WA Senate Take Two: Preview, Live Comments And Post-Count

Seats Called: 3 Lib 1 ALP 1 Green 1 PUP

Summary and general comments (edited to update as required):

Liberal Senators David Johnston and Michaelia Cash, ALP candidate Joe Bullock and Greens Senator Scott Ludlam are all elected based on party quotas.  Palmer United Party candidate Zhenya Wang is close to quota and will be elected on micro-party preferences.  Liberal candidate Linda Reynolds (who would have won in the first election had it not been annulled) and Labor Senator Louise Pratt were contesting the final seat as of election night.  Both Labor and the Coalition have little over half a spare quota in their own right and the outcome will be determined by the preferences of other parties including the surplus votes of the Greens and, once they cross the line, PUP.

The hope for Labor in the post-count was that the pattern of postal votes relative to ordinary votes would be different to how it was in the 2013 federal election, when the Liberals performed nine points better on postal than ordinary votes.  In particular, it might have been thought Labor's dire final week (with constant bad publicity surrounding Bullock) would have meant that its vote before the last week had been higher.  Postal votes in so far are very consistently showing that the Liberals' performance relative to booth voting is at least as strong as in 2013, possibly stronger, and for this reason Reynolds will win.



A possible tipping point apparent earlier in the count was whether PUP crosses the line before the HEMP Party is excluded or only when the HEMP Party is excluded.  The former could assist Labor by up to 3000 votes in terms of the difference in net margin between the parties (compared to this not happening by the smallest possible margin), but the latter currently seems more likely, and Labor looks like ending up much too far behind anyway.  On primaries as they stood at the end of polling day and until Tuesday evening, Labor would have won because below-the-line votes would have caused PUP to cross before the HEMP Party were excluded.

Going into this election, there were four candidates who had half-won the first time around (Dropulich, Pratt, Ludlam and Wang) and of those, Dropulich was not going to win again.  The question then was which of the candidates left in limbo in the original election would lose, or whether all might win at the expense of the Liberal Party's third seat.

The answer is that Ludlam and Wang have bolted home while it is Labor who have missed out.  For an Opposition to lose a seat in a by-election atmosphere, while a government struggling for popularity retains all its seats, is a disaster born of a campaign which would have survived being uninspiring had it not also descended into farce.  Not only have Labor won only one prize out of six, but that one is the booby prize, and the victory by Bullock could go on to hurt the party more than if they had wiped out altogether.

While much less complicated than the 2013 WA or Tasmanian Senate counts, following this count has still been a complex modelling exercise and the article below contains large volumes of detail.

NB During the post-count this article scrolls from the bottom with the most recent updates at the top, directly below this section.

Post-Count

Note: The term "parallel calculator" below refers to a variant of the ABC calculator in which the PUP vote has been increased to make PUP reach their quota before the exclusion of the HEMP Party.  At stages in the count this looked likely to occur but it it eventually narrowly failed to happen.

29 April: The button has been pressed and it's all over.  The key issues from the distribution of preferences: as at the exclusion of HEMP, Palmer United were still 1180 short of quota so the tipping point was not activated, the party having made up only around two thousand votes on BTLs.  However, the margin ended up being much higher than the ABC calculator expected with Reynolds defeating Pratt by 12,127 votes. This probably reflects what I noticed earlier: that missing the tipping point made more of a difference the closer PUP got to it, because it increased the size of their overwhelmingly Liberal surplus. But I doubt that's the full explanation for the difference between the margin and the ABC projection and suspect close examination will show other factors (such as a higher rate of BTL leakage away from Labor or to the Liberals).  Analysis from readers who have the time to look at this more closely is very welcome.

29 April: Today's the day - button press at 2 pm (4 pm in the east) and final results and the distribution should come through soon after.  I'm working in town today so I'll wrap it up when I get home, probably a couple of hours later.  Final projected margins: ABC 8,109, parallel 7,558 (PUP shortfall to be made up on BTLs to get elected before HEMP exclusion 3,170). 



26 April: 88.54% now counted and the margins keep dropping: 8,119 by the ABC and the parallel calculator has 7,567.  The PUP shortfall keeps growing, now at 3,171.

24 April: Still more votes have been added and with a very respectable 88.41% now counted, the ABC's margin is 9,255 and the parallel calculator has 8,567.  Whichever way it goes the margin has come down a lot more than I projected it to do, apparently (see James Simpson in comments) because unknown to me, the AEC set aside non-ordinary BTLs and added them to the count last.  As a result the margin has come back to about the area of the earliest projections made after the night.

The deadline for postal votes passed on Tuesday and there are no votes shown as unapportioned.  So the button press may not be too far away!

23 April: Votes have continued to trickle in and with 87.64% counted, the one percent added in the last week has closed the gap substantially.  The ABC's margin is 10,298 and the parallel calculator has come down to a mere 9,393.  PUP's requirement to activate the tipping point has blown out to 3,021, which introduces more doubt about whether this will happen, though I suspect PUP's BTL performance will be better than in 2013.

Thursday 17th late night:  The primary count has now advanced to 86.61% counted.  Parallel calculator margin is 12,626 while the ABC's margin is 13,629.  PUP currently only need to gain 2,574 on BTLs to activate the tipping point and I think this probably will happen now, but whether it does or not is completely academic.

Wednesday 16th late night: We're now over 83% counted, with probably not much to go, and the various calculator margins have only widened by a few hundred votes over the last 5%, compared to the dramatic rise before that.  Probably the greater share of absent votes in the mix has a lot to do with that and perhaps the margin will still pull up below 15K (it's currently just under 13K with the tipping point included).  We're still well away from the button press which I think is expected towards the end of the month.

