The AEC published a fact sheet today with their estimates of margins for each electorate changed by the recently-completed federal redistributions.
There are a few pretty big differences between their estimates and my own, and I wanted to explain what I think is going on. The AEC were able to provide me with some of those estimates in advance to try and understand what is going on.
Firstly, I don’t have any particular insights into how the AEC transfers parcels of votes from one seat to another. I’ll explain at the end a bit about how I do this, and part of the process that I think I need to change that may have affected a couple of margins, but for most of this post I’m just exploring how you combine those votes once they’ve been transferred.
Firstly this shouldn’t be seen as any kind of criticism of the AEC. They are in a very different position to myself and as a government agency sometimes they will feel constrained to not make political judgements. They are also using quite sophisticated but old software to do their calculations and can’t just make tweaks on the go, while it’s easier for me to fiddle with my redistribution code. Although having said that, there is one big change I want to make and I haven’t found time to do it yet, so all of us have that issue sometimes.
Dilemmas with redistribution estimates mostly take place when an electorate takes in new areas. In the case of Curtin, for example, that seat has only lost territory, so there is no issue with deciding how to deal with mismatched 2CPs.
I see four different categories of seats.
Firstly, there are the classic seats. Generally these are simple enough. Two-party-preferred votes are calculated everywhere. The differences here seem to be quite small or sometimes there are no differences at all.
To be honest I am happy for there to be small differences as it shows that all of us (myself, Antony, William, the AEC) are doing our own work and should increase confidence in the overall analysis. They may also be explained by differences in how we treat postal votes and other votes without a specific geographic location, since that is the most complicated.
Secondly, there are non-classic seats where the 2CPs don’t match. Usually this is where a non-classic seat has taken in parts of neighbouring classic seats, but it can also include where two different non-classic 2CP combinations have been combined.
There are eight cases of this issue: Fowler, Goldstein, Grayndler, Kooyong, Melbourne, Nicholls, Wannon and Wentworth.
For my own methodology, I have calculated the difference between the 2CP and 2PP in the part of the seat which is comparable, and then applied that to the 2PP in the non-comparable bit. Understandably the AEC has not done this.
Unfortunately their method is a bit nonsensical. To take Wentworth as an example, Allegra Spender’s seat has added parts of the seats of Sydney and Kingsford Smith. Both of these areas are strongly left-leaning with strong Labor 2PP majorities. So there is no doubt that her margin relatively to the Liberal Party should go up from the actual 2022 margin of 4.2%. But the AEC’s calculations include the Liberal 2PP for the new area while ignoring the Labor 2PP in those areas, and thus make the margin much closer.
There are also three neighbouring seats on the north shore where every part of the new electorate had a 2CP between a Liberal and a teal independent in 2022. The AEC has decided that it is not their place to judge if these independents should be treated as interchangeable, and thus their margin calculations for Warringah, for example, don’t credit Zali Steggall with Kylea Tink’s votes. But for myself, and presumably other analysts working for media organisations, it’s an easy decision. You just add Steggall, Scamps, Tink and Boele 2CP votes together, as if they were a party.
Finally, there are three seats where the entire seat had a Labor vs Greens 2CP: Wills, Cooper and Sydney. It appears that the AEC’s system was not designed to be able to combine these 2CPs. I don’t fully understand what the system is doing instead, but in the case of Wills, for example, they have the Labor margin as much greater than my estimate. Wills has added a lot of very strong Greens areas from Melbourne that should be cutting the margin. The AEC is looking at fixing this bug but it won’t be done before the next election.
So for the avoidance of doubt I think there may be problems with the AEC’s margins for the following non-classic divisions: Bradfield, Cooper, Fowler, Goldstein, Grayndler, Kooyong, Mackellar, Melbourne, Nicholls, Sydney, Wannon, Warringah, Wills, Wentworth. In some cases the changes are quite minor but in others there are obvious problems. In particular Grayndler, Wentworth and Wills stand out.
Finally, I should mention an issue I’ve had with my methodology which I think has affected my Wentworth margin.
Right now my federal redistribution process evenly distributes each portion of votes across the area where those voters came from. This works great for election day votes but for categories like postal votes, where the whole electorate has a single undifferentiated bucket of votes, it means that I am not taking into account geographic differences across a seat. For example, if the most pro-Labor part of a seat is redistributed, ideally I would be moving a portion of postal votes that is also skewed towards the ALP. That is how I do it for state and local redistributions, but not currently for federal redistributions.
Until now I haven’t noticed big problems caused by this issue, but I think it may have led me to overstate Allegra Spender’s margin in Wentworth. The actual margin in 2022 was 4.2%. Antony Green has estimated it at 6.8%, while my estimate is 9.0%.
There is no doubt that Spender should be stronger on the new boundaries, but I think my margin is stronger because I have taken in an unskewed share of postal votes from the seat of Sydney, which includes extremely left-wing areas like Newtown and Glebe. While the area moved into Wentworth around Potts Point, Darlinghurst and Woolloomooloo do lean to the left, they don’t lean as far as the areas further west.
I will be trying to find time to modify my redistribution code before the election, but I just wanted to flag this.
Elsewhere, Antony Green has also explained why the AEC margins don’t match his figures (including a table comparing his figures to the AEC, myself and William Bowe).