of psephologists to explain their plans for producing this estimate. One key point they raised is that they didn’t want to base their estimate on historical trends in other elections or in other seats.
So the AEC’s formula starts with the three-candidate-preferred count, the second-last round of the distribution of preferences. So far, so good. I think any estimate should start from that position, since it limits the field to just three leading candidates, with roughly equivalent options in a wide range of seats. It is then a question of calculating what proportion of Boele’s 3CP votes flow to Labor or Liberal.
The AEC’s formula calculates how many preferences flowed to the Labor and Liberal candidates from other minor candidates. In the case of Bradfield in 2025, Labor gained 6,136 votes between the primary vote and the 3CP count, and the Liberal Party gained 4,082 votes. Labor thus gained 60.1% of those preferences. The AEC then applies that formula to the Boele 3CP votes to produce a Labor-Liberal 2PP. While I have used the seatwide numbers, the AEC applies these figures at the polling place level – something which seems to produce very slight differences in the final figures. Where I have used the AEC’s formula in this post, I have performed the calculations at the booth level.
The issues with this logic are immediately obvious. It is firstly dependent on the pool of votes for other parties. If a right-wing minor party runs and gives their preferences to the Liberal Party, that will improve the Liberal position in the formula, even though it may have no relevance to how teal preferences flow. These votes split three ways between Labor, Liberal and independent, but those flowing to the independent are ignored. Logic would suggest that teal voters are more likely to preference Labor than Labor’s proportion of votes from minor candidates when the teal remains an alternative option. And that is indeed what happens when you apply this formula to 2022 results in similar seats and compare it to the actual 2PP figures.
There is a substantial gap between the real figure and the AEC formula in every seat, ranging from 3% in Curtin to 16.3% in Kooyong. The AEC formula always underestimates the Labor 2PP, often by quite a bit. I can’t conduct this analysis on 2025 results in those other seats, because we don’t yet have the 3CP counts for these seats.
So now we get to the actual data that the AEC has kindly shared in advance of publishing the data themselves. They have an interest in their process being explained and people understanding what the data actually shows, and why they have published it. Yesterday afternoon, the AEC shared the 3CP figures and their estimates of the 2PP for every polling place in Bradfield with myself and other election analysts.
First, the 3CP data. This is the first seat for 2025 where we have final 3CP figures. Labor polled over one quarter of the vote at this second-last stage of the count. As a raw figure, or as a ratio of independent 3CP vs Labor 3CP, Bradfield 2025 is only comparable with Bradfield 2022 or North Sydney 2022, but we will see next week how the 3CP looks in other seats with strong independents in 2025.
Interestingly the Liberal 3CP was under 42%, yet they almost won. About 8% of the electorate preferenced Labor, then Liberal, then Boele. That makes up over 30% of Labor’s 3CP vote.
The AEC’s 2PP estimate for Bradfield 2025 is 54.95% for the Liberal Party. I think this is a significant overestimate of the Liberal 2PP, as it would have been in 2022. To explain why, it’s worth looking at one statistic – the proportion of voters who end up with the independent in the 3CP who preference Labor over Liberal. The AEC’s estimate implies this figure is 59.2%. Yet every one of these similar seats in 2022 had a rate of at least 68.8%, and in Kooyong it reached 80.1%.
The figure was 70.3% in North Sydney and 72.6% in Bradfield. I would expect the 2025 figure would end up somewhere in this range, in the low 70s.
If you apply the Bradfield 2022 figure, you get a 2PP estimate of 50.6% for the Liberal Party. I will be using this figure for the time being. Next week we will be able to calculate the equivalent figure for seats like Mackellar and Warringah for 2025. If there is evidence that this figure has substantially shifted for similar seats, I would change my Bradfield estimate.
This would mean that there was a 5.53% 2PP swing in Bradfield, which would rank 39th in the country for the biggest 2PP swings to Labor.
This also means that the national Labor 2PP would end up as 55.25%, which would be a 3.12% national swing.
Interestingly it would also mean that the Labor 2PP and the Independent 2CP in Bradfield were almost identical. I think this makes sense considering the large Labor 3CP, and the substantial numbers of Labor votes that leaked to the Liberal Party on the 2CP.
All of this is a bit unfortunate, that we don’t have real figures to work with. But there is a history of psephologists having to make estimates. Psephologists like Malcolm Mackerras had to estimate 2PP figures prior to 1983 for all seats where the distribution produced a winner before it reached its conclusion. Indeed I am planning estimate independent 2CP figures for seats where independents narrowly missed out on the final count when we have the 3CP figures next week.
It is understandable that the AEC wants to avoid making political assumptions when doing these estimates but I think they would be far more robust if they did. Ideally the AEC wouldn’t be involved in making estimates, but the nature of their IT system forces them to do so. Hopefully they can improve their systems before the next election to avoid these problems. If data is simply unavailable due to court cases, it should be possible to leave it blank, and leave it to analysts like myself to make estimates.
This also means that I will need to add the 2025 federal election to my data repository. In the past I have only included state and local data because the quality of the AEC’s data is so excellent. But I will want to produce my own 2PP data at the booth and seat level for Bradfield, and combine that with the AEC data for other seats to produce files that can be used in the same way as past AEC files. I will also want to produce my own file for 2CP seat-level figures, as there are a number of seats where the AEC’s 2CP swing figures are wrong. But this is not at the top of my list.