In conceding her seat is lost, Louise Pratt has made very strongly worded attacks on the party's preselection decisions and on Senator-elect Bullock, who she describes as "deeply homophobic".

Monday 14th late night: With 78.43% counted the margin on the ABC calculator is 13,422. The parallel calculator has it at 12,637 and has PUP needing to make up a plausible 2394 on BTLs to cross at the relevant stage.

Sunday 13th: The margin is continuing to blow out, having just gone over 10K on the ABC calculator, and it is probably heading for 15K or more.  The main feature driving this is the low Greens vote on postals, and in the end it looks like we will be able to say that the Liberals' win of the third seat was not even all that close, and just looked close on the night because of uncertainty as to whether the post-count would follow past patterns.  In the end the post-count has been even more favourable to the Liberals than expected. 


Saturday 12th, 8:10 pm (6:10 pm)  Quite a deal of counting today.  Now at 75.23% counted and the lead on the ABC calculator is just shy of 10K, or 9K on the parallel calculator.  The gap PUP need to make up on BTLs for the tipping point to activate has been rising slowly and is now at 2124.

Saturday 12th, 11:40 am (9:40)  James Simpson in comments has spotted a new dimension to the tipping point saga in that now on the calculator the Liberal Democrats are excluded before the Voluntary Euthanasia Party, meaning that the last exclusion before HEMP is now the Nationals.  As it turns out PUP receive nothing above the line on the Nationals exclusion and only the preferences of the Sports Party on the exclusion of VEP.  I don't believe this affects the mechanics of the tipping point since the key question is what happens to the HEMP preferences, and not in what order the preferences before that are distributed.

Friday 11th, midnightish: 73.55% counted.  The blowout continues with Labor now nearly 7000 behind on the ABC calculator, or 5681 behind on the parallel version.  On some of the more conservative projections the margin should stop growing soon, but I'm not sure it will!

Friday 11th, 9:10 (7:10): 73% counted.  We are starting to see some early prepolls as well as absents and postals but the margin keeps blowing out - now 5626 or 6538.  And still there are a lot more postals to go.

Friday 11th, 6:20 (4:20): Now that I am satisfied this one's a done deal I won't be updating it as often. At present with 72.28% counted, the parallel calculator margin is 4071, the calculator margin is 5377 and the gain PUP needs on BTLs for the tipping point between the two to activate is 1497.  The latter is growing quite slowly and I still don't rule out PUP crossing on Lib Dem preferences.  But whether they do or not affects only the margin, and that not by all that much.

Thursday 10th, 9:40 (7:40): Some more absents and postals from Brand saw a small easing in the margin on my parallel calculator to 2513 (remember: this deficit represents Labor's best case on current figures).  The ABC calculator margin is 3869.  But in a further sign of how much over this is, Brand is now the most counted electorate at 82% counted.  It might only have a few percent to go.  Yet Brand is one of Labor's strongest electorates.  Some others that are much less helpful have a lot more post-counting left.

Thursday 10th, 9:20 (7:20):  First postals from Canning, Stirling, Swan and Tangney and extras from Brand have been counted and it's a familiar story; with 71.21% counted the gap with tipping point is 2693, while PUP move further away from the tipping point assuming all votes are ATLs (1602).  The ABC calculator gap without tipping point will be well over 3500 when the latest postals are added.   I should also note here that if PUP don't cross at the tipping point, but are closer to it than the ABC calculator expects, then that increases the value of their surplus, which goes almost entirely to the Liberals.  So it could be that the flow of BTL votes to PUP actually increases the margin above the calculator's final projection, if PUP fail to get quota on the LDP votes.

I'm not personally much concerned about modelling the margin and won't be putting a lot of work into it.  My concern is the result and on that score I am seeing no reason to doubt that the Liberals are winning the third seat. 

Thursday 10th, 8:40 (6:40):  The postals are back now (minus 100) so the count is at 70.31% with today's additions so far being dribs and drabs of ordinary voting.  Even on those Labor is down some more - now 2743 votes behind without the tipping point, or 1303 with it.

Thursday 10th, 6:10 (4:10): No idea why the postal count has been down on last night's for a few hours now.  If these are the real figures then Labor's net position is about 200 votes worse than it was at the end of last night but we will need to wait and see if those numbers return.

Thursday 10th, 3:20 (1:20):  Figures are moving around a little but it seems there is again some data re-entry going on.  I'll wait for the postals to get back to where they were before rerunning modelling.  (As an aside it would be really useful if the AEC system kept old figures until they were overwritten with new data - taking data out before replacing it is quite confusing to follow.)

Thursday 10th, midday (10 am):  No new figures but something interesting to keep an eye on is the modesty of the Green vote in postals compared to ordinary votes.  This is a normal thing to see but in this case it seems to be unusually pronounced.  In the five electorates with some postals counted, in 2013 the Greens polled 10.2% in ordinary voting and 6.4% in postal voting.  But in this election in those five electorates so far, the Greens are averageing 16.1% in ordinary voting compared to 7.9% in postal voting - the surge in ordinary voting is very much more pronounced, even proportionally.  That's compatible with the Greens picking up a surge late in the campaign - as is the fact that for the first time in quite a while they didn't underperform their prior polling.  For those looking for the smoke signals of the Bullock Factor there could be a few points of it here (driving votes from Labor to the Greens), or there could be other explanations.  Anyway we will need to wait to see if this early trend lasts.  And if it does, the Liberal margin will blow out by even more than people are projecting.

