Projected ALP swing | Projected TCP | ALP win probability |
Monday night
Labor had slightly the better of a second batch of postals, breaking 2945-2908 in their favour on two-candidate preferred, after those counted on election night went 4118-3721 to the Liberals.
Sunday night
Today’s counting consisted of rechecking and the addition of 338 formal votes from special hospital teams and electronic assisted voting. Of note in the former case was the correction in the Langwarrin booth that had inflated the Liberal swing there from 5.0% to 11.1% on two-party and from 7.6% to 13.7% on the primary vote. The latter figure was cited by News Corp’s James Campbell as evidence the Liberals had done better in the “richer, Tealier part of the electorate”. A similar argument by in a jointly written column for the Financial Review by Tim Wilson and Jason Falinksi, who lost their seats to teals in 2022, hangs on the slender thread of the Mount Eliza North booth — the one that gave the Liberals false hope when it was the first to report on Saturday, and which turns out to have had the biggest Liberal swing. Left unmentioned is that the other three election day booths in Mount Eliza, which each had two to three times more votes cast at them than Mount Eliza North, recorded below par swings of 1.0% to 1.6% (each of the electorate’s three pre-poll centres, including the one in Mount Eliza, swung by 4% to 5%). In point of fact, a geographical pattern to the results is difficult to discern.
End of Saturday night
The 3.9% two-party swing currently recorded against Labor in Dunkley can only be described as unremarkable. It is worse than the 1.3% average for first-term governments out of the twelve previous contested by-elections going back to 1983, but that includes some notable successes for governments in the first-flush of their honeymoons, including the 6.4% swing to Labor in Aston last year. All but two of the twelve were conducted in the government’s first year in office: in the two that weren’t, there were anti-government swings of 2.7% (last year’s Fadden by-election) and 6.1% (the Canning by-election in 2015, held days after Tony Abbott was deposed by Malcolm Turnbull). Another minor contingency is that Labor did badly out of the ballot paper draw, with the Liberal in first position and Labor last, whereas Peta Murphy was second behind an independent in 2022.
Evidence that by-elections caused by deaths are easier on the incumbent party than those caused by resignations seems to me rather thin. The average 4.8% swing in seats at by-elections caused by the deaths of government members calculated by The Australian is, by my reckoning, actually slightly higher than an overall 4.2% average in government-held seats over the same period. A linear regression analysis I conducted testing for death, disqualification, first-year and first-term effects going back to 1972 turned up no statistically significant evidence for any of them.
The Liberals’ 6.7% gain on the primary vote likely reflected reduced options for right-of-centre voters, with 7.9% up for grabs from the absent United Australia Party and One Nation. The other right contenders, independent Darren Bergwerf and the Libertarian Party (then the Liberal Democrats), were also in the field last time, and respectively made up a little ground and no ground. Conversely, the entry of Victorian Socialists meant there was more competition for the left-of-centre vote, although their 1.7% only partly accounted for a 3.8% drop in support for the Greens. Animal Justice gained 1.0%, and it seems likely Labor was able to hold level on the primary vote through net gains from the Greens that balanced out net losses to the Liberals.
Talk of a danger to Labor from apathy-driven low turnout does not seem to have been borne out. The votes of 74.2% of the enrolled voters have been counted, already in excess of the 72.5% at the Fadden by-election of last year, and likely to reach 81% after around 12,000 outstanding postals are processed. However, this will still leave it short of the 85.6% in Aston.
Election night commentary
10.34pm. The final result for the night is a 4.0% TCP swing to the Liberals from the Carrum Downs PPVC.
10.08pm. Frankston PPVC TCP result swings 4.5% to Liberal, compared with an election day swing of 3.3% and an existing Labor margin of 6.2%. Only the Carrum Downs PPVC remains to report for the night on TCP.
9.24pm. All election day booths are in on primary and TCP, leaving only the TCP counts for two of the three pre-poll centres, the results of which are easy to broadly predict. All but a fraction of the outstanding vote will consist of perhaps 15,000 postals, of which the Liberals would need about 64%, as compared with 53% from the postals already counted.
