Wednesday, September 11, 2013

2013 Federal Election Late Counting: Tasmania Senate

Tasmania Senate: Current Assessment
2 Labor, 2 Liberal, 1 Green certain
Final seat between PUP, Liberal, Sex Party
Current assessment: All three parties have realistic chances.  There is no clear favourite, but PUP are in my view narrowly the best placed. 

The AEC currently expects to declare the result (subject to challenges) at 2 pm Thursday.  The button is expected to be pressed on Wednesday 10 am. 


Final Result Thread: Final result and analysis will be posted on the new thread here.


NB A report published on Tasmanian Times (18 Sep) claiming the Sex Party has won was a bad case of premature ejaculation.  Counting has not finished, the button has not been pressed yet and the ABC Calculator is not an exact model of what happens when it does. TT has since acknowledged the unclarities in the situation.

Summary: 

I have added this summary (which will be edited as needed) because this article is getting very long and very complex.  What follows below is my full analysis of the current Tasmanian Senate late count, some of which is highly technical in nature.  The current situation is that three different parties - first Palmer United, then the Liberals, then the Sex Party, have been shown as leading on the ABC Senate Calculator at various points.  However the calculator assumes all votes are above-the-line ticket votes, and in practice preferences flow slightly more weakly than the model expects, especially if a party has lots of below-the-line votes.



Family First at one stage appeared to have an outside chance of beating the Sex Party and hence getting elected on a series of preferences, but they have made no progress in the post-count and that chance appears to now be gone.

Currently the Sex Party are being projected as winning comfortably, but that is because of a key point at the count where they are shown as beating Labor and then winning on a series of preferences.  However, because of below-the-line votes, they will not get as many preferences up to that point as the calculator suggests.  Their current calculator lead is just under 800 votes following a huge gain in very late counting.  At various stages I've estimated the lead required as somewhere between 700 and 1000 votes.  On careful analysis of the vote tallies now, I am not convinced that their lead is quite enough, but there is a significant chance that it will be.  That chance rests largely on the strong possibility of higher than normal leakage within the Labor ticket. 

Should they fail, the crunch point in the count comes when there are only three parties remaining: Palmer United, Liberal, Liberal Democrats.  This point is not showing on the current calculator.  If Palmer United are last then the Liberals win the seat, otherwise Palmer United win.  The calculator method shows the Liberal Democrats ahead by 1477 votes at present, but the Liberal Democrats rely on a lot of preferences to get to this point, and are more likely to shed votes because of below-the-line votes and their low profile in the state.  PUP will also shed some votes because of below-the-line votes, but in all probability not as many.

The size of gap needed on the calculator for the Liberal Democrats to exclude PUP at this point is not at all clear, and very difficult to model, but an extremely rough estimate is 1650 votes.  This estimate has an error margin of several hundred (if not more), meaning that the outcome is uncertain unless the gap is massively l above or below this figure.  On the current gap, PUP seem more likely than not to survive this exclusion assuming they are still in the race.

The impending "press of the button" (actually, execution of a computer routine) will determine the result.  The apparent margins are so tight that no amount of modelling has been able to determine which of the three parties will win with much confidence until the preference distribution is done.  It is also possible that some margins will be very close and that the pressing of the button may not be the final word on the result.

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

Note 1: you can follow late counting updates for various House of Reps seats here.   

Note 2: Analysis of this seat has been revised because of late changes in the ordinary vote totals and also a modelling error.  These combined affected PUP's modelled position by around 1100 votes, changing my assessment above.  Some out-of-date sections have been removed.  The latest assessments are always in the updates at the bottom.

Note 3: The Sex Party is apparently (as of late Monday) going to take the calculator lead for the final position very soon, when they pass Labor's total.  At this point the calculator will cease showing the gap between the LDP and PUP, which is probably the most important gap unless the Sex Party can gain enough of a lead to protect itself from below-the-line votes.  

