Goldstein Count History And Partial Recount

GOLDSTEIN (Vic, IND vs Lib est 3.9%)
Tim Wilson (Lib) has provisionally won by 260 votes
AEC has authorised a partial recount

Wilson will win unless large errors are found during partial recount

RESULT: Wilson won by 175 votes.   

---------------------------------------------------------------------------------------------

Updates

Updates will be posted here scrolling to the top.

Saturday 1:55: It's finally over, Wilson has won by 175 votes and the seat will be declared.  

Saturday 31st 11:35 am: The end cannot be far away as corrections are now showing to absents, dec prepolls and all booths except Hampton PPVC where only minor changes are expected though I do not know how many postals are to be checked.  Wilson leads by 164 following the expected Brighton corrections which actually cut his lead by 65.

Friday 7:30: Further confirmation sighted that there is a correction of at least 50 votes coming in the Brighton PPVC booth tomorrow as a result of a bundle of 50 being double counted previously.  Also that the PPVC checking is close to done and the partial recount should end tomorrow, possibly by mid-afternoon.

Friday 5:30:  I have heard that there are still plenty of postals to be recounted tomorrow, and also that there is another substantial but not decisive correction that may occur in Daniel's favour (pending further checking) - Wilson would still have a lead well in three figures. 

Friday 30th 3:10: Wilson drops another 13 on postals, apparently 7 to informal and 6 didn't exist.  (Means a slight miscount in the number of votes in his pile). Lead now 233.  Not known to me if this is all the postal checking and especially whether all Daniel's postals have been checked too.  Still two large PPVCs outstanding - one of them Hampton which had issues earlier - and no movement in other declaration votes.  

Friday 30th 11:45: A larger than usual correction in the Moorabin West booth where Wilson has lost 11 primary votes and Daniel has gained 2.  Wilson lead down to 246 but only four ordinary booths unchecked (plus non-ordinaries).  Two of the remaining booths to be rechecked are large prepolls.   

Thursday 29th 7:30 pm: Another 20 booths are showing with updates from today and the net change from those booths was ... zero! Wilson's lead remains at 263.  There are only nine booths yet to be reviewed; also I am seeing no changes in figures for any kind of non-ordinary votes so I suspect they are still to be done as well.  The lead went as low as 259 and as high as 269 in what passed for fluctuations according to one T Wilson who was paying more attention through the day than I was.  

Wednesday 28th 6:30 pm: A rather impressive 17 booths are showing with updates already from today but not much seems to have happened.  Wilson's lead went at least as high as 267 votes but has now dropped back to 263.

Main Article (28 May 9 am)

It's my policy here to start a new thread when a federal seat qualifies for a recount.   In this case it's a partial recount, but I also feel that it's useful at this point to write a fresh narrative of the history of this unusual count which has caused a great degree of interest (and been the subject also of a lot of incorrect claims).

Goldstein stands to be the only Coalition gain in an otherwise terrible election result.  The seat was considered vulnerable in the leadup to the election but received less attention than Kooyong, which was often considered at greater risk and saw an even scruffier campaign but has been retained by Monique Ryan.  

The election night count started well for Zoe Daniel who was projecting as ahead early in the night.  A screenshot of the ABC website possibly taken around 10:30 shows an ABC projection of "IND Retain" with 52.2% projected 2CP based off 65.3% counted.  Around 10:40 with 70.4% counted the seat came up on screen with the primary votes and the projected win showing but no projected 2CP.  However Antony Green was more cautious in his verbal commentary saying "we're giving it away" (based I believe on an automated calling system) but that he thought Daniel had won the seat and would have a look at the figures later.  Shortly after this Daniel celebrated victory. I am not aware if there was any basis for doing so at that point other than the ABC projection.  

Overall projection issues with this seat included the within-electorate prepolls having a bigger primary vote swing to Wilson than the day booth votes and the postals having a bigger one, but I suspect there were other problems with the early figures caused by the handling of booths that had been brought into the electorate.  In any case by the end of the night the ABC had the seat back in doubt.

It became clear over coming days that early postals were terrible for Daniel whose lead rapidly disappeared.  Wilson took the lead on the Tuesday after election day at which point I moved the seat to expected win status.  Many networks called it around that point and on Wednesday Wilson declared victory.

