Author: William Bowe

I’ve been too consumed by the minutiae of the count to have accumulated any deep thoughts about the result, or even the polls. But I can recommend the assessment of Matthew Knott of Nine Newspapers, for identifying the relevance of Nate Silver’s axiom that “almost all polling errors occur in the opposite direction to commentators’ predictions”. A case in point being my own instinct that Labor couldn’t possibly doing as well as as BludgerTrack’s end-of-campaign reading of the situation, which gave Labor a lead of 53.2-47.8. My own projection of the national result currently has it at 54.2-45.8, but this is generally reckoned (in both senses) too conservative — a display at The Guardian appears to have it at around 54.6-45.4. On either reading, this is Labor’s best result since 1943, and was exceeded in this time by the Coalition only in 1966 and 1975. The primary vote, of course, is another matter — Labor’s currently stands at 34.7%, compared with BludgerTrack’s 32.6%; the Coalition is at 32.2%, by far the worst result in its modern history, compared with BludgerTrack’s 32.9%; while the Greens and One Nation are at 11.8% and 6.3%, compared with BludgerTrack’s 12.5% and 8.0%.

Not everyone agrees with me about this, but I don’t think it can reasonably be described as any sort of failure on the part of the Australian polling industry. In dismissing the notion that even its 2019 performance counted as such, Nate Silver pointed out that its roughly 3% error was exactly normal by international standards — though this rather glossed over the extent to which the industry’s failure on that occasion lay in the herding-related uniformity in the size and direction of its error. There were at least a few suspicions abroad that something similar was happening this time, and a general reluctance to believe what some polls were seeing — including some mid-campaign blowouts from Roy Morgan that were hardly credited by anyone at the time, but which proved about on the money — may have helped prevent polls and their aggregates from landing nearer the mark.

As it stands, the measure of any given pollster’s accuracy relative to its rivals was a simple function of how high it came in for Labor. As The Guardian’s display shows, line honours were shared by Resolve Strategic, RedBridge Group and Roy Morgan, and trailing the field was the Coalition’s hapless internal pollsters, Freshwater Strategy (apart from an Ipsos poll that was apparently half a point worse in having Labor’s lead at 51-49, which is news to me — the only polling I’m aware from it at any point during the term was limited to leaders’ ratings). Not included in this assessment was RedBridge Group/Accent Research tracking polling, which did very well for much of the campaign in pointing to a 3.5% Labor swing across 20 marginal seats that ended up swinging 4%, only to fall short with a 2% swing at the last. As an indication of how much better the polling industry performed than certain other areas of the media-political complex, the publisher that commissioned this polling persistently instructed readers to share its delusion that it pointed to a Labor minority government.

Another aspect worth noting of the news media’s horse race coverage was its acceptance of the Coalition’s claim that polling was failing to measure a revolutionary transformation in preference flows, such that the precedent of 2022 offered no guide on this score. Pollsters did in fact tweak their preference models in anticipation of weaker flows to Labor — BludgerTrack’s relatively good performance on two-party preferred had a lot to do with its persistence in applying 2022 election flows, which I must confess was more down to indolence than insight. We won’t actually know the true story here for a couple of months, when the Australian Electoral Commission will provide two-party preferred data from the unprecedented number of seats where the two-candidate counts include independent or minor party candidates, and — better yet — preference flows broken down by party.

We do, however, have 118 seats where the two-candidate counts are between Labor and the Coalition, and from which we can observe how well applying preference flows from 2022 would have done in projecting the two-party preferred. And they do in fact suggest that flows to Labor were quite substantially weaker than last time, such that Labor would have scored 55.4-44.6 across these seats on 2022 preference flows but in fact managed 54.5-45.5.

Click here for full display of House of Representatives election results.

Adrian Beaumont update at 11:50am Saturday: I wrote a long article on the Senate for The Conversation yesterday.  I believe Labor is likely to gain five seats from the Coalition.  Labor’s national primary vote in the Senate is slightly higher than in the House.

Friday

Sorry to disappoint Senate fans once again, but I’m kicking my promised post on that subject down the road for another day. Meanwhile:

Grey. By general acclaim, this should be added to the watch list, though my system is still calling it for Liberal because it’s giving independent Anita Kuss next to no chance of making the final count ahead of Labor. However, the AEC’s incomplete 3CP count has it very close (something may be amiss here though, because percentages are provided that don’t add up to 100%). I’ve tried making use of those figures in my projections, but what they are coming out with doesn’t accord with talk I’m hearing of scrutineers’ reports, which is that Kuss is doing a lot better on preferences from right-of-centre sources than I’m allowing for.

Bradfield. The Liberal lead fell from 237 to 209 mostly due to postals breaking 260-215 in favour of Nicolette Boele (maintaining an unbroken run of improvement for Boele across six batches of diminishing size, from 41.0% to 54.7%). The first batch of out-of-division pre-polls went 472-465 to Gisele Kapterian, who also made a net gain of ten on re-checking. I’m expecting about 1500 more each out of absents and out-of-division pre-polls, 200 provisionals, and let’s say another 200 late-arriving postals since there have already been more than I was anticipating. Boele will need about 53% of them, when there seems no particular reason based on the preceedent of 2022 to expect them collectively to lean one way or the other.