These are the formulae for calculating totals for the two parties when this happens:

Liberal Democrats = LDP+SR+STG+KAP+DLP+FF+ASP+HEMP+SPP+S&F+NCT+AI+96*

Palmer United = PUP+SOL+AFLP+(PP-96*)+Greens+Labor-(3 quotas)


(* This number is approximate only and will change slightly)

Flowchart!  The fight for the last seat is so complex I've decided to put up a flowchart.  As usual I have no presentation skills whatsoever so if anyone wants to whip up a neater one feel free to send it to me.  These are just the most plausible scenarios; there might yet be others.  Note that "staying ahead" means in the real cutup, not on the ABC calculator, as the calculator is a crude model only. Click for a larger version:



========================================================================

(Note that this flowchart is now out of date - all Family First chances are gone and it is no longer clear that ALP staying ahead of Sex Party is more likely than not.)


Sorry to take so long about this but it's finally time for the obligatory post-count thread for the Tasmanian Senate! The on-the-night vote was quite a surprise because I never expected that of all the minor and micro parties vying for fragments of the Tasmanian vote, one of them, namely Palmer United, would actually get a massive head of steam up in the last week of the campaign and poll almost seven percent.  In pre-election testing I had often tested a PUP vote as high as the 4-5% range and found that their preference flow was not very good and that they invariably lost.  By the time (in the last few days) it became very plausible based on polling that PUP would get more than that, there was much too much else going on for me to notice it and check it out.

The effect of the enormous PUP vote has been that just about everyone else's preference snowball is trashed, because nearly all the preference snowballs relied on getting ahead of PUP and adding their preferences to a collection. 

The ABC issues a Senate Calculator here but it is only a rough model of the actual count.  There are at least two problems with using it to forecast the final outcome:

1. The calculator uses the vote as it is now as a base.  It does not attempt to model where the vote will be after the addition of all postal, prepoll and absent votes.  Parties may rise and fall significantly through the post-counting process, and in some cases even minor random fluctuations could affect the process. 

2. The calculator assumes all votes cast are above the line, meaning that preferences flow entirely according to party tickets.  In practice in Tasmania this is significantly untrue, with something like 11% (thank you all!) having voted below the line.  This particularly impacts on parties that depend on lots of preferences from other parties to build their way up.  It is the reason Family First lost the final seat to the Greens in 2004 despite being shown as winners by the calculator.  (In the end that result was not even all that close.)

While simple compared to some other state counts, this is Tasmania's most complex Senate count ever.  It is much tougher than 2004, which was effectively a race in two for which the modelling ground-rules were very simple.

It seems to me that something is About To Happen and if it happens some people may become Very Confused.  That is, that as postal votes are added, the calculator will soon "show" Palmer United Party as no longer winning the last spot, and the Liberal Party winning it instead.  This article tells you why that happens and why, at this stage, you should ignore it.

The reason for this is that the PUP vote has been steadily falling as postal votes have been added.  Their total of all "ordinary" votes (on the day booth votes plus pre-polls cast within the electorate) was 6.91% but their vote has now dropped back to 6.73%.  This is because PUP only built up steam in the last week or so of the campaign, and are not an incumbent party, so their postal vote is not much over 3% (and adjusting for the unequal rate of postal vote counting around the state does not increase it much.)

Those who were watching the calculator closely on election night would have noticed that at various points of the count PUP oscillated between winning and not winning, before the poor calculator decided it had had a gutful and started pretending that the party didn't even exist.  The main factor influencing whether or not the calculator showed PUP as winning was a certain point late in the count where three parties are very close together: (Click for larger version)


The Liberals, Liberal Democrats and PUP have each been shown as last here at different points.  If the Liberals are last, they are excluded by the calculator and elect PUP.  If the LDP are last, they are excluded and also elect PUP.  But if PUP is last, PUP is excluded, electing the Liberals.

As the PUP vote has tracked downwards in the post-count, they have moved ever closer to dropping into last place.  The calculator now has them less than 200 votes ahead, and soon they will probably fall "behind" the LDP and cease showing as winning in the calculator. (Update Wednesday morning: this will happen very soon as the "margin" is down to 17 votes.)