8.50pm. Carrum Downs PPVC also yields unremarkable numbers on the primary vote, and my Labor win probability now says 100%.
8.45pm. Frankston PPVC has reported on the primary vote, producing swings almost identical to election day. Nonetheless, the ALP RETAIN call is almost retracted: probability now 0.9902938.
8.42pm. There was apparently a second batch of postals – I had been assuming there would only be one for the night. David Speers says Labor is calling it. Perhaps they’re reading my blog. More likely they know what’s coming in the unreported PPVCs.
8.41pm. For what it’s worth, my Labor win probability has ticked over to 0.9920559, beyond the 0.99 at which my database starts saying “ALP RETAIN”.
8.40pm. Still nothing from the Carrum Downs (12,957 votes cast) and Frankston (12,468) pre-poll voting centres, which are the outstanding points of interest for the night’s count. Labor doing very slightly better on preferences than 2022, with two factors cancelling out: the absence of One Nation and the United Australia Party on the one hand, and the slump in the Greens on the other. With the overall major party vote up 7.6%, it has also mattered less, Labor’s likely win being built on a primary vote lead.
8.08pm. The night’s postal count is likewise a bit better for the Liberals than election day votes, but the 5.9% swing is below what the Liberals would have needed overall. Probability estimate now at 97.0% — I call it at 99%.
8.05pm. The Mount Eliza pre-poll booth is in sooner than I thought — very slightly better for the Liberals than the election day vote, but not enough to change the underlying metric. Another PPVC to back that up is likely to result in my system calling it for Labor.
7.57pm. What I expect over the next hour or so is the outstanding election day booths to report without changing the projection much. The next really interesting data point will be the first pre-poll voting centre that reports, which may or may not show a decisively different dynamic on the early vote. One of the three PPVCs, Mount Eliza, received a relatively modest 4503 votes and probably won’t take too long – I would estimate a bit after 9pm, but these things can be highly variable.
7.51pm. And now the ABC projection is stronger for Labor than mine. I think I may have been wrong in the previous update — this looks like it’s based on booth-matching of the TCP results, and not based on projections of the final primary vote and preference flows as mine are.
7.49pm. As noted, a weak result for the Greens. My “projected primary vote” bar chart records a candidate separately from “others” if they are projected to clear 6% — the Greens are poking around this level, causing their bar to disappear and reappear.
7.47pm. David Speers just cited a 3.3% swing, which would be based entirely on the two-candidate preferred count. My estimate further factors in the booths with primary votes in only — ten as compared with twenty — and has it a bit over 4%. The ABC projections being cited by Antony Green are doing something similar.
7.46pm. Twenty booths now in on the primary vote – over half the election day booths – and the projected two-party swing is settling at a bit over 4%, compared with the 6.3% the Liberals need to win.
7.45pm. If you’re enjoying the Poll Bludger’s results feature, please consider a donation through the “Become a Supporter” button at the top of the page.
7.41pm. Now it’s down to 86% Labor win probability, and the projected swing of 4.2% is now almost exactly the same as the ABC’s.
7.39pm. Further booths — now 15 on the primary vote — leave my Labor win probability estimate in the low nineties, suggesting the most recent results have been a bit below par for them.
7.34pm. Leaving my projections aside, it’s not disputed that the swings point to a 3.9% improvement for the Liberals in terms of their primary vote as compared with Labor’s, which does not suggest they will win.
7.29pm. The projection is now based on the swing in preference flows rather than my pre-determined educated guesses of how preferences will flow, and this has in fact given Labor a further boost. My system is more bullish for Labor than the ABC’s, which I’m a bit nervous about.
7.25pm. My projection is still based on preference estimates, but based on the TCP results that are in to this point, this appear to have been highly accurate. Nonetheless, my projection of the final TCP is about 1% strong for Labor than the ABC’s.