The other thing coming through in the postals is that PUP were always going to get a lot of votes at this election.

Wednesday 9th, 10 pm (8 pm): 70.28% counted.  Many of the votes added tonight are postal votes;
11,138 are in the count already.  And we are seeing what we expected to see once postals arrived: the Coalition charging into a genuine lead.  The PUP-to-tipping-point margin is now 1238 but more importantly even if PUP cross before the tipping point Labor are now behind by 1050; otherwise they are down by 2508.  And there are a lot of postal votes to come.  This damage has come on the addition of less than 1% of the count total and not even from an entirely unfriendly set of electorates.  On postals, the Liberals are running ten points above their booth vote in Perth and Durack, eleven points up in Brand and Curtin and thirteen points up in Hasluck.  The theory that postals might not be as bad for Labor as booth voting (or even as bad compared to booth voting as they normally are) is being destroyed.  I see no basis in evidence for the idea that Labor can save this from here and am not far from calling it already.

Wednesday 9th, 9 pm: With 69.87% counted, Labor has just lost the parallel calculator lead as well.  The Liberals lead on it by a far from secure (but likely to grow) 44 votes and the tipping point margin is 942.  Labor's gains on absents from earlier today appear to have now been wiped out.

Wednesday 9th, evening:  Certainly looks like minor updating in 11 of the 15 electorates so far.  The parallel-calculator margin is now 447 and the notional PUP-to-tipping-point margin is 954.  (I'm editing these figures every now and then. Last update 9:00 pm eastern time, 69.79% counted).

I've fixed a small teething issue with my parallel calculator which was allocating votes of the two ungrouped candidates to Labor; in reality they are likely to lean Liberal since van Lieshout, who is getting slightly more of them, is a right-wing-fringe "Christian".   The correction, applied to all today's updates below, reduces the margin in favour of Labor by around 250 votes.

The number of Brand absents counted has been revised downwards to 1653.

Wednesday 9th, 7 pm (5 pm): There have been slight changes in many electorates in the last few hours.  Possibly this is just data re-entry after checking; at the moment Labor has improved to a parallel-calculator margin of 642 but it remains to be seen if this will last.

Wednesday 2 pm (midday):  69.51% counted.  The addition of 1734 absent votes in Brand has put Labor back in the lead assuming that the tipping point is crossed, but this is not representative as absent votes tend to favour the ALP compared to other non-ordinaries, and should especially do so in Brand.  I have written my own parallel calculator spreadsheet to compensate for deficiencies in the ABC's and it shows a margin of 538 to Labor on current figures assuming the tipping point is in their favour, which it would be as PUP are only 786 shy of it assuming all votes are ATLs.  The margin without tipping point is currently given by the ABC as 1295 to the Liberals.  Remember that this is not representative of what will happen in the rest of the post-count so Labor should draw no comfort from this apparent and very slender lead.  I am still projecting the Liberals to probably win.

Tuesday 9:42 (7:42 pm): Changes happening quickly now.  Labor back in front on parallel calculator, but only by 215 votes compared to around 1000 at the start of the day. 69.39% counted.  Antony Green relates that the mysterious early eight votes on polling night came from Eucla mobile, from an area which is on a different time zone to the rest of WA.  Roller-coaster vote changes tonight are attributed partly to re-entry of data from mobile booths.  Antony also notes that the Sports Party and #Sustainable Population Party are going to lose about 10% of their votes when keying errors are fixed.  Sports Party preferences go to PUP and then the Liberals while that lot with the gimmicky name split theirs 2/3 to Labor and 1/3 to Liberal.

Oh, my apologies for being slow with the PUP BTL modelling.  A white knight kindly assisted by providing me with a great 2013 BTL file but I'm still having some computing power issues, and paid work on another election simulation earlier today has exhausted me too much to deal with it tonight.

Tuesday 9:40 pm (7:40 pm): Another update (69.35%) and Labor rebounds.  I am now running a version of the calculator in which I put PUP's vote over the tipping point by the barest margin possible and leave everyone else's vote the same. The latest update of this version (hereafter called "parallel calculator") has Labor down by 132, implying that the outcome if no more votes are added would be back to too close to call.

Tuesday 9:30 pm (7:30 pm): Another update and PUP are now only 711 shy of the tipping point on the ABC calculator. Labor are still losing more on the calculator's bottom line, but the gap if it is assumed the tipping point is crossed has not changed since the 8:15 update.

Tuesday 8:40 pm (6:40 pm): Liberal notional lead at 2504.  This currently exaggerates their real lead because of the tipping point (which PUP are getting closer to) but they're in front.

Tuesday 8:15 pm (6:15 pm): More figures added and the turnout has dropped to 69.18%.  During the 2013 counts there were cases in which the turnout displayed by the AEC regressed because of a technical issue (especially prevalent in the Tasmanian count) but in this case it could well be that vote changes have been caused by formality checking.  Anyway whatever has happened Labor's position after these changes is now very bad indeed.  Palmer United are still quite close to the tipping point (being only 1055 votes shy of it assuming all votes are ATL, which means that in reality they would cross before it) but the deterioration is most clear on the final distribution.  The Liberal lead has blown out to 2298.

Furthermore, if, when using the ABC calculator, I fiddle the PUP vote to put them over the tipping point, Labor now still loses.  The calculator margin in this instance is 1183 votes to the Liberals, but that's equivalent to about 900 votes at this stage of the actual count.