The other important point in the count - and not just because the narrow winner of my prestigious #54 vote is eliminated - is this one (again, click for larger):


At this point in the count, despite amusingly receiving thousands of votes via Family First, the Sex Party is eliminated.  However, there is a relatively small gap of 0.24% of the vote between the ALP and the Sex Party.  Should the Sex Party be ahead of the ALP at this point then Lin Thorp (ALP) is excluded here instead of them.  What happens then is a familiar story to those who've been following the interstate counts:

1. The ALP preferences elect Peter Whish-Wilson (GRN).
2. The Sex Party receives preferences from both the Greens and Labor via Whish-Wilson's surplus and jumps over the LDP.
3. The LDP is excluded with slightly more than half of its preferences going to the Sex Party and the rest going to PUP.
4. PUP trails the Sex Party and is excluded.
5. PUP's massive primary vote goes to the Sex Party which thrashes the Liberals for the final seat off less than 1.5% of the primary vote.

A freakish outcome that would be indeed - a four left two right scenario in a state that returned three Liberal and only one Labor MHR. 

When Is A Lead Not A Lead?

But the problem with all the above is that some Tasmanians can count past one.  At each step at which preferences are transferred, the receiving party can only count on the percentage of the votes that are above the line, while a given below-the-line vote may or may not flow in the same way as the party ticket. If we assumed that voters for all parties voted below the line to the same degree, then each party would only be sure of c. 89% of each preference transfer, while the remaining c. 11% would scatter to a large degree except perhaps where the two parties were very similar.

Looking at the Liberal/PUP/LDP tipping point, 90% of the Liberal votes at that stage were 1 Liberal, meaning that those that are below the line are only a risk if they leak out from the ticket while there is another Liberal still to preference.  (The rate of this happening is negligible.)  72% of the PUP votes at this stage are 1 PUP, with the rest coming mostly from Labor and the Greens.  But for the Liberal Democrats, only 24% of their vote at this stage is their own, with the remaining 76% coming from twelve different micro parties - Smokers Rights, Stop the Greens, Pirate Party, Stable Population Party, HEMP Party, Sex Party, Shooters + Fishers, No Carbon Tax, Aus "Independents", Family First, DLP and KAP.

These micros between them would, according to the calculator, be bringing c. 20,000 votes to the Liberal Democratic table.  But if we assume that 89% are ticket votes, the remaining 2200 are not at all guaranteed to flow to the Lib Dems, and they are exposed to possible leakage of those votes, especially as they are an obscure party compared to PUP and the Liberals.  Even if we assume that the Lib Dems keep 60% of these preferences with 20% going to each other party (a generous assumption for the Lib Dems), that would make their position relative to PUP 1320 votes worse than the calculator shows.

But even that's an underestimate, because micro-party voters are more likely to vote below the line than major party voters.  This may because they are more politically aware or fussy or may be because micro-parties are so prone to cut dodgy deals.  Based on the 2010 Tasmanian election, voters for micros may be 25% more likely to vote below the line.  Looking at SA and WA with similar length ballots last time, it's possible that figure is even an underestimate.  But keeping it conservative, I'd expect the LDP to shed at least 1700 votes from BTLs not following the ticket by the time they reach the three-way nexus.  Probably many more.  PUP's exposure is much smaller (probably about 750 votes worth, for a loss of 300 or so under similar assumptions to the above.)  I'd estimate 1400 votes as the minimum lead the LDP would need to have over PUP on the ABC's calculator for PUP to lose at that point of the count.  Probably much more.

[This paragraph updated] I originally tried to model where the count might end up by using the existing postals to project the remaining postals, and the prepoll voting centre votes to project the early prepolls, of which there are not that many in Tasmania.  I have not attempted to project the absent votes, which I expect will alter things little (perhaps a little in Labor's favour).  Even with this done I originally projected the LDP as only "beating" PUP at the critical point by 680 votes on the calculator, which would mean in practice that PUP are very likely to survive this exclusion, even if the calculator says they won't.  However, using the prepoll voting centres to project the early prepolls was a mistake in the case of the Labor vote (see 12 Sep update) and furthermore shifts in Labor's position have weakened PUP's position.

The Sex Party's Dirty Snowball: Can It Work?

The Sex Party snowball has the same issue as the LDP problem.  In my current projection [updated] the Sex Party beats Labor by a few hundred votes on the calculator.  But Labor's vote there is all its own with not a single ticket preference from anyone else, while the Sex Party is likely to be exposed to about 1250 (or more) below-the-lines at a stage in the count where they still have five alternative destinations.  The Sex Party not only will miss out on most of those BTLs but might even lose more of them to the ALP plus the Greens than it retains.  So, PUP has more room up its sleeve in avoiding defeat through this route than the party totals suggest.