7.22pm. And I must again stress a potential weak link in the probability estimate, which is that it doesn’t account for the potential for a very different dynamic on early reporting election day and later-reporting pre-poll votes, which have been known to happen. For such reasons do I not call the result until it reaches 99%.
7.20pm. With nine booths in on primary and three on TCP, we’re past the point where I can follow which of the newly reporting booths weren’t in already. The most recent result was clearly a less good one for Labor, checking what had been their steady progress on the probability estimate.
7.16pm. Seaford North is the sixth booth in on the primary vote, and it’s swung very modest. My Labor win probability continues creeping up, now at 72%, though still well short of where you would call it.
7.14pm. Also a big drop for the Greens, who often struggle at by-elections, but perhaps not by this much.
7.13pm. Frankston Heights East now in on, I think, both primary and TCP — big 10.9% on Liberal primary vote, but Labor almost steady and 5.1% swing on TCP. Likely dynamic is that Hanson and Palmer votes who were preferencing Liberal anyway are shifting to them on the primary vote, neither being in the contest this time.
7.12pm. Second TCP result is Carrum Downs West, with a swing of about 1.5% to Liberal. A big unknown over the next two hours or so will be whether pre-polls have a different dynamic from election day results — we likely won’t see those results until about 10pm.
7.07pm. The fourth booth on the primary vote, Mount Eliza Central, is comfortably Labor’s best so far, looking like a swing in their favour on TCP — a stark contrast with nearby Mount Eliza North.
7.02pm. The ABC’s projections gel with my own, which is always reassuring from my perspective.
7.01pm. My Informal column seems to be working now, so the problem there may have been on the AEC’s side.
6.58pm. Labor’s bad Mount Eliza North result in on two-candidate preferred, a clean 10.0% swing to Liberal. That suggests a better preference flow for Labor than I was estimating, so their win probability nudges up from 50.6% to 53.0%.
6.56pm. The next two booths, Carrum Downs West and Frankston Heights East, are less bad for Labor — projection now down to the wire.
6.50pm. My system is working, except seemingly for the Informal column of the booth results table.
6.48pm. The first booth in is Mount Eliza North, and while it’s only 579 votes, I have the swing north of what the Liberals need to win, based on preference estimates.
6.28pm. An explainer of the map at the bottom of the results page. At the moment all booth locations are indicated by white dots. When the booth reports on the primary vote, it will change colour to indicate the party that “won” the primary vote there. When it reports on two-candidate preferred, the dot will be replaced by a number indicating the two-candidate preferred result there, colour-coded according to the winning party.
6.18pm. Something to consider while you wait is the electoral geography of Dunkley, which can be observed through the 2022 booth results map at the bottom of the Poll Bludger’s by-election guide. Far the strongest part of the electorate for the Liberals is affluent Mount Eliza at the southern end; Labor is strongest at the northern bayside end and Carrum Downs further inland, and somewhat weaker at Langwarrin at the southern inland end.
6pm. Polls have closed. The link at the top should be operable now and, God willing, the three charts above recording the projected swing (which uses estimates of the final primary vote shares and preference flows, based on the currently available information), the final projected two-candidate preferred result (which is the 2022 result plus the projected swing) and a calculation of the incumbent party’s (in this case Labor) probability of winning the seat based on the the projected TCP and the proportion of the expected total vote that has been counted.
5.40pm. With twenty minutes to go before polls close, welcome to the Poll Bludger’s coverage of the Dunkley by-election count. After 6pm, you will find through the above link live updated results, including full booth details in both tabular and map display (click on the button at the bottom of the page for the latter) and swing-based projections and probability estimates. The main chart displays on the top right will also be shown at the top of this post, if all goes well. This post will offer live commentary as the results come through, the first of which I imagine will be in a bit before 7pm.
Banging ’em out.
Postal ballot papers counted = 17.7k
Postal envelopes awaiting processing = 16 + whatever else comes in
Liberal 50.06%, Labor 49.94%. Liberal ahead by 22 votes.