Unless the current figures are mistaken in some way, or are based on formality checking of only a part of the overall count (for instance), Labor is now in a truly losing position based on current figures - and it won't get any better from here.

Tuesday 6 pm (4 pm): Another trickle trickles in with 69.27% now counted.  Three hospital teams, one mobile booth, Perth PPVC and most of the BLV PPVCs still to add.  Still no non-ordinary votes added.  PUP's calculator position has dropped back slightly to 14.17% (meaning they need to gain 1180 votes on BTLs to cross before the tipping point and elect Pratt - there I'm using real quota figures) while the Liberal notional lead at the end of the count is now 1152.  At the moment Labor would still win (because of PUP's advantage on below-the-lines would easily exceed 1180), but today has been a bad day for Labor so far and this could be a familiar experience once postal vote counting commences.  By the way the Votes By Division screen looks like a useful place to keep an eye on what non-ordinary votes have come in when they do.

Monday 10 pm (8pm): Still plowing through the modelling but meanwhile the AEC has updated with 68.95% now counted (it will probably end up in the mid-80s).  The ABC calculator model shows the final margin at a meaningless 430 votes to the Liberals, and also shows PUP slightly closer to the tipping point at 14.20% (14.29% required, but the figure does not account for PUP's advantage on BTLs).  So Labor's position has improved and if this was all then they would win, but we haven't had the dreaded non-ordinaries added yet.  We're still waiting on the BLV PPVCs, 12 hospital teams, a mobile booth in Durack and the Perth PPVC, among booths that might have votes.  Also note that below-the-lines are now assumed formal on the night, but some small percentage of them won't be.

The comment by user Alaric at Tally Room shows what I am looking at here: in the 2013 election PUP tended to get preferences from BTLs from both sides of politics, and Alaric estimates PUP could gain 2800-4300 votes by the tipping point, compared to what would happen if all votes were above the line.  At present PUP only need to gain 872. If the expected blowouts in postal voting patterns materialise then PUP might end up needing to gain several thousand for the tipping point to activate, so the Liberals should still win if the differences between booth and postal voting in the past hold up.

Monday: I hope to do some detailed modelling using BTLs on the question of how much BTLs might be expected to improve PUP's position at the tipping point, given that it is likely PUP (and also Labor and Green) will go backwards in post-counting.  It may be that it turns out that the contribution of BTLs is too slight to overcome PUP's likely backwards drift, or that Labor go so far backwards that even the tipping point doesn't save them, but we will see.  Anyway this is going to take me a while.  Fortunately the AEC BTL files from 2013 are an excellent data source for modelling this question.  This is the primary task I hope to perform over the next few days: Take every party other than ALP, Lib, PUP and HEMP that contested in 2013 and for each, obtain the breakdown in their BTLs between the following: Reynolds, Pratt, Wang, lead HEMP candidate.  Then I can estimate the number of BTLs for each party and from that see to what extent BTLs might improve PUP's position at the tipping point, bearing in mind that PUP's preference flow should be somewhat better this time than last time given their stronger performance.  As Ben Raue has pointed out the distance between PUP and the tipping point if we assume all votes are above-the-line is projected to blow out from the current 0.11% to something like 0.5%, and that does seem like rather a lot.  (The below-the-lines that are distributed prior to that point could be worth about 1% of the total WA vote, but I'll have a more robust estimate later.)

Monday 12:35 am (10:35 pm Sunday):  I've now scaled the Liberals back from clear favourite to slight favourite for the last seat.  A reason for this is the tipping point mentioned below.  At the point where it is reached, Palmer United will currently have received only .1188 quotas in preferences if all votes are treated as above-the-line.  Labor will have received .2163, the Liberals .3773 and HEMP .4572.

However, we know these don't quite flow 100%; some of them are below-the-line.  PUP have the least to receive and therefore are the least exposed to leakage, while HEMP are the most exposed, with the additional disadvantage that their candidate is obscure and from out of the state.  PUP are therefore the most likely to overperform their expected preference gain to this point in the cutup.

PUP are only .0078 quotas off the tipping point on the calculator at the moment.  A lot of these preferences are coming from micros that typically have below-the-line rates in the 7-10% range, higher in cases.  Plus PUP have access to leakage from the Liberal, Labor and Green surpluses.

It's entirely possible that on current primaries PUP would actually be over the tipping point and Pratt would therefore win.  How this will pan out after a few weeks of postals remains to be seen (and if Labor drop back too far then even the tipping point doesn't save them).  The count should move in the Liberals' favour but I think they are actually starting behind.

Another issue for the Liberals in the current cutup will be BTL leakage from the Nationals ticket. 

Sunday 8 pm (6pm) Tipping Point:  Hitherto this count has been seen as a straightforward preference slog, but I've identified what seems to be a critical tipping point.  This concerns the point at which Dio Wang (PUP) is elected.  At the moment he is narrowly short of election after the exclusion of the Liberal Democrats (at which point only PUP, Labor, Liberal and HEMP are in contention).  He is then elected on HEMP's preferences, and a surplus created.

However, if I adjust the calculator input by increasing PUP's vote by a little over .1% and decreasing all others slightly, the outcome suddenly and sharply switches to Pratt (at the closest point, I found that an increase in the PUP vote of .01%, ie 100 votes, generates a 3300 vote swing in the calculated outcome). This happens because if PUP are elected on the exclusion of the Liberal Democrats with a tiny surplus, then PUP's preferencing of the Liberals has no impact, and the preferences from micro parties that are transferred when HEMP are excluded will all flow at full value.