Kooyong. Monique Ryan’s lead fell from 724 to 661 as the first batch of absents broke 984-942 to Amelia Hamer, who also made a net gain of nineteen on re-checking. That’s probably over half the absents accounted for, and the remaining postals are unlikely to be appreciably helpful for Hamer. Her hope lies in out-of-division pre-polls, of which I had previously been suggesting 3000 could be expected, but after closer observation I believe it is more likely to be 4000. Others things being equal, Hamer will need about 58% of them — in 2022 the Liberals got 48.2%, which was 2.7% better than they did on ordinary votes, suggesting 50.2% if the pattern holds this time.

Bendigo. The two-candidate Labor-versus-Nationals count is taking its time catching up to the primary vote count, but Labor leads by 1.2% on what’s been counted, I’m projected them to a lead of 0.7%, and those who have been following the situation more closely than I have expect them to hold out.

Longman. The LNP lead is down from 289 to 231 after a second batch of absents broke 268-214 to Labor and the first out-of-division pre-polls broke 361-320, redressing by a 361-320 break in the latest postals to the LNP, the latter having exhibited little of their usual tendency to improve for Labor in later batches. I expect there will be a further 1700-or-so each of absents and out-of-division pre-polls, which will close the gap with about 100 to spare if they behave as they have so far, which they may or may not do. Here as elsewhere, there can’t be many more postals outstanding.

Bean. The first absents broke 820-679 to Labor, cutting independent Jessie Price’s lead from 195 to 54. That should be most of them — there should only be enough outstanding to exactly account for the independent lead if they behave like the first batch, though as I keep stressing, absent batches can be a bit variable. I would expect about 1200 postals, though since both batches so far have broken about evenly, that fact doesn’t offer much of a guide. That likely leaves the matter to be decided by how upwards of 2500 out-of-division pre-polls go — that absent votes and, to a lesser extent, pre-poll voting centres favoured Labor is presumably encouraging for them.

Bullwinkel. Things continue to trend Labor’s way here, their lead increasing from 333 to 634 after absents broke 595-364 in their favour, postals went only 143-142 the other way, and re-checking of early votes gave them a net boost of 71.

Thursday

I promised a review of the Senate a few days ago that hasn’t been forthcoming, but that will hopefully be rectified this evening, and I promise it will be worth it. Today’s developments from another place:

Bradfield. Nicolette Boele has suffered a blow in her already difficult struggle in chasing down Gisele Kapterian’s lead, with the first batch of absents breaking 477-427 against her, reflecting an unusually weak flow of preferences to Boele. This more than cancelled out rechecking that cut 50 votes from Kapterian’s total and 17 from Boele’s. Postals are now breaking about even, today’s batch going 243-238 to Kapterian. Kapterian’s lead is now 237, out from 215 yesterday. Boele now needs to hope for much better from around 2000 outstanding absents (not impossible, since these can vary from batch to batch depending on where they are sourced), out-of-division pre-polls (less likely) and what I presume will be a couple of hundred late arriving postals.

Kooyong. The odds on Monique Ryan holding out shortened after today’s postals broke 1966-1958 her way, her trajectory over three batches being 37.9% to 42.0% to 50.1%. Together with the effect of minor rechecking changes, Ryan’s lead has gone from 723 to 724, leaving Amelia Hamer to hope for something unusual to happen on maybe 1000 outstanding postals (which if anything seem likely to favour Ryan from now on), 3000 absents (which should also favour Ryan if 2022 is any guide) and as many out-of-division pre-polls (here at least Hamer can probably count on a few hundred votes).

Ryan. The AEC’s indicative three-candidate count is more or less complete, and finds Greens member Elizabeth Watson-Brown leading Labor’s Rebecca Hack by 30.44% to 29.79% in the race to make the final count and win the seat ahead of the LNP on the other’s preferences. I have used these numbers to revise the flow of lower order candidates’ preferences between the three in my projection, which hasn’t made much difference to what is now a 74.7% probability estimate to the Greens, since the changes involved a drop for the LNP and an increase for both Labor and the Greens. My model tends to get too conservative in estimating probabilities for leading candidates at this stage because, in the absence of hard information on how may votes remain to be counted, it errs on the high side.

Menzies. The first favourable batch of postals to Labor, breaking 1039-902 and pushing the lead out to 1300, marks an appropriate occasion to draw a line under this one.

Longman. Labor are hanging in here after the first batch of absents broke 383-280 their way, together with their usual modest dividend from provisionals, which broke 127-87, collectively cutting the LNP lead from 471 to 289. However, my system was clearly in error earlier today when it projected a high probability of a Labor win – how prescient it proves in continuing to lean slightly in their favour (now that I’ve replaced its dubious projection of primary votes with the raw results) remains to be seen.

Bean. The fresh two-candidate count between independent Jessie Price and Labor has caught up with the primary vote count, and it finds Price leading 48,353 to 48,158. If I were using the implied preference flow here I would be all but calling the seat for Price, but the truth is I have no idea what the outstanding vote types might do in a race involving an independent with no history, so I’m sticking with my existing preference estimates for no other reason than that the very close contest they project seems about right.

Bullwinkel. An already tough fight for the Liberals got harder with the first batch of absents breaking 524-429 to Labor. The long delayed two-candidate result from the Lesmurdie North booth also broke 402-316 Labor’s way, boosting their lead from 86 to 333.