By the end of the first week Wilson's lead had blown out to well over 1000 and it looked highly unlikely to even be close.  But in the second week Daniel started closing rapidly, though she was seldom if ever closing quite fast enough.  This closing was possible because although there was a (non-redistribution-adjusted) 2.8% two-candidate-preferred (2CP) swing against her on ordinary votes (day booth and prepoll votes within the electorate combined) and a 4.8% swing against her on postals, on absents the swing against her was only 1.6% and on out of division prepolls the swing was to her, by 0.8%.  A factor in the latter was that for whatever reason the Liberal primary vote on out of division prepolls was poor.  There also seemed to be a late absents/early absents dynamic where the later batches of absents were better for Daniel, perhaps because of where they were from.  

The postal shift spawned many conspiracy theories (frankly if anyone has evidence of anything untoward in nursing homes, take it to the AEC and the police rather than posting baseless claims on social media).  In fact there are many possible explanations for what ended up being a moderate 2.4 point difference in swings from the ordinary votes, including:

1. The postal voting pool generally contains fewer COVID-concerned voters than in 2022. Voters choosing to vote by post in 2022 to reduce their COVID risk are likely to have been to the left of the average postal voter.  There is a general national trend that postals,relative to ordinary votes, were somewhat worse for Independents and Greens in 2025 compared to 2022.

2. Orthodox Jewish voters, who vote heavily by post, were probably more likely than other voters to change their vote/preference away from Zoe Daniel and Monique Ryan over Israel/Gaza/antisemitism issues.

3. I understand the Coalition stepped up or refined postal vote targeting in Victorian seats.

4. There may have been late swing in voting behaviour (though indicators from other vote types in this seat are mixed).  

Whatever its causes the postals pattern plus a minor correction (see below) saw the lead shrink rapidly with Wilson 128 ahead at the end of the indicative 2CP count.  At this point Daniel made some media comments suggesting she was considering requesting a recount citing some comments about the count by Simon Jackman.  I dismissed those arguments (at least in the form reported by media) on my Goldstein page.  Firstly these arguments referred to a change in the Hampton prepoll booth, but this was an obvious error that had been corrected.  Secondly they referred to the idea that a larger voting population since the under-100-vote recount guideline was formalised justified a higher recount threshhold of say 140 (the margin is almost twice that now), but as turnout increases, the chance of an error that is a fixed percentage of turnout does not remain the same.  Thirdly there was reference to "lumpiness" in the count, but corrections in early stages of counting are not unusual.  

Some other support for a recount on social media came from the leaking of some group chat messages by Wilson in which he seemed stressed about the rate at which Daniel's scrutineers were"knocking out" his votes and suggesting he needed more scrutineers to do the same for informal votes in Daniel's pile.  Aside from showing that Wilson still needed to work to make sure of a win he had already claimed (or at least to make it easier to get it declared faster), there wasn't anything to see here - a  normal candidate to scrutineers type conversation.  Scrutineers don't decide whether votes are formal, the AEC does; the "knocking out" referred to is simply challenging apparently informal votes so the AEC can make a ruling on them.

Things were, however, to become unusual later.  

Corrections in this count

This count has seen an unusual number of corrections.  Before running through these I should explain the stages that all counts go through.

Each vote is first counted in an initial primary vote count for its booth (or for declaration votes these will occur in batches - eg the postals will be counted in several goes, usually starting with batches of around 2000 but becoming smaller later in the count).  Then each vote is counted in an indicative two-candidate preferred count, which forms the running 2CP count for the leaders for the seat.  Subsequently the initial primary and 2CP counts are reviewed in a second round of counting called fresh scrutiny. The initial count and the first indicative 2CP count are mostly done under great time pressure by casual workers on election night and errors in this process are pretty common  Fresh scrutiny tends to fix nearly all of them.  

The indicative 2CP count is a shortcut for election watchers to have an idea who is winning and also to help tell the AEC when it is safe to formally declare the seat because it is mathematically certain who will win.  But it does not itself supply the winner.  The winner is the person who wins at the end of the next stage, the distribution of preferences.  In this stage candidates are eliminated starting with the candidate in last place, and as each candidate is excluded the votes they were holding when excluded are passed on to whoever the voter put next of those candidates still in the race.  This is the standard process that all seats go through at some stage.  At the end of the distribution of preferences the AEC can choose to declare the count finished or to conduct a full or partial recount.

Some errors in this count (not an unusual number) were picked up and fixed by the end of the fresh scrutiny phase:

* In the Hampton prepoll Daniel had been credited with about 500 primary votes that were actually primary votes for the Greens (many of which flow to her as preferences anyway).  This was an obvious error because the Greens' share of the primary vote in the booth prior to fixing it was only 2.5%.  (Most often this error occurs when a bundle of 500 votes gets tallied to the wrong candidate - it's not super-rare and a similar thing happened in Solomon with far greater impact on the 2PP there.)