But when HEMP are excluded, if PUP are still in the race, then the preferences of each of the Democrats and Katters Australian Party, which later split between Labor and Liberal, go to PUP and become part of the PUP surplus. This means their contribution to Labor is greatly devalued because the PUP surplus is dominated by the PUP ticket votes, which go to the Liberals.  The impact of this on the final ALP vs Lib outcome is thousands of votes.  (If PUP are still in the race when HEMP are excluded, then the half of the Motoring Enthusiast preferences that favours Labor becomes available to Labor rather than being trapped in a tiny surplus, but they are not worth much to Labor because of the same devaluation.)

This tipping point would provide hope for Labor except that the likelihood is that the PUP vote will go down after non-ordinaries rather than up.  But it is something to keep an eye on as numbers roll in in post-counting.

Sunday afternoon: Ben Raue points out that a problem for Labor is that the proportions of non-ordinary votes this time around is different to in 2013; absent, provisional and out-of-electorate pre-polls are way down and postals are up.  If it's the same people switching to a different method of non-ordinary voting then it might not matter, but it's quite possible (given that the turnout is down) that some voters who would have otherwise cast a non-ordinary vote just didn't bother voting at all.  And even if some of the same people are switching, non-ordinaries favour the conservatives anyway.  Raue projects a margin of .0297 quotas and William Bowe has .033 quotas.  Projected leads of around 5000 votes are close, but not that close.  It will have to be a lot closer than that by the close of primary counting for Labor to find something on below the line voting.

Sunday 2PP estimate:  Thought I'd have a go at translating the Senate results to a rough 2PP estimate.  This is a slightly tricky exercise; 4.7% of voters voted for parties that didn't contest any Lower House seats in 2013. Also it's possible that Palmer United took a greater share of its vote from the Coalition than normal (though I actually don't see any evidence to support this hypothesis).  Anyway using the WA lower house preference flow from 2013 where available, failing that the federal 2013 preference flow, failing that an estimate based on my view of how right or left leaning the party is, I get the vote share recorded as equivalent to a Lower House 53:47 to Coalition 2PP.  Which is really about what the polls were saying about federal Lower House voting intention in the state all along.  The difference from the polls to the result is that the PUP/Others vote rose and the Green vote didn't tank in its normal manner (if anything it rose too). So we could be seeing the "it's a by-election, no-one cares" factor when it comes to major parties trying to scare voters away from voting for the minors.

Sunday morning:  Morning all.  Thanks for the high level of interest last night.  Counting is currently at 68.72% completed and will probably end up somewhere in at least the mid-80s. Late counting last night saw the ALP vote continue to improve slightly and it now stands at 21.76%, but the Greens have dropped to 15.88.  Mainly as a result of the substantial drop in the Green vote, the Liberals are back in the ABC calculator lead albeit by a flimsy 629 votes.

While the division of Moore showed as having all polling places in, this is no longer the case.  A total of 90 polling places are still remaining to be included.  However when these are examined, it turns out they are:

* 14 BLV (Blind and Low Vision) booths including every electorate bar Moore.  These receive few votes.
* 15 Mobile polling places, including seven in Durack.
* 44 Special Hospital teams, covering every electorate but especially Curtin (8)
* Seven airport prepoll voting centres (PPVC) in the electorate of Swan. These all recorded no votes in 2013.
* Six divisional office PPVCs in Hasluck, Perth, Stirling, Curtin, Tangney, Swan. These all recorded no votes in 2013.
* Carnarvon, Perth and Christmas Island PPVCs
* The booth of Port Hedland (Durack)

Perth PPVC with 1471 votes is fairly large, Carnarvon and Port Hedland are around 500.  Christmas Island is actually in the NT electorate of Lingiari and its interstate PPVC didn't trouble the scorers last time and presumably won't contribute a lot.  The Special Hospital teams add up, however: these racked up over 7000 formal votes between them in 2013.

In 2013 the Coalition's final vote share outperformed its booth share (including PPVCs) by 0.27 points (Libs up 0.38, Nats down 0.11) while Labor's final vote share went down by 0.18 points, the Greens' by 0.11 and PUP's by 0.22.

Labor will need a markedly different pattern in non-ordinary voting this time around.  That's not impossible given that the Labor vote may have tanked late, but it would be surprising if non-ordinary voting actually favoured Labor.

As the count unravels it will be worth looking at the below-the-line voting patterns to see if they offer Labor any succour but I'm suspecting at this early stage that they won't.  There are a lot of comments about Labor voters voting below the line for Pratt but these only assist very marginally - to the extent that they slightly reduce the impact of any leakage from Bullock to other parties.  What Pratt needs especially is BTL flow from the various parties feeding the Liberal ticket, including PUP when they go over quota.

On The Night

Comments will scroll from the bottom of this section to the top.  Refresh for updates.

2:02 (12:02) Goodnight all! Just wanted to stay up past the end of daylight savings!  Not sure how much more there is to come tonight with 65% counted but whatever might be left won't give us much idea of the outcome in what could be a very close ALP - Liberal contest for the final seat.  If there is indeed no activity on Sunday I might take that opportunity to model postal votes as a guide to where things go from here.  For the moment the calculator lead Labor has is by no means a safe one.

2:48 (11:48) Count continues ... slowly.  Not sure exactly when they'll stop; early on I understood they were going til midnight.  Anyway with 64.5% counted there is still not much change with Labor's calculator lead at a far from safe 1.52% (14,111 votes).