Wednesday

I have obtained information from the AEC on which booths have gone into its incomplete three-candidate counts, resulting in meaningful revisions to my projections in the following seats:

Ryan. The 3CP count finds the Greens are under-performing my model on estimates from lower-order candidates, increasing their risk of having Labor closing the narrow primary vote gap and excluding them from the final count, in which case Labor will win the seat. Specifically, the Greens’ win probability is in from 79.6% to 64.0%.

Flinders. Here my three-way preference estimates were apparently about right: replacing them from the ones that can be inferred from the 3CP count increases the Liberal win probability from 87.3% to 89.2%. The shift probably reflects the fact that I now think there slightly more of a chance that independent Ben Smith will fail to make the final count, in which case Liberal member Zoe McKenzie’s win is certain. Even if he does make it, my system deems her victory very likely, projecting a two-candidate result of 52.1-47.9. However, this continues to be based on preference estimates — the AEC is not conducting a Liberal-versus-independent throw, though it may feel inspired to do so if the 3CP ends up confirming that Smith will make the count. Failing that, we will have to wait for the full preference distribution.

Richmond. Lower-order preferences are flowing a lot more strongly conservative than my model had counted on, in this case extinguishing whatever chance the Greens had of getting ahead of Labor to make the final count.

Two further seats warranting special mention, as I’ve retracted my system’s calls:

Bean. After continuing to mistrust my projection based on a still incomplete Labor-versus-independent preference count, I have reinstituted a two-candidate projection based on preference estimates, judiciously tweaked to reflect what the count seems to be showing. It had already withdrawn its call of the seat for the independent before I did so, and it now reckons it to be lineball.

Bendigo. I’ve done the same thing here, such that it is no longer calling the seat for Labor, as it was for a time today. But it still finds the seat leaning in their favour.

Elsewhere:

Bradfield. Gisele Kapterian only increased her lead today from 178 to 215, as postals followed their usual pattern in getting less conservative the later they arrived, in today’s case breaking only 495-468 (51.4%, compared with 58.3% of all previous). By my reckoning there should only be about 400 to come, though my reckoning might be out — it assumes 79.2% of postal vote applications will yield formal votes, based on what happened here and to a lesser extent in North Sydney in 2022 (postal votes can arrive up to two weeks after the election and still be admitted to the count if they were sent early enough). By this stage though, the bigger factor is absents and out-of-division pre-polls, of which there will be about 3000 each, together with a handful of provisionals. They respectively leaned independent and Liberal in 2022, but this can depend heavily on where the boundary is, since many of them are cast in booths just outside the electorate, and these have changed substantially with the redistribution. All you can really say here is that you would rather be ahead than behind.

Kooyong. After only rechecking was done yesterday, Monique Ryan’s lead shrank today from 1002 to 723 as postals broke against her by 2281-1649. However, this was a marked improvement for her on the first batch (42.0% rather than 37.9%), and rechecking of early voting centres improved her position by 332. I expect there to be a further 5000 postals, and the improving trend to Ryan would need to halt for them to get Amelia Hamer ahead. The precedent of 2022 suggests Ryan will gain about 300 on absents and break even on out-of-division pre-polls, but the caveats just noted for Bradfield apply here also — as does the concluding remark.

Done and dusted, or just about:

Goldstein. Postal vote batches continue to get less bad for Zoe Daniel, but are still breaking strongly against her and inflating a Liberal land that now stands at an unassailable 1362. My system isn’t calling it because it’s not doing data-matching, but I won’t continue following the seat on this post.

Melbourne. My system is calling this for Labor by a rather comfortable margin, on the basis of what now looks like a well-founded projection of preferences. I won’t continue following this one.

Menzies. My system is calling this for Labor. Postals broke 4258-3519 to Liberal today, cutting the margin from 1655 to 1145, but I only expect around 1200 to come, and absents should favour Labor.

Wills. My system is calling this for Labor, and I see no reason to doubt it. Nothing more from me on this one.

Longman. This continues to drift slowly away from Labor, today’s postals breaking 518-461 to push the LNP lead out to 471.

Fremantle. The Labor-versus-independent two-candidate count has caught up with the primary vote here, leaving Josh Wilson 1582 ahead of Kate Hulett, who has twice done well but not quite well enough. Another 4000 postals should widen the gap beyond the point where Hulett can hope to pull any rabbits out of the hat on absents or declaration pre-polls.

Continue reading “Late counting”

I have a new thread about this one that will follow late counting in seats by rather conservative results system still considers in doubt. This post attends to the ones that it is recording as having changed hands, or close enough to it — I have made the cut-off point a 95% probability rather than the usual 99%. First though, a plug for my paywalled article in Crikey yesterday on the likely make-up of the Senate, where Labor and the Greens between them look set for a clear majority they didn’t quite get to after the 2022 election, despite having respectively lost the services of Fatima Payman and Lidia Thorpe (erratum: I have Pauline Hanson and Malcolm Roberts the wrong way around in the article — it was the latter who was up for re-election). And second, here is a podcast I did yesterday in a state of hopefully not too obvious sleep deprivation (still ongoing) with Ben Raue of The Tally Room:

The summary below encompasses fourteen Labor gains from the Coalition based on post-redistribution margins from the 2022 election, which involves three complications: Bennelong was a Labor-held seat that became notionally Liberal (just) following the redistribution; Labor had held Aston since winning the by-election there in April 2023; and the defeated incumbent in Moore was elected as a Liberal in 2022 but contested the election as an independent. There are also two Labor gains from the Greens, both in Brisbane, and the special case of Calare, which former Nationals member Andrew Gee has retained as an independent. If these are the only seats that change hands, the final numbers will be Labor 88, Coalition 48, independents ten, Greens two and one each for Katter’s Australian Party and the Centre Alliance. The other new post for today focuses on the undecided seats that could potentially change this calculation. Relevant to this question is a point made in relation to each seat below: fairly consistently, Labor did best on election day votes, second best on early votes and – so far – worst on postal votes. However, the first batches of postals are fairly reliably the most conservative.