(Original conclusion has been removed - out of date)

Update (11 Sep):  The event I spoke of is about to occur with PUP now only 17 votes "ahead" on the calculator.  However the cause was not postal votes but some of the last batches of ordinary votes, which bounced very badly for PUP.

The upshot of this was that my projection now has the LDP calculator "lead" over PUP with all votes in out to 1062 votes, which I would still expect to not be nearly enough (especially as it has been reported that there are 36,000 votes to be entered, suggesting a BTL rate more like 12% than 11%), but justifies caution about calling the race at this point, especially in case PUP performs poorly on absent votes for some reason.

The projection has the Labor lead over the Sex Party now at 550 votes.  As noted the Sex Party must beat Labor at this point to be elected and will need a substantial buffer to compensate for BTL assistance.

It is also worth noting that even under the votes as they now stand, Family First have come closer to winning than people may realise.  In one version of the data input for this model I mis-typed the projected No Carbon Tax Climate Sceptics vote at 0.58% instead of 0.058%.  The model responded by "electing" Family First.

Update 2 pm Sep 11: Here we go!  As expected the calculator now awards the seat to the Liberals

But as noted above the projected margin is much too small so at this stage PUP are still ahead. 

When statistics for above and below the line votes for different parties are available it might be possible to get more accurate estimates of what lead the LDP lead for PUP to be defeated.

Note that Antony Green discusses this situation here.  Two very minor points:

1. Antony writes: " Third, there is a higher incidence of above the line voting on the Green and Labor tickets than with minor parties, which means the flow of Labor and Green preferences to PUP is likely to be relatively solid." In 2010 the BTL rate for Greens voters was actually higher than for micro party voters overall in Tasmania.  Whether that is the case this year isn't clear (unless Antony knows something I don't!) but since the surplus PUP receives from the Greens is mostly Labor votes by value, the basic point stands - the votes PUP is getting are likely to be more solidly ATL than most.

2. Antony writes: "More counting is to come, and no doubt the calculator may flip-flop between Chandler and Lambie from now on." There may be some flip-flopping in the short term but I expect the calculator lead for Chandler to soon become locked in and to increase.

UPDATE 12 Sep: (Revised 4 pm) I took this update down for half an hour because of a glitch on the AEC website that briefly had figures turning into quicksand on the screen in front of me.

I may as well revise it while I'm at it.

PUP are now 281 votes behind on the calculator following the addition of more postal and some absent votes.  However as noted above, they would easily win on such figures and would have to shed at least 1000 votes (and that's a very conservative estimate) to be any real risk of being excluded before the Liberal Democrats from here. There are several items of good news for PUP in today's count:

1. Their share of postal votes is now 3.51%.  This compares with a figure of 3.24% in my model yesterday, a small difference but they all add up.

2. The number of postal votes remaining (as of now between about 4400 and 8000) is not large.  Most of the postal votes are already included.

3. We have not seen many absent votes included so far but those we have seen are not much different to the average for the electorate they come from.  Absent votes are difficult to model because we do not know where a given sample comes from at booth level but what we've seen so far suggests that PUP will not do badly on absent votes and that absent votes may even boost their position above my projection.

There is something else that I think is important.  The Liberal Democrats' lead candidate is Clinton Mead.  He is from NSW.  Nobody much in Tasmania has a clue who this guy is.  When there is a preference split between Lambie, Chandler and Mead, Chandler has a reasonable profile, Lambie has some profile, and Mead has none.  Voters voting below the line who prioritise candidates they have heard of will not preference the Liberal Democrats.  That said, if there are lots of voters preferencing a few candidates then donkeying the rest, Mead will do well on those.   It doesn't matter if Chandler is doing better on BTLs for profile reasons than Lambie, because the real issue is Lambie vs Mead; if Lambie stays ahead of Mead then BTLs aren't likely to be an issue.

Hopefully before much longer we will have a clearer idea of BTL rates for different parties.  These are still being worked on and the existing breakdowns of allocated votes won't be representative.

I think PUP are in a good position here.