2:27 (11:27) The signoff comments of Truth Seeker on modelling methods are well worth a read, especially the one about substituting for booth matching.  Truth Seeker's modelling of where the Labor and Greens votes could end up was more stable than the ABC's and apparently a better model.

2:21 (11:21) Still no change really.  63.3% counted and the calculator margin is currently 15,255.

Moore is the first electorate to be all done for the night.  Brand and Cowan have just one non-reporting booth apiece.

1:55 (10:55) It's not looking likely we'll know the result of the final seat tonight.  Labor's calculator lead hasn't made any progress in the last half-hour and is hovering around 14,000.  That sort of margin could easily be undone in postcounting if Labor fails to increase it tonight.

1:45 (10:45) With 60% counted some electorates may well be getting into their prepoll centre counting. Most electorates have only a handful of booths still to report.

1:21 (10:21) Labor has now jumped to a calculator lead of over 15,000 votes (nearly 2%) but bear in mind that this is on actual primaries and prior to postals on which Labor could well fall back.  What is probably happening is that as we get into bigger metro booths the Greens vote is soaring and pushing Labor up, since Labor get the Greens surplus.  The Greens' result is excellent ... but it is also another example of how the party does best when there is not much on the line.

1:03 (10:03) And there we have it; as the Labor vote approaches the giddy heights of 21.5%, the ABC calculator model has switched to showing Pratt as its prediction on current numbers.

1:00 (10:00) Truth Seeker has noted that his "median scenario" now shows 2-2-1-1 but that allowing variation from the current figures makes 3-1-1-1 more likely.  The question now is whether Labor can continue to improve as more votes are added, and reach a favoured position by the end of this very long night. (Apparently they're going to midnight western time!)

12:51 (9:51) Further improvement for Labor, now on the threshhold of crossing to the lead on the ABC calculator (0.6% behind; was briefly 0.3%).  Bit of a lesson here about dubious modelling as concerns some of the very low models of the Labor vote early on.

12:35 (9:35) ALP now trailing by just 1.27% (update a few mins later: now 1%) by the calculator.  I'm giving the seat simply as "undecided" above as there is still potential for Labor to improve. It is my view that the Labor campaign for this by-election has been abysmal (with the choice of #1 candidate the biggest single problem) but their chances of getting over the very low bar to elect two candidates should not be written off just yet.

12:22 (9:22) Labor have broken 21! The ABC (estimated) calculator is now online - note that it is a rough model only.  The bottom-line result there is only 2% between the major parties, and that's off lower primaries for Labor than at present.  It's possible Labor will close down this gap because of the defects in the projections that I keep mentioning, but Labor's tendency to perform poorly on postal votes compared to the Coalition won't help if so.

12:04 (9:04) Improvement in the Labor vote is slowing with nearly 26% now in the can (it's been around 20.6-20.8 for a while).  But they may yet improve a little more for the reasons noted.  Likewise as urban booths come in PUP are falling back, but not that rapidly.

11:51 (8:51) Once upon a time the big test for the Australian Labor Party was believed to be a primary with a 4 in front of it.  For the first time tonight the ALP primary starts with a 2! As noted in my previous comment I wouldn't be surprised to see it still rise a little from here.

There is some talk about micro party chances.  As a general rule while there is still no-one near the final vote we shouldn't write off a snowball by someone or other completely.  But it's generally not as easy as the simulations say.

11:36 (8:36) I'm having some trouble crediting, as projected by the ABC, that the Labor vote won't really rise at all from here.  There are still rural electorates where the small booths would be more conservative than the big ones but the small booths would have been counted first.

11:21 (8:21) I'm not seeing any basis for realistic doubt that Scott Ludlam will be what you might call "re-elected". Holding off on calling the PUP seat for now as they are tracking for below a quota of primaries.

11:11 (8:11) Truth Seeker projecting 3-1-1-1 (yikes) as most likely, followed by 2-2-1-1 with 2-1-1-1-1 an outside chance.  I believe those projections aren't adjusted for below the line votes (if this is correct, 2-1-1-1-1 becomes less likely still.)  Whatever the result Labor is tracking for an abysmal vote.  It can count itself lucky but take no comfort if it salvages two seats from here.

11:03 (8:03) More booths have come in in those electorates and the Greens projection is going down, though it is still extremely high. Curtin has a naturally high Green vote and some of the best Green booths in it are small.

As an overall comment on the votes counted so far there's a loose seat (beyond the so-far likely-looking 2-1-1-1) that we'll have to see a big change in trend or a lot more data before we know who might want to claim it.

10:55 (7:55) The huge projections for the Greens on the ABC computer seem to be driven by big swings from a few booths in in the electorates of Brand and Curtin.

10:46 (7:46) ABC now finding large swings against both majors to the Greens and PUP in projections.  Some other projections online are more muted.

10:38 (7:38) I am finding the evidence that PUP are travelling very well at this point to be compelling. They're on a whopping 13% raw primary now.  All the matching models are projecting them with a swing of around +6% at present.

10:34 (7:34) ABC now showing votes across the bottom of the screen that are actually their projections.

10:29 (7:29) Shooters and Fishers dropping as expected, now down to 2.5%.  No sign yet that any particular micro is threatening to poll all that much in terms of primary vote.

10:24 (7:24) ABC projection has swing against Labor, to Green (+6%!) and PUP, with Others down, but from a very small sample and without booth matching.  Treat with a fair amount of caution still, for reasons other than just the small sample size.

