Using 3PP data to understand three-cornered races

I was recently alerted to a new source of data on the AEC website. The AEC has always published the distribution of preferences at the seat level, as well as a more detailed dataset showing the flow of preferences from each candidate’s primary votes to the two-candidate-preferred count. They have now published this data at the booth level. This means you can see how many votes in each booth flowed to other candidates as the count progressed. Until now, the only booth-level data we had was the primary votes and the top two count.

You could probably use this data to make some interesting maps (how the flow of Greens or One Nation preferences shifts across a diverse electorate) but that’s for another day. I’m finding it most useful now because it allows us to recalculate a three-party-preferred count for electorates that have been redistributed, and there were two in particular I wanted to look at.

The Greens were not far off getting into the top two in Melbourne Ports in 2016. While the Greens candidate trailed Labor MP Michael Danby on primary votes by 3.21%, this gap shrunk to 1.12% after preferences from minor candidates were distributed. We already knew that this primary vote gap had narrowed in the renamed seat of Macnamara following the redistribution, but until now I haven’t done a 3PP count.

The new seat of Canberra has a big question mark floating over it. The best Greens areas in the ACT were previously split between Canberra and Fenner, but the new inner-city seat (confusingly taking the name of Canberra, while the old Canberra was renamed Bean) now concentrates these areas in one seat. My primary vote calculations put the Greens 14.1% behind the Liberal Party, but with over 4% of the vote going to the Bullet Train party. I was curious to know whether that gap would narrow once those Bullet Train preferences were distributed.

The answer is below the fold:

Firstly, here are the numbers. I haven’t bothered to calculate the figures for all of Victoria, since I had to download the data one seat at a time, but it was just as easy to calculate for all three ACT seats, so that’s what I did. I’ve shown the relevant stats for the pre-redistribution and post-redistribution seats:

Melb Ports pre28.7043.7227.5851.381.12
Macnamara post28.2543.8227.9351.230.32
Fenner pre48.4133.6817.9263.8915.76
Canberra pre44.2839.1916.5358.4622.66
Bean post45.9338.8515.2258.8623.63
Canberra post44.3734.6420.9962.9513.65
Fenner post48.6935.6915.6261.8220.07

My estimate is that the gap between Labor and the Greens in Macnamara narrows from 1.1% to 0.3%. It’s worth noting that redistribution estimates are not perfect. In particular the methodology for distributing special votes could do with some work – I just distribute them in proportion to voter movements, not taking into account that the voters in the area which moved may not be representative of their whole old electorate. So a margin of 0.3% is the statistical equivalent of a tie.

The news is less good in Canberra. The 3PP gap between Liberal and Greens is only slightly narrower than the primary vote gap, suggesting the Greens didn’t benefit much from Bullet Train preferences. Of course, the Greens are a long way away from winning in Canberra, but they will presumably be putting up a much more serious fight in the seat than they have in past lower house races in the ACT, and may also benefit from the absence of a sitting Labor MP. I wouldn’t be surprised to see that 13.7% gap narrowed significantly, but it will be hard to close it completely and make the top two.

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

About Ben Raue

Ben Raue is the founder and author of the Tally Room.If you like this post, please consider donating to support the Tally Room.