New South Wales:

Banks. One of the election’s many unexpected Coalition casualties was David Coleman, just months after he was unexpectedly elevated to the foreign affairs portfolio in part because of his reach within a Chinese community that accounts for a substantial chunk of the seat’s population — for all the good that did anyone concerned. Coleman went in with a 2.9% margin and came out, on my current projection, with -2.6%. The swings were 7.3% on the day, 2.8% on pre-polls, and — so far at least — 1.3% in his favour on postals.

Bennelong. A technical Labor gain, having had a post-redistribution Liberal margin of 0.1%, for which my usual formulation of “accounted for by a 9.4% swing” feels like understatement. That Labor was up 13.3% on the primary vote and the Liberals down only 5.6% reflects the gap in the 2022 numbers created by the seat’s absorption of much of North Sydney, where around a quarter of the vote went to Kylea Tink. The swings to Labor was 10.9% on the day, 7.3% on early votes and 5.3% so far on postals, though the latter will likely increase as further batches are added.

Calare. Former Nationals member Andrew Gee held his seat as an independent from 23.9% of the primary vote, easily overhauling the 30.3% for the Nationals candidate after securing most of the preferences of teal independent Kate Hook (15.9%) and Labor (10.2%) to hold a 6.3% lead that late counting is unlikely to change much.

Hughes. A result no one saw coming was Labor’s win in a seat Labor last held before John Howard came to power in 1996. A post-redistribution margin of 3.2% was accounted for by a 6.2% swing, giving Labor a margin I project to 2.8%. In the absence of two independents who polled 17.5% between them in 2022, Labor gained 11.2% on the primary vote while the Liberals were only down 3.5%. The swings to Labor were 7.7% on election day, 4.5% on pre-polls and 2.1% on postals-thus-far.

Victoria:

Aston. James Campbell of News Corp related mid-campaign that Labor “hadn’t bothered” to poll the seat they gained from the Liberals at a by-election in April 2023, and every indication was that the Liberals regarded it as in the bag. I am projecting a 3.4% Labor margin from a swing of 6.0% compared with the 2022 election, almost exactly equal to the 3.6% Labor margin at the by-election. Swings were 6.7% on both election day and early voting, and 3.0% on the postals counted so far.

Deakin. The script for the election did not involve the Liberals losing seats in Victoria, but so it proved in Deakin, where I project a 3.2% swing off a Liberal margin that redistribution reduced from 0.2% to 0.0%. Swings were 4.6% on the day, 2.3% on early voting and 1.1% the other day so far on postals.

Queensland:

Bonner. Labor’s second ever win in a seat created in 2004 was not close, the LNP’s 3.4% margin accounted for by a swing I project at 8.7%. Primary vote swing for and against the major were around 10%; the Greens were down 4.7% on a strong performance in 2025, no doubt reflecting an increase in the field from five to eight candidates, and competition from Legalise Cannabis in particular. Swings: 9.3% on the day, 8.4% on early voting, 7.4% on postals so far.

Brisbane. Both in 2022 and 2025, this seat came down to whether it was Labor or the Greens who made it to the final count and defeated the LNP on the other’s preferences. In 2022, Labor scored eleven more primary votes than the Greens, a gap the latter closed on Animal Justice preferences. This time Labor is up 4.9% and the Greens are down 1.5% (with the LNP also down 3.3%), a gap the Greens would need nearly every preference from lower order candidates to close. A two-candidate count the AEC was conducting between the Greens and the LNP on the night has been junked, and it is now in the early stages of a count between Labor and the LNP that will only confirm the former’s winning margin.

Dickson. The day that Peter Dutton feared when he pitched for a safer seat before the 2010 election arrived at a particularly inopportune moment, from a swing of similar dimensions to a number of seats in outer Brisbane: 7.8%, compared with a margin of 1.7%. Dutton has a projected 34.9% primary vote, down 7.2%; teal independent Ellie Smith’s 12.8% kept a lid on Labor, up 2.0%, and contributed to a 5.7% drop for the Greens to 7.3%. Swings were 9.5% on the day, 7.0% no early voting and 3.7% on postals so far.

Forde. My system still gives the LNP a sliver of a chance, but I’m quite sure I’ve never seen a lead approaching 3000 votes slip away at this point in proceedings. Further discouraging the notion is that postals are not favouring the Coalition as they are elsewhere, the swings being 6.8% on the day, 6.1% on early voting and 6.4% on postals.