10:16 (7:16) More figures coming in and Sports Party dropping back to around 1%, with LibDems currently the second best performing micro.

10:08 (7:08) Sports Party polled another bunch in O'Connor.  This is actually interesting (if the first one wasn't an error) - everyone has assumed the Sports Party will get the same trivial vote that they did before, but they have had a lot of publicity from the first result.  If they can muster a few percent they might become a factor.  That said it seems that their support is very on and off in specific booths.

10:02 (7:02) Given that the votes in are largely from O'Connor, PUP seem to be going better there than last time.  However we don't yet know how representative the booths in so far are.

9:56 (6:56) Nearly a thousand votes in, mostly from O'Connor.  Shooters and Fishers are now leading the microparty race but they won't stay on 4.5% as less rural votes come in.  Sports Party no advance on the mysterious 18.

9:47 (6:47) Whoah! Sports Party showing as being on 18 votes out of 268.  These 18 are all from one booth in O'Connor.  Could be an error.

9:44 (6:44) Sam Dastyari (ALP) reporting suggestions of low turnout (often a bad sign for Labor.)

9:41 (6:41) The AEC has 47 votes in now, all from O'Connor and hence the heavy conservative skew is unsurprising.  A good start for HEMP with two votes!

9:40 (6:40) Still nothing yet bar the mystery eight votes.

9:15 (6:15) Oh this is silly.  The eight phantom (or are they?) votes have returned.  Hopefully just testing!  (The appearance of those eight votes on the bottom of the screen on the ABC is even sillier. Someone override the dumb computer please.)

9:10 (6:10) Senator Abetz saying it would be perverse if the Liberals lost their third seat having clearly won it the first time around.  He's right.  But it would also be perverse if both major parties continually ignored warnings about the critical defects in the Senate system that have contributed to putting us where we are. Yet they did.

9:05 (6:05) Polls have closed. No real results in yet (I'd give it, oh, 40 +/- 15 mins.)

8:50 (5:50) Reports of a broken ballot box at Maida Vale with votes spilling onto the floor.  Uh-oh.  These votes in quarantine.

8:40 (5:40) Results here: Virtual Tally Room.  Oddly, eight votes are showing despite the polls not having yet closed.  The votes are attributed to the electorate of O'Connor. (Update 15 mins later: they're gone now. Don't tell Clive; one of them was his.) (Update days later: these votes were cast in a different timezone, in which the polls had closed so they'd been counted, which explains the mystery.)

8:30 East (5:30 West) Polls close in half an hour.  Some bad news from Antony Green: apparently votes won't be reported to media by polling place tonight.  This means modelling during the count will have to make do with breakdowns by electorate, but I'd expect some serious skew in projections based on those (because small rural booths tend to report before large urban ones).

=====================================================================
Preview

Firstly my apologies for taking so long to get around to putting up anything about this weekend's sequel to the astonishing failed election for six Western Australian Senate spots at last year's federal poll (previously chronicled in WA Senate Squeaker).  A combination of remote field work (now finished) and focus on my home state's state elections have stopped me writing anything about it until now.  However this will serve as my one-stop thread for this weekend's by-election - some preview comments, whatever live comments I want to make, and any post-count coverage that may be needed.  Whatever I say live on the night will be by way of addendum to Antony Green's commentary on ABC News 24, which will be well worth checking out to see what kind of modelling approach he takes.  I may also be posting comments at Poll Bludger in reply to comments there.

A Senate by-election is a very unusual thing, and nothing quite like this one has happened before.  The Constitution requires half-Senate elections to be held without an accompanying House of Representatives election under certain circumstances, and these were held in 1953, 1964, 1967, and 1970.  However, these elections became notorious for their by-election-like feel.  The last three in sequence are especially instructive - from the state of the Senate after the last time the houses had been in sync (1961) the Coalition managed over nine years to shed four seats, with Labor losing two. The DLP gained four seats and Independents two. Yes, the DLP were supposed to take seats away from Labor.  A perception developed that governments that allowed half-Senate elections were asking for trouble, and governments ever since have usually tried to avoid them.  (A Hare-Clark state by-election in Tasmania in 1980 saw something similar, with the government of the day dropping a seat to a Democrat.)

This time it's a little bit different.  This single-state half-Senate election isn't the government's fault (except to the extent that both major parties have had ample opportunity to fix the Senate system mess over many years and haven't done it).  Whereas voting for minor parties was a possible release valve at the half-Senate elections in the 1960s, in this case it's voting for micros in a broken system that has got us into this mess, and there might be some reluctance to do it if it means potentially being sent back to the polls yet again.

To this stage, no recent public polling truly specific to the Senate by-election has been seen.  Senate polling is a fickle beast anyway, but perhaps without the distraction of a House election, it just might work this time. Instead, what we have to go on are indirect indications of voting support levels for the parties at Lower House level, both in WA itself and federally.

The national picture continues to show Labor leading the 2PP vote narrowly, with a swing of around 4.5% since the election.  My aggregate in the sidebar currently has Labor with a 51:49 lead, up from 50.4 two weeks ago following results of 52 (Newspoll and ReachTEL), 50 (Galaxy), 53.5 (Morgan by last-election preferences) and 49 then 51 from Essential.  The smoothed version of my aggregate suggests the national pattern has seen no lasting change yet this year.