Turnout = 83.49%
Informality = 4.13%
Current swing to Labor = -3.56%
A good strong finish for Labor! This time, the flow of late postals to Labor has been very strong.
92 postal envelopes awaiting processing.
G at 11.51 am
Down to 91. Meanwhile, here is a pointed observation from Dr Bonham yesterday:
“Labor is now very close to winning the postal count and the 2PP is now 52.71; it is likely to finish between that number and 53. The Australian published an incorrect article today referring to a 10% drop in turnout; the count does not finish until all postals that can be admitted are received 13 days after polling day. The turnout is currently 83.5% and there should be about 1% or so to come; the turnout decline will be smaller than at least 16 of the last 20 by-elections, potentially 18. Media should not publish turnout doomery articles without consulting with the AEC or someone who has a clue.”
https://kevinbonham.blogspot.com/2024/03/dunkley-by-election-live.html
The Dunkley turnout in 2022 was just over 90%.
Nothing new to add. Maybe 2 (3? 1?) small counts of postals left over the next week. Full distribution of preferences early next week?
Couple of small comments based from https://www.abc.net.au/news/elections/dunkley-by-election-2024/commentary to fill in some time.
“Caused by an election night counting error corrected by the more experienced staff conducting the check count.”
– There’s not necessarily an absolute, direct correlation between an error happening on with election night counting staff and that the errors are picked up because you have more experienced staff afterwards. There’s often a scattering of votes in the wrong pile in any number of random venues. It can be easy for even an experienced person to accidentally put a paper in the wrong pile or just look too quickly at a number when hundreds have already been before – though yes, experience does help. It’s more that you have a fresh set of eyes on it, in a slightly fresher environment with a single staff body of people who are more likely to have experience.
“The Liberal vote went down around 150 votes and Labor up 150. A bundle of 50 with the wrong top ballot is one of the most common count errors. I suspect three bundles of 50 Labor votes were tallied on the wrong pile or there was a data entry error when reporting the totals.”
– I touched on this in a couple of spots earlier. Counting errors on the night aren’t necessarily strictly counting ones; they can often tabulating or reporting ones. A primary aim is to ensure that you are able to reconcile the total numbers of ballot papers you started with to the total number in front of you from the ballot box. In this scenario, everything reconciled.
AEC bundes ballot papers in 100’s and any left over. So maybe it was a 100 and around a 50. It likely wasn’t a wrong ballot on the top – it was a whole pile/s just placed in the wrong collection of bundles, then counted by bundle. This can happen but can be picked up as these things can be looked at by more than one person. Always gotta be careful. It’s unlikely to have been a data entry error in this case; otherwise it probably would have been a whole value, such as two complete candidate figures transcribed or reported in an incorrect order.
“Or Langwarrin was check counted first because the change in vote share since 2022 looked odd.”
– Someone saw something somewhere that caught someone’s attention – perhaps it was the outlier swing. Even if it hadn’t been explicitly noticed and check counted first, it still would have been picked up whenever the check count was undertaken.
140 postal envelopes awaiting processing.
216 postal envelopes awaiting processing.
220 postal envelopes awaiting processing.
242 postal envelopes awaiting processing.
Postal returns close tomorrow.
243 postal envelopes awaiting processing.
245 postal envelopes awaiting processing.
264 postal envelopes awaiting processing.
It looks like its only you and me left here G…. The final lot of postals have also broken strongly to Labor – to the extent that Labor has gained 56 more postals than the Liberals. In this election, the late postals have been much more favourable for Labor than those counted first. Something to ponder.
That might be the end of the Dunkley count – 0 envelopes awaiting processing.
Postal TCP finished at Labor 50.16%, Liberal 49.84%. Labor ahead by 56 votes.
Final (tenative) swing away from Labor – 3.53%.
Turnout:83.72%
Informality:4.13%
Full distribution of preferences to come.