Griffith. The 3.3% drop in Max Chandler-Mather’s primary vote does not of itself explain his defeat in a seat where he outpolled Labor 34.6% to 28.9% in 2022. The decisive point was the swing from the LNP, who were down 4.1%, to Labor, up 6.0%, resulting in Chandler-Mather facing Labor at the final count, rather than the LNP as he had done in 2022. The AEC was conducting a Greens-versus-LNP two-candidate count that it has pulled in recognition of that outcome, leaving my system to rely on an estimated 70-30 break in LNP preferences in favour of Labor over the Greens in projecting the final result.

Leichhardt. The retirement of Warren Entsch presumably had something to do with the biggest swing in Queensland, presently at 10.1%, off an LNP margin of 3.4%. Booth results suggest Entsch was known and liked in Indigenous communities, but nowhere was the swing insubstantial. It was 11.4% on election day, 9.7% on early voting and 4.7% on postals so far.

Petrie. Another Queensland seat Labor has long had trouble shaking loose, this time doing so off a 5.8% swing accounting for a margin of 4.4%: 6.9% on election day, 5.4% on pre-polls and 3.4% on postals so far. I note that, with the exception of lineball Longman, Labor has won all the seats in Brisbane that formed part of Kevin Rudd’s statewide sweep.

Tasmania:

Bass. Northern Tasmania had one of its trademark changes of heart at this election, with noted Liberal moderate Bridget Archer’s 1.4% margin was demolished by a 9.8% swing, panning out to 10.2% on the day, 8.9% on early voting and 8.1% on postals so far.

Braddon. Unlike a lot of other places covered here, there were rumblings about northern Tasmania, but not like this: Labor’s Anne Urquhart, hitherto a Senator, picked up a 15.3% swing, possibly helpd by the retirement of Liberal incumbent Gavin Pearce. Labor was up 17.3% on the primary vote and Liberal down 12.3%, the gap reflecting the 7.8% vote in 2022 for independent Craig Garland, now in state parliament. The swings were 15.3% on the day, 14.9% on early voting and, unusually, higher yet on postals-so-far at 16.5%.

South Australia:

Sturt. The Liberals’ last seat in Adelaide, retained by 0.5% in 2022, swung to Labor by 7.2%: 8.5% on the day, 5.9% on early voting and 5.5% on postals-so-far. Their primary vote was down by more than Labor’s was up (8.9% and 4.6% respectively) because of independent Verity Cooper’s 7.3%.

Western Australia:

Moore. The Liberals’ last seat in Perth, retained by 0.7% in 2022 (bumped up to 0.9% by redistribution), swung 3.8%: 5.0% on the day, 2.6% on early voting and 3.1% on postals-so-far. Presumably not helping was incumbent Ian Goodenough, who ran as an independent after losing Liberal preselection and polled 9.9%, declining to direct a preference to his old party on the how-to-vote card. The Liberal primary vote was down almost exactly the same amount while the Labor vote primary hardly changed.

Click here for full display of House of Representatives election results.

I’m not presently in a position to offer the result the commentary it undoubtedly deserves, so hopefully the live results feature speaks for itself. Rest assured that this will be supplemented with voluminous analysis of late counting in close races in the weeks to come. Two changes keen observers might note from the close of play last night are that Bendigo is now rated a lineball contest between Labor and the Nationals – which would be an extraordinary success for the latter given the general course of events – and in allowing for the possibility that an independent who is on only 12.6% of first preferences will snowball to victory in Calwell. These changes result from me changing the candidates designated for my three-way projection, which had originally excluded the potential winners.

I have just finished writing an article for Crikey on the Senate race, which should be available to subscribers later this morning (UPDATE: Read all about it). The upshot is that the Coalition will only win two seats in each state, and could be reduced to one on a worst case scenario in Tasmania, which will reduce it to its lowest representation since parliament assumed its present size in 1984. Labor and the Greens look likely to have a majority between them, even without accounting for further left-of-centre votes from David Pocock (who romped home in the Australian Capital Territory), Fatima Payman and Lidia Thorpe.

Click here for full display of House of Representatives election results.

Live commentary

8.30pm. Just popping in to say my results system seems to be working well, and if you’re finding it of value, perhaps you might consider making a contribution through the “become a supporter” buttons at the top of the site or the bottom of each post.

6.21pm. My system has successfully processed the first booth, which is the CBD pre-poll result for Fowler.

Preview

Polls have closed on the eastern seaboard, so welcome to the the live thread for discussion of the results for tonight’s federal election. I will be doing honest work this evening in the data room at the Nine Network (on which Ally Langdon, Peter Overton, Charles Croucher, Andrew Probyn and Liz Daniels will be joined by Chris Bowen, Bridget McKenzie, Katy Gallagher and Christopher Pyne), so I won’t be in a position to provide live commentary, though I may sneak in the odd explainer about technical aspects of the count and things that are going on with my live results.

On that front, the link above will take you to one of two entry platforms, the other of which is a map display that will colour in as results are reported to reflect who the system deems to be ahead or to have won, respectively indicated with a lighter or darker shade. Scroll over an electorate on the map and you’ll a window with a two-party pie chart and other essentials of the result. Click and the results page for the seat will be revealed, featuring progress totals, projections, probability estimates, a table recording votes and swings for each booth and, if you click on the button at the bottom, a booth results display map. The results system is based on a three-candidate model that hopefully had the bugs ironed out of it when it had a run at the Western Australian election in March. This will prove especially useful in Macnamara, the Greens seats in Brisbane and possibly one or two other places I can’t presently foresee.

