EDIT 6:44PM: We’ve now got new figures for both Bellevue Hill and Bondi Beach. In both cases we saw the preference flow to Kerryn Phelps increase significantly, and between the two of them Phelps’ 2CP lead has increased by almost 1000 votes. It now looks very unlikely that extra postal votes will put Dave Sharma in front.
EDIT 9:31AM: We now have the first batch of postal votes – about 5,463 votes. Dave Sharma is winning 64.4% of them after preferences. This suggests he is roughly polling enough in the postal vote to have a chance of winning – the big question is how many postal votes are left to be counted, and whether adjustments in the election-day vote might give Phelps some breathing room. It’s also worth noting that I didn’t factor in informal postal votes when calculating Sharma’s threshold to win, up from 63% to 63.6%.
ORIGINAL POST: Since I posted last night, we had some shifts in the count following the inclusion of pre-poll votes. In short, it is no longer certain Kerryn Phelps will win. We are awaiting the results of the postal votes, which could see Dave Sharma win if he does well. On the other hand, there’s also reason to suspect a counting error in two booths which may be underestimated Phelps’ vote.
I’ll run through both of these issues in this post, and try and briefly explain why I think we had problems predicting this result last night.
Firstly, let’s look at the pre-poll votes. When I clocked off last night we had received the final figures from all 35 ordinary election-day booths, and Phelps was leading on 54.4% of the two-candidate-preferred count.
Then later last night we received the figures from the four pre-poll centres. Phelps won majorities (comfortable but not massive) in Haymarket and Paddington and barely scraped by in Waverley. Sharma won big at the Rose Bay prepoll centre, which was one of the biggest. This produced an overall result of Sharma 55.1% in the pre-poll vote, 9.5% better than he did on election day.
So, the postal votes. According to the AEC they have sent out just under 12,900 postal vote applications. But not all of these will be actual votes – some of these people will have used a different method to vote. At the Bennelong by-election, 77% of PVAs turned into actual postal votes, which translates to almost 10,000 votes.
If there are 10,000 postal votes remaining, and no others (in reality there could be maybe 400 provisional and absent pre-poll votes to be counted as well), then Sharma would need 63% of those votes to win.
Could that happen? That would be 8% better than he did on pre-polls, and 17% better than on election day. It seems hard, but the pre-poll trend does give weight to the theory that Phelps gained ground in the final days of the campaign, in which case you’d also expect her to do badly on postal votes. Another theory for the gap between the election-day and pre-poll vote could be orthodox Jewish voters choosing to avoid voting on Saturday, which would also play out in the postal vote. The Liberal Party generally does better on postal votes, and major parties tend to have better machinery for getting out the postal vote.
And then there’s the second wrinkle.
Kevin Bonham last night noticed that, when you plot the two-candidate-preferred vote against the primary vote for Kerryn Phelps, there are two booths where the 2CP is much lower than you would expect.
I’ve calculated how much of the vote from the 14 other candidates flowed to Phelps over Sharma at each booth and compared it to her primary vote to produce this chart:
Those two booths stand out pretty clearly. Vaucluse also has a surprisingly low preference flow, but it’s also her worst booth on primary votes. It’s also a place where Labor and Greens made up a relatively low share of the vote flowing as preferences. So while I think Vaucluse is worth checking, I don’t think it’s necessarily wrong.
Bondi Beach seems like a particular concern. 79% of non-Phelps-Sharma votes were from Labor or the Greens, yet only 57% of preferences flowed to Phelps.
If you adjust Phelps’ preference flow to 80% in Bondi Beach and 70% in Bellevue Hill, she would gain 426 votes, and Sharma would lose those same votes. This shift would increase the proportion of the postal votes needed for Sharma to 67%, which seems a lot more difficult.
I think it’s also interesting to dwell on why we had this problem. Pretty much everyone agreed on calling the result (indeed I was about half an hour later than some others), but I don’t think we had the sufficient data to do so. Normally we rely on matching booth swings to previous election results, but we didn’t have any past 2CP data for Phelps. Despite some references to a “projected” 2CP, the 2CP figure was just the raw data from the booths reporting so far.
So while Phelps did well early on, and has clearly gained a tremendous swing, we didn’t really have any way to judge whether the trend was high enough to stay in majority territory, and we still don’t really.
There may also have been some unique factors around the Rose Bay booth, which did not exist in 2016. There was no local prepoll booth in the harbourfront area in 2016, and it appears that this pre-poll booth locked up quite a lot of Sharma’s strong harbourfront voter base.
I’ll keep following this story and keep you updated with edits to this post.