Update (Sep 12, 10 pm) Sex Party Closes In, AEC Computer Unsafe, And So (sometimes) Is My Modelling!

Indi and Fairfax have missing votes; Tasmania has disappearing votes.  Now you see them, now you don't; a bit like our fox population.

Some may have noticed that on the current version of the ABC calculator the Sex Party is closing in, and is now only 233 votes behind on the calculator at the critical point at which the Sex Party can snowball to victory on Labor preferences if it gets ahead.  Meanwhile PUP are shown as 880 behind at the critical juncture with the Liberal Democrats, figures based on which, if the decline continues, things start getting sticky for them.  The figures showing have 86.02% counted, and are the same on the ABC and AEC sites.

The first set of figures surprised me as the ALP - Sex Party gap had been pretty steady at around 700 through the post-count.   I did investigate and found that the ALP had gone down on ordinary votes (which are now all at least roughly counted), from 33.1% when I did my projection to 32.9%.  That's a difference of 532 votes, which explains a lot of what is going on, and indeed puts the ALP quite close to falling behind the Sex Party on the calculator. 

The ALP trending to fall well behind the Sex Party could make things interesting, because the Sex Party is not as vulnerable to below-the-lines as the Lib Dems (since the crunch point comes sooner in the count).  Put Labor, say, 1300 behind at that point and Tasmania may get Senator Robbie Swan.

While trying to work out what was going on, I looked at the divisional figures for the electorates, and found a repeat of the problem from earlier today.  The figures currently showing are wrong.  They are wrong because earlier today there were as many as 4972 postals from Denison, 4730 from Franklin, 4366 in Lyons, 4258 Bass, 3424 Braddon, that had had their results tallied and added in.

Looking at the current postals there have been minimal shifts in most electorates, but almost two thousand postals that were counted in Denison have disappeared from the online electorate tally.  That's the second time today that this has happened (there was an earlier such glitch affecting most Tasmanian electorates).  Something is apparently causing new figures to be overwritten by old ones. 

Looking at the postals counted in Denison on the current count, Labor has 35.85%.  Assuming the remaining total break similarly, their position improves by 57 votes compared to the calculator when the other 2000.  The Sex Party performs worse on Denison postals than state ordinaries, but one of its feeder groups, Australian Independents, performs much better.  So probably Labor's position in the current calculator is only understated by about 50 votes at the key point - which doesn't make a huge difference; the Sex Party really are closer than they were yesterday.  

And there's more.  At this point it's time to post the dreaded Reversed Ferret, which appears whenever I get something wrong and have to change my modelling.  It's especially fond of Senate elections, for some reason:


When I attempted to model Labor's likely performance on out-of-electorate prepolls yesterday, I looked at within-electorate prepolls and found Labor were getting about 30.8% in those.  But I have also noticed that in 2010 Labor performed 8.4 points worse on out-of-electorate prepolls than on ordinary votes, a much bigger difference than the 2.3 points in within-electorate prepolls, and enough to propel the Sex Party to about a 300 vote calculator lead, all else being equal.  And indeed, checking 2010 closely, it turns out that within-electorate prepolls weren't a good model for out-of-electorate prepolls that time either (the Labor vote was 5.5 points different.)

The projected lead for the Sex Party over Labor still isn't enough to get them the seat, because of the high proportion of BTLs they would have to hold in a congested field.  They would need to improve further than expected as the count continues.  But at this rate it is very possible that the ABC calculator will soon see its third different leader!

And there's still more!  The hit to Labor's ordinary vote, whatever caused it, has also caused a blowout in PUP's position, because Labor votes feed into PUP, so whatever Labor loses, PUP loses too.  So PUP have blown out on the calculator to trail by 880.  Throw in what I have discovered about Labor's bad performance on out-of-electorate prepolls (which adds another 600), and consider there are still more postals to come (including the vanishing Denison batch), and the LDP's eventual lead (assuming the Sex Party are eliminated) creeps up to something like 1700 -  at which point it's quite unclear PUP have enough advantage from the leaking micro-party BTLs to get over the LDP and win. It becomes necessary then to look finely at who has what numbers of BTL votes and whether there are enough to make a difference to a quite substantial gap.