The AEC is conducting Labor versus Greens counts in Grayndler, Sydney, Cooper, Wills and Canberra; LNP versus Greens in Brisbane, Griffith and Ryan; Greens versus Liberal in Melbourne; Liberal versus independent in Warringah (Zali Steggall), Wentworth (Allegra Spender), Mackellar (Sophie Scamps), Goldstein (Zoe Daniel), Kooyong (Monique Ryan), Indi (Helen Haines), Bradfield (Nicolette Boele), Cowper (Caz Heise) and Wannon (Alex Dyson); LNP versus independent (Kate Hook) in Groom; Nationals versus independent (Andrew Gee) in Calare; Labor versus independent in Fowler (Dai Le) and Clark (Andrew Wilkie); LNP versus One Nation in Maranoa; Bob Katter versus LNP in Kennedy. The information for South Australia, the Northern Territory and Western Australia will be provided when voting closes there.

If the choices in the Greens seats in Brisbane can be faulted for not including Labor where it seems possible or even likely (in the case of the seat of Brisbane) they will win, the results will in fact likely come down to whether Labor makes the final count, on which a two-candidate count would offer no guidance no matter who was chosen. This point can be illustrated with reference to Macnamara, where this evening’s Labor-versus-Liberal count is unlikely to tell us much: barring a better-than-expected result for the Liberals, Labor should win if they make the final count, and the Greens should win if they don’t. We may see a repeat of the ad hoc three-candidate counts that the AEC conducted for some of these seats after the 2022 election in the days to come. It is here that the Poll Bludger results feature comes into its own, as it integrates probability estimates for who makes the final count and what chance they have when the get there. However, this will of necessity be based on pre-determined estimates of preference flows unless and until three-candidate counts are conducted.

Now for some high wonk factor detail about the specifics of the count and the projection of the results. The Australian Electoral Commission relieves those of us in the election industry of the burden of producing “historic” data, namely estimates of how each polling place or vote type voted at the last election, a matter complicated by changes in locations and electoral boundaries. It is on this basis that swings can be calculated on a like-for-like basis, and a determination made as to whether they are big enough to cause the seat to change hands. The challenge it has faced on this occasion has been greater than ever because of the unprecedented number of independents who won seats last time, and the fact that nearly all of them were in states where redistributions have occurred. This means most of them are running in areas where they weren’t on the ballot paper last time, meaning there is no good way of determining their “historic” vote in these areas.

In a couple of cases, the AEC’s processes have led it to data that is of little value. The most glaring case involves Wills and Cooper: inner Melbourne seats where the threat to Labor from the Greens has intensified with the addition of territory from the seat of Melbourne. For reasons too arcane to go into, the AEC credits Labor with respective margins here of 9.0% and 8.9%, which makes little account of Greens strength in the newly added areas (my own determinations have it at 4.2% and 7.8%). Presumably to keep the historic two-candidate totals at this level, the AEC has produced two-candidate histories in booths around Carlton North and Fitzroy North that are wildly out of line with how people actually vote there. To pick one example: the Greens are allocated 30.4% in the Merri booth in Wills, where Adam Bandt got 65.9% in 2022. By contrast, the primary vote history has the Greens at a sensible 57.6% in this booth, suggesting the mathematical impossibility of them losing votes after preferences. If such booths are the first to report in these seats, the AEC’s early projection for the Greens in these seats will be massively inflated.

Independent incumbents, who in every case are teals, have been allocated zero primary votes in areas that were not in their electorates in 2022, meaning they will be credited with swings in these booths equal to the entirety of their vote. This presents my results system with a problem, as primary vote swings are a crucial element in how the final result is projected. If the AEC cannot be faulted for not crediting independents with invented numbers, they have in effect done so on two-candidate preferred – in this case with mostly plausible results. The AEC will use this data to produce useful swings and projections of its own based entirely on the two-candidate count. But it doesn’t do me any good, because my system is geared to make use of the primary vote, which gets counted first and is thus ahead of the curve.

There is one independent seat where I believe the AEC’s two-candidate determinations in areas now to the electorate will be of as little value as those noted for Wills and Cooper, namely Wentworth. In December, the AEC published its estimated margins for redistributed independent seats, which you can read more about here. The historic booth data matches these margins in every case except Wentworth, where it seems the previously published estimate of a 0.6% margin for Allegra Spender had it the wrong way around. The historic booth data in fact provides for a Liberal margin of 0.6% in this seat, such that Spender will need to record a swing in her favour to be projected as winning. I do wonder though if something might have gone awry in the AEC’s calculations, as its historic two-candidate numbers weigh heavily against Spender in such Liberal-unfriendly areas as Woolloomooloo and Kings Cross. So here too, the AEC’s early projections will be unreliable – specifically, heavily weighted towards Spender – if these booths happen to be among the first to report.

My approach to deal with these problems, which I can only hope won’t run into any major bugs, involves abandoning my usual effort to project preference flows based on booth matching, and – in the case of the independent-held seats – also to make no effort to project the primary vote. The numbers in both cases will simply be based on whatever the progress count happens to be. There are two independent seats whose newly acquired areas aren’t big enough for me to consider any of this a problem, namely Fowler (which gains a part of the suburb of Weatherill Park) and Wannon (which gains rural territory at the eastern end of the electorate).