Evidence that swings in WA are fundamentally different from the national picture is not that strong.  The Newspoll quarterly figures show the Coalition down from their 58.3% 2PP result at the federal election to 54%.  In primary vote terms this consists more or less entirely of a five-point movement from the Coalition to the Greens with nothing to Labor, but I'd be very surprised if it actually pans out like that on Saturday night.

Recent Nielsen samples have been a little weaker for the Coalition which has averaged 52% 2PP in them, and have also had the Greens vote slightly lower (at an average of about 13%).  Some private polling discussed on the Truth Seeker blog is also said to show a mid-teens Green vote but the Greens are perennial election under-achievers compared to their leadup polling.

Another issue in WA polling is the size of the Palmer United vote.  There are rumours it could be large, based on Morgan state breakdowns (which I would not place much weight on at all) and the aforementioned unstated-source private-polling.  Certainly PUP are strafing the landscape with expensive yellow to an apparently even bigger degree than they did in the Tasmanian state election.  At the latter, PUP underperformed most of their polling and a promising surge in support for them had collapsed by polling day, but it is hard to believe the WA PUP campaign could be quite as awful (nor as utterly bereft of rational appeal) as the Tasmanian one.

Of course, as we saw last September, what this or that major or minor party polls means very little if the micro-parties have a quota tied up between them and it flows tightly enough based on group ticket voting to elect one of their number.  There is a general vibe-of-the-thing sort of view that this time will be different and that we will not be seeing names like Dropulich, W rising to the top of the pile out of nowhere.  For the evidentiary basis for that view, see Glenn Druery. Mr Druery, who's been known to know a thing or three about this game, reckons that the Liberal Democratic cluster (LDP, Republicans, Smokers Rights, Stop the Greens) have made life hard for PUP by preferencing the Liberals.  I assume the basis for this would be that the LDP doesn't want any more PUP Senators on the crossbench since the PUP/Motorists voting bloc reduces its Senate power.

There has been some fuss about the chances of the HEMP party, which seems to have done the best deals of the micro-parties on the left. HEMP isn't even a blip in the Truth Seeker simulations, and in my brief experiments with the ABC calculator, I've found it hard to get HEMP close to the line.  One problem for them is that the Sex Party has preferenced Labor first.  I have found some theoretical chance for the HEMP party if it polls a good primary (say 2%) and the votes for both the Coalition and the Greens are down on current polling (say 41% and 11%) with Labor and PUP also mediocre (29% and 6% respectively).  But that scenario involves having, say, 13% of the vote between the various micros (which seems a bit high given the adverse publicity for micro-party voting), and even then it's a calculator output that's close enough to be undone by below-the-line voting.

Generally the utterly prosaic 3-2-1 (Liberal-Labor-Green) has been the favoured outcome for this election pretty much everywhere. It is only in the last week that 2-2-1-1 (with PUP depriving the Liberals of a seat) has started to get a close look.

For mine, the evidence underlying the PUP surge to above the 7% or so they seem to need is still sketchy, and I'd like to see it blessed by a public poll with a decent sample size. I may have more comments later this week, otherwise live comments and post-counting will be posted at the top of this article on Saturday.  (For eastern-staters, be aware things won't get serious until after about 9:30 pm eastern time.)

Other resources:
Poll Bludger thread and guide and pre-election thread
Tally Room guide
Truth Seeker simulations
ABC Elections

Wednesday 2 April: Truth Seeker now has an increasingly formalised WA Senate aggregate.  This currently has the Coalition just north of 41, Labor on 27, the Greens around 13.5 and PUP a bit over 7.  On this basis 3-2-1 is indicated as most likely based on the probabilities cited with 2-2-1-1 probably the next most likely.

Thursday 3 April: William Bowe's article for Crikey (possibly paywalled) raises two other prospects: 3-1-1-1 and 3-2-0-1 (with Palmer United winning a seat in either case).

I have found that 3-2-0-1 is viable if one assumes simply that the apparent polling surge for PUP is largely real and the well-documented polling surge for the Greens largely imaginary (or else out of date and now deflated).  In my limited experiments a PUP vote of around 9.7% or more to a Greens vote of around 10.8% or less is about the range in which it might start to happen.

To get 3-1-1-1 to fly I had to assume both the Greens surge and the PUP surge were fully real and take the Labor vote down to around 24 (eg Coalition 40 ALP 24 Green 14.6 PUP 10.5).  This would be basically a straight swing from both majors (more or less equally) to PUP and the Greens, which seems unlikely given the known federal data for WA as it would amount to a 2PP swing to Labor of only a point or so.

Some goalposts should probably be set in advance here: the retention of three seats can be regarded as a good result for the Coalition if it occurs, but by no means a miraculous one (and likewise, winning only two seats would be a disappointment but no disaster).  For Labor, winning only one seat this time would be a massive failure.

Friday 4 April: William in this Poll Bludger post adds another option to the table - yes, there is speculation even of 2-1-1-1-1; a picnic for the minors with Greens, PUP and HEMP all getting up.  (Hey, it happened in South Australia, but that's a bit different.) This would require not only the Coalition vote to be weak but the Labor vote to collapse into the very low 20s.  There seems to be a view about that this is actually a genuine risk, and you don't have to look far to see a reason for this: Joe Bullock.  Recent revelations involving Labor's #1 Senate pick have included brown paper bags full of union cash, an old conviction for assault, him calling Labor's rank and file membership "mad", and him not always voting for his own party.   It's also worth reading the putdown by former state MLA (ex Labor turned independent) Larry Graham. 

The betting markets (not that they deserve much attention after SA) are still tending towards 3-2-1.