* In the Malvern prepoll Daniel's share of preferences on the initial count appeared impossibly weak.  This was corrected.

* In the Hampton day polling place Ben Raue noticed that there was an issue with preference flows in the Hampton day booth; it appeared the numbers had been inverted on second data entry.  This was fixed before the distribution of preferences started.  

It is possible, particularly in seats that haven't had much attention from scrutineers, for mistakes to happen even after the fresh scrutiny phase.  However in close seats with a heavy scrutineering presence, mistakes are nearly always caught and corrected, and most of the changes that go on in the distribution phase involve tiny numbers of votes being decided to be informal that were previously formal, because a repetition or omission of a number had been missed early in the process.  Now and then one sees a change with a double-figure impact on the lead resulting from some other error, for instance in Bradfield there was just one of these, worth 22 votes - one change of a few dozen is similar to other casess.  

The Goldstein distribution of preferences saw at least five significant margin changes that weren't just the usual trickle of a vote or a few votes here and there.  Wilson's margin increased by 96 votes in the Oakleigh prepoll and 172 votes in the Hampton prepoll (on top of the previous correction there).  The margin then decreased by 274 votes on postals before increasing again by 44 votes on absents and 38 at Beaumaris Central.  (My thanks to Ben Raue for keeping a list of all these and their exact sizes).  All up the margin had gone up to 444, down to about 170 before finishing the distribution at 260.  There was also a small correction in Wilson's favour before any of these five that may have been unusual too but I have not determined the details yet.

My understanding is none of these five larger corrections involved votes physically in the wrong pile.  All involved data entry errors - in one case a failure to overwrite the initial 2CP split from a postal batch with the new one, and in at least some of the others the 2CP split being entered incorrectly.  Unusual in a close count and I think the AEC will need to review its data entry checking processes to work out what happened here.

Partial Recount Background

The AEC has a standing policy, formalised following the 2007 election, to automatically grant a recount request if the margin at the end of the count is 100 or less.  In the years from 1984 to 2007 this policy had been applied informally except for Bowman 2007 (64 votes) where a request was rejected.  There had been three rejections in the 101-120 vote range, but Hinkler 1984 (237 votes) was accepted for recount for reasons unknown to me at this stage.  

For Goldstein 2025, Daniel requested a full recount citing the fluctuations seen in the distribution of preferences.  If there were so many data entry errors in the distribution, does this show potential for similar errors in the primary vote counts?  It may be that it doesn't, but even if so it's a very hard sell in optics terms.  What the AEC has done is to allow a partial recount that will check the primary votes for both candidates and the informal votes.  These are the votes for which the fresh scrutiny numbers have not yet been further reviewed - the votes moved in the distribution of preferences are not going to still contain significant errors.  In my view this decision by the AEC is very sensible.

The partial recount is expected to take around four days and I will post comments on it as it goes.  Generally recounts do not change the margin much (see Bradfield thread).

One of the common questions I have had is what happens if the partial recount now moves the margin below 100 votes, does this trigger a further automatic full recount?  The answer is no; in this case most of the votes will have already been recounted and the potential for a move of anything near 100 votes if the remaining votes were recounted is negligible.  However if the result after the partial recount was extremely close (I would think especially 20 votes or less), the AEC could decide to extend the recount.  This has been left open.

At some point a winner will be declared ("what if it's a tie?" has been dealt with on the Bradfield thread) and the declared winner will be seated; the loser will then be able to challenge in the Court of Disputed Returns if they can find any basis to do so.  

Partial Recount updates

Partial recount updates will be posted at the top.  The recount starts 9 am Wednesday but I am on fieldwork for that day so updates will be scarce barring anything sensational.  

Before I start here's a graph showing how Daniel's preference share in booths in Goldstein relates to how she stands on primary votes relative to Wilson (small special hospital booths have been deleted):

There are no massive outliers here (the nearest thing to one in the top left is Caulfield South) but the relationship isn't particularly strong and Daniel's preference share is reasonably flat across the different booths.  If a booth is above the best fit line, Daniel is doing well on preferences relative to where she stands on primaries, so any significant correction that might happen in primaries would more likely favour her.  If a booth is below the line, Daniel's preference flow is weak relative to the primary votes, so a correction is more likely to favour Wilson.