The Australian reports the concluding Newspoll for the campaign has Labor leading 52.5-47.5 on two-party preferred, out from 52-48 a week ago, from primary votes of Labor 33% (down one), Coalition 34% (down one), Greens 13% (up two) and One Nation 8% (steady). Anthony Albanese is on 42% approval (down one) and 52% disapproval (steady), while Peter Dutton is on 32% (down three) and 60% (up one) and Albanese’s lead on preferred prime minister is unchanged at 51-35. The poll was conducted Monday to Thursday from a sample of 1270.

YouGov has weighed in with its final national poll, conducted last Thursday through to yesterday, which is distinctive even by the standards of recent polling in the weakness of the major party vote: Labor is on 31.1%, down from 33.5% in last week’s regular YouGov poll, with the Coalition on 31.4%, up from 31.0%. The Greens are up from 14.0% to 14.6% and One Nation is down from 10.5% to 8.5%, with independent up from 5.0% to 6.7%, Trumpet of Patriots up from 2.0% to 2.5% and other up from 4.0% to 5.2%. Presumably incorporating data from the latest MRP poll, the preference allocations have been tweaked: Labor’s share from One Nation is down from 33% to 31% (and from 35.7% at the 2022 election), but its independent share is up from 59% to 65% (63.8% in 2022). This pans out to 52.2-47.8 to Labor, in from 53.5-47.8 last week. Anthony Albanese is up one on approval to 43% and steady on disapproval at 49%; Peter Dutton is down three to 33% and up three to 57%; and Albanese’s lead on preferred prime minister is out from 50-35 to 51-34. The poll was conducted last Thursday to yesterday from a sample of 3003.

Also out today was a poll was a poll from DemosAU showing Labor on 31%, the Coalition on 33%, the Greens on 12% and One Nation on 9%. Labor was credited with a two-party lead of 52-48 – whereas in the past this pollster had been using preference flows from the 2022 election, this time the Coalition was credited with the 73% of One Nation preferences they received at the Queensland election last October, which would reduce Labor’s two-party by about 0.7%. The poll had a bumper sample of 4100, lending credibility to state breakdowns recording 50-50 in New South Wales (a Coalition swing of about 1.5%), 53-47 to Labor in Victoria (a Coalition swing of about 2%), 54-46 to the Coalition in Queensland (no swing) and 56-44 to Labor in Western Australia (a Labor swing of about 1%). The poll was conducted from Sunday to Wednesday.

The print editions of the News Corp papers have a final wave of the RedBridge Group-Accent Research tracking poll of 20 marginal seats, in which Labor’s blowout 54.5-45.5 lead over the previous two weeks has moderated to 53-47, implying a swing to Labor of 2%. The accompanying report is light on for detail – nothing on primary votes, field work dates or sample size – but there is likely to be more from a forthcoming online report.

Roy Morgan has a poll with Labor leading 53-47 on respondent-allocated preferences, unchanged on the poll released on Monday, from primary votes of Labor 33% (down one), Coalition 34.5% (steady), Greens 13.5% (up one) and One Nation 6.5% (down one). The two-party result based on 2022 preference flows is 54-46, likewise unchanged. The poll was conducted from Monday through to today from a sample of 1368.

That just leaves the only poll that matters – and BludgerTrack, which ends with Labor’s lead at a new peak of 53.2-47.8, exactly equal to Bob Hawke’s winning margin in 1983. I must stress that this is not at all what I expect to happen, and I will be happy to admit I was wrong if I am proved right, so to speak. The caveats were laid out in yesterday’s post: as well as a tendency of polling over recent years to overstate Labor’s primary vote, BludgerTrack’s application of preference flows from the last election almost undoubtedly flatters Labor, and I will probably come up with something different when I crank the machine up again next term. While I do not believe the more extravagant talk of One Nation flows to the Coalition matching those of the Greens to Labor, I do think YouGov, Newspoll and DemosAU have likely been judicious in bumping up the Coalition’s share from 64.3% to upwards of 70%, and that Labor’s 85.7% share of Greens preferences from 2022 is unlikely to be matched. That should be enough to knock a point off what BludgerTrack is crediting Labor with, which would make tomorrow night all about whether or not Labor can make it to a majority.

Late mail from party sources related via media reports:

• Samantha Maiden of news.com.au reported yesterday that Labor internal polling showed them leading in the Coalition-held seats of Sturt and Bonner, respectively by 53.5-46.5 and 51.5-48.5, and tied in Bass. In rather sharp contrast to talk coming from the Liberal camp, the polling shows Labor holding Whitlam by 56-44 and Gorton by 55-45. Labor is said to lead the Greens by 54.5-45.5 in Wills, where party sources say the Greens are withdrawing resources to focus on their seats in Brisbane. Maiden quotes a Liberal source saying Labor’s expectation of a final seat tally somewhere between 72 and 78 is “credible”.

• A different story emerges from the Liberal sources of The Australian, whose theory of victory involves winning Aston, Gilmore, McEwen, Tangney, Solomon, Paterson, Werriwa, Gorton, Hawke and Bullwinkel from Labor, Goldstein, Curtin and Kooyong from independents and Ryan from the Greens, pushing its seat count into the seventies and giving it the whip hand in cross-bench negotiations over Labor, who would be reduced to 67. The sources invoked went so far as to rate eight gains as definite and ten as likely. Concerns were nonetheless acknowledged about Bradfield and, unusually for such a bullish assessment, Bass and Braddon were “considered too close to call but the Coalition is confident of retaining them” (a bit of cognitive dissonance that might be thought to colour the rest of the assessment). Labor was “also facing neck-and-neck races in Bennelong, Robertson and Lyons”, though why they would be excluded from the list of seats about which the Coalition was “hopeful” if this were so is unclear. Then there is Lingiari, ̶which is considered tough to poll”, and presumably for that reason alone is rated “too close to call”.