So that is my latest figure. Liberal 2PP in Bradfield of 50.6%, and a Labor national figure of 55.25%.
I found the Bradfield result quite strange this election. In 2022 I predicted that if a “teal” wanted to win, they had to get their first preference above Labor, and push the Liberal candidate below 45% – and every candidate who did that won. In 2025 Boele was above Labor, and even combining Kapterian and Yin’s vote the Libs were at 42.16%. It feels like Boele should have won by several percentage points – not come down to the wire.
Given the fact that if we follow the trends in 2025, it seems like Labor actually wins the seat… It would be a cruel irony for Labor to have won the seat more convincingly then Boele in spite of the ‘tactical voting’ narrative. The Labor to Teal preferences don’t seem as disciplined as the Teal to Labor preferences compared to every other teal seat. This is of course, pending on if they *actually* decide to count the 2PP after the court challenge.
Great work Ben, and thanks for the quick turnaround. You deserve more donations…
Whilst this is just speculation, I think the IND 3CP flow to Labor 2PP will be slightly higher than in 2022. Firstly, the mood to vote against Dutton seemed even higher than that to vote against Morrison in 2022 and this seems evidence in many overall preference flow to Labor being higher than in 2022 – refer Kooyong (I think you had a chart of this is a previous post).
Also, there must be some relationship between this flow, and the proportion of IND 1 votes that where in the IND 3CP and the IND 1 votes would flow higher to Labor than others (perhaps with the Greens being an exception). I haven’t done the numbers but I would guess the IND 1 votes in Bradfield 2025 was a high proportion of the IND 3CP. Its higher than North Sydney 2022. It wouldn’t need to be much higher than 72.6% for the 2PP result to flip.
It is striking how similar the Bradfield 2025 results are to the North Sydney 2022 results – on primary and on 3CP. It’s the change in preference flows that make the difference in 2PP and 2CP. Whilst the official AEC 2PP result for North Sydney was Labor 48.7% there remains good evidence that there was a large error in the Willoughby PPVC and the result may well have been around 49.5%. All this shows that a small difference in preference flows from 2022 in Bradfield could increase Labor 2PP beyond 50%
Another notable stat is that the IND to Labor gap on primaries was 6.7%, and on 3CP was 6.8%. Such a low increase (0.1%) has never occurred in Teal history. I believe lowest has been 2% in North Sydney in 2022 (I am possibly missing some where the primary vote gap was around 15% and therefore the increase to 3CP margin are hardly worth thinking about).
This means some may be surprised that Labor managed to get from a 20.3% primary to 25.8% on the 3CP and may make them reconsider estimates of Labor’s 2PP
That’s a good point about the pre-poll 2PP error in North Sydney in 2022. That would increase that preference flow in North Sydney.
Indeed it would increase the preference flow to 72.6%, exactly the same as Bradfield!
I have to say that I find this a very unfortunate call by the AEC. They now have a highly misleading number on their tally room page which I’m sure will mislead a number of individuals interested in the 2PP. It would be better to simply not have a 2PP at all published, or failing that, to produce a “dummy” result that purely and clearly is only for the purposes of archiving rather than an intended “estimate”. I also wonder if the “estimate” means the AEC will feel the need to count the 2PP even after the court case window passes – hopefully they will do so.
Thanks Ben for that calc. I didn’t mention it thinking it would impact that North Sydney 70.3% value but I was intrigued as to why it was lower than Bradfield 2022.
Just noting that the rate would have to increase from 72.6% to pretty close to 74.5% for Labor 2PP to crack 50%. We will have to wait for all the similar 3CP’s next week and hopefully you can find time to update the post or add a comment as to how it changes (or doesn’t change) your view.
I trust – similar to Adda’s concerns – that this work by yourself convinces the AEC that they have a highly likely to be quite erroneous Bradfield 2PP figure in the election archived dataset and that when the appropriate time comes, they do an actual count and if their system does not allow the archived version to be over-written/updated, then they issue a press release with the actual count figure. The general public looking at the AEC website may not notice it, but election analysts like yourself will be able to correct your own datasets with the actual result and bring that to readers attention in the future).
And the AEC should be provided funding to update their systems…
I’ve never seen you state your view definitively about the potential Willoughby PPVC 2PP error, Ben.
Do you think it is an open and shut case of a preference transposition error that was not detected? Makes one wonder how many other errors could be in the 2PP results that aren’t the 2CP result. The AEC never responded to e-mails or twitter messages saying they might have an error and could they look into it. Perhaps it’s not an error, but it would be good to know if it’s just some amazing random outlier.
To be honest I don’t think much about it. I think I looked into it once years ago and it seemed really obviously a mistake. But it hasn’t been worth the bother to factor it into my analysis since the error is baked into the AEC data.
Yes, and I don’t blame you – but that’s the shame of it. Its baked in and there doesn’t seem to be an AEC process where interested observers can alert the AEC to the potential of an error and they can look into it and close it out one way or the other..
It seems odd it publicly noteworthy commentators and analysts such as yourself and Kevin Bonham think there’s likely an error, but nothing at all gets done about it…