Three new national polls to report. Known unknowns include a big sample DemosAU poll that should be along later today, and the inevitable one-last-Newspoll.

• YouGov has a third and final MRP poll, constructing demographically based estimates for all 150 seats off an impressive sample of 35,185, conducted from April 1 to 29. This maintains the pollster’s recent form of strong results for Labor – and, perhaps more to the point, weak for the Coalition ” offering a median projection of 84 seats with the Coalition on just 47, the Greens on three, and 16 for independents and others. Projected Labor gains are Banks, Bonner, Braddon, Menzies, Moore and Sturt from the Liberals and Brisbane from the Greens, while the Coalition is further projected to lose Bradfield, Calare, Cowper and Wannon to independents. A detailed display allows for results to be explored at seat level. The national voting intention results are Labor 31.4%, Coalition 31.1%, Greens 12.6%, One Nation 9.3%, independents 8.1% and others 7.6%, with Labor leading 52.9-47.1 on two-party preferred.

• The final Freshwater Strategy poll for the Financial Review credits Labor with a two-party lead of 51.5-48.5, out from 50-50 in its mid-campaign poll, the first lead for Labor in this series since March last year. The primary votes are Labor 33% (up one), Coalition 37% (down two) and Greens 12% (steady). Leadership ratings are particularly encouraging for Labor, with Anthony Albanese up four on approval to 41% and down four on disapproval to 44%, Peter Dutton down one to 35% and up four to 51%, and Albanese’s lead as preferred prime minister out from 46-41 to 49-39. The poll was conducted Tuesday to Thursday from a sample of 2055, which is nearly twice as big as usual.

• The new online regional news publisher The Gazette has a second DemosAU national poll for the week, recording a two-party lead for Labor of 51-49, compared with 52-48 for the first. The primary votes are Labor 29%, Coalition 32%, Greens 12% and One Nation 9%, as compared with 29%, 31%, 14% and 9% in the earlier poll. It was conducted Sunday to Tuesday from a sample of 1974.

There remains the small matter of how accurate all this will prove to be, and the fresh memory of a general failure in 2019 that was in fact nothing special by international standards. The question of whether this was an aberration in a long-term record of strong performance by the Australian polling industry gets a negative answer from political scientist Luke Mansillo, who has been behind an aggregation model for The Guardian that long appeared quixotic in indicating a Labor primary vote in the high twenties, below what any individual poll was saying. The latter continues to be the case with its current central estimate of 30%, but its two-party preferred measure now has an eminently believable central estimate of 51.5-48.5 in Labor’s favour.

A sense of why such a model should be so bearish with respect to Labor can be gained by comparing final week polling for the most recent elections federally and for the five mainland states with the actual results. The table below goes into detail for two pollsters who have covered enough state elections to make them worth the effort, followed by all final week polls combined for each election (no row is featured for South Australia as the only such poll was the Newspoll).

Thirteen of the sixteen polls covered here overestimated the Labor primary vote, but these errors have tended to be obscured by weaker readings for Labor on two-party preferred, which have shown no bias one way or the other. This was notably the case at the 2022 federal election, at which the last polls by Newspoll and Ipsos both had Labor on 36% – well clear of an actual result of 32.6% – but did well enough in recording two-party preferred at 53-47, compared with an election result of 52.1-47.9. The suggestion of pollsters under-estimating preference flows to Labor is at odds with an emerging narrative about tomorrow’s election that was covered here in depth yesterday.

My own BludgerTrack aggregate, which was given the seal of approval of Laura Tingle of the ABC on 7:30 last night, abandoned the notion of correcting for past observed bias after 2019, and now presumes only to offer a snapshot of what the polls are saying, warts and all. Adjustments are made, but their aim is to reduce distinctions between, on the one hand, Pyxis Polling and YouGov – both of which have had charge of Newspoll at different points throughout the past three years – and the various other pollsters. In two-party terms, these adjustments amount to about three-quarters of a point in Labor’s favour for Freshwater Strategy and half a point for Resolve Strategic, and about a third of a point in the Coalition’s favour for RedBridge Group.

As the table shows, correcting BludgerTrack for the general pattern of errors over the past few years would essentially involve moving about 1.5% from Labor to the “others” column. To extrapolate that to various estimates that were discussed in yesterday’s post on the subject, that would reduce Labor from its present 53.0-47.0 lead in BludgerTrack to 52.3-47.7 based on 2022 election preference flows; to 51.7-48.3 on YouGov’s current preference model; and to 51.9-48.1 on what I take Newspoll to be doing. The latter two are a fairly comfortable fit for what The Guardian’s model says. Then there is what I described yesterday as the “maximal” model, which accommodates what some observes take to be an historic blowout in the share of preferences the Coalition is about to receive from right-wing parties. This tips the balance all the way to 50.4-49.6 in favour of the Coalition by giving them 80% of a 7.9% One Nation vote and nearly as much of Trumpet of Patriots’ 3.5%.