PUP's position, which I thought was very strong yesterday, is now under siege on two different fronts.  The outcome is very unclear and it may even be that it remains unknown until the completion of data entry and the hallowed pressing of The Button.

Update (13 September): Denison back on board

The missing Denison votes have returned (for now) with 86.58% counted and the Labor over Sex Party calculator gap has, as predicted, widened slightly.  It now stands at 307 votes.  I will be logging the percentage of vote counted now and then as insurance against disappearing numbers.

1 pm 13 September: 87.14% counted now; Labor "leads" Sex Party on the calculator by 282, Lib Dems "lead" PUP by 959.

Update 13 Sep: And Then There Were Four

Just when you thought three contenders was confusing enough, it is my melancholy duty to inform you there are four.  Family First's Peter Madden is struggling because a number of FF's early feeder parties polled too poorly while one of their key feeder parties (PUP) polled much too well.  But he's not completely out of it just yet.

A 0.3% probability of a Family First win popped up on the Truth Seeker simulations today, but considering how the post-count is likely to go, I think it is more than that, although it's still a remote chance.

This is the crunch point for Family First:

At this point on the current calculator Family First are excluded.  Even if I rig the numbers to get them over the Sex Party at this point, they are still excluded behind Labor at the next point - but only by a few hundred votes, and Labor could well lose that in post-counting.  Assuming Family First actually beats Labor at that point, then this preference snowball takes off:

* Family First gets over the Sex Party.  When the Sex Party is excluded, their own preferences to their credit go nowhere near Family First, but those they have received from Shooters and Fishers and Stable Population Party do.

*Family First gets over Labor.  Labor are excluded and their preferences go to the Greens.  Whish-Wilson is elected and both Labor and Green preferences now flow to Family First in preference to the Liberal Democrats, Palmer United Party and Liberals.

* Labor and Green preferences put Family First over the Liberal Democrats.

* And now virtually every micro-tiddler whose preferences pooled with the Lib Dems, excepting only the Sex Party and the Pirate Party, goes to Family First.

* PUP are excluded way behind the Liberals and Family First.

* PUP preferences go to Family First.  Family First wins!

So how likely is this?  At the key point, the Sex Party has its own vote plus the above-the-lines of the Country Alliance, HEMP, Stable Population Party and Shooters + Fishers.  Family First has its own vote plus the above-the-lines of No Carbon Tax, Aus Christians, Aus "Independents", AFLP, Senator On-Line and Rise Up Australia.  Each has whatever below the lines it has picked up from these parties and also from Smokers' Rights, KAP, Stop the Greens, the Pirate Party, the DLP, within-ticket leakage from the majors (which will be negligible), the Australian Republicans and Andrew Roberts.  I put Roberts in bold because he is easy to forget about as an ungrouped independent, but he does have 267 below-the-line votes, and is about as anti-gay as Peter Madden.  It's quite possible his preferences will favour Family First. 

The calculator's gap between the Sex Party and Family First was 931 votes on ordinaries but it has now shrunk to 770.  The main reason for this is that the Sex Party have a low postal vote compared to their booth vote, and a possible reason for that is that their vote grew in the last week on the back of awareness spread by yours truly that they were one of the few parties not preferencing Family First.  [Edit: See the comment below by Mark Duffet and my response - it turns out this is a national issue and not just a Tasmanian thing, although it is (so far) strongest in Tasmania.] If the Sex Party have a low postal vote, they are likely to have a low early out-of-electorate prepoll vote as well.  So the gap may reduce further.  And looking at the mix of parties whose below-the-lines contribute to this race, well over half the votes come from parties more likely to be sympathetic to Family First than the Sex Party.  So it is just plausible that Family First can bridge this gap, although it's a big ask.  Also unless Labor slumps even more than expected in further late counting, it's likely that Family First's high dependence on below-the-lines will keep them behind Labor.  But on current trends, even Family First can't be ruled out.

Any advance on four?

13 Sep 8pm: Now with 88.34% counted with most of the Absent votes counted today.  Labor and PUP did well on these.  The key figures by the calculator are:

* Sex Party over Family First: Slight increase to 808.  As discussed above this will probably come down when more postals and early votes are added, and probably overstates Sex Party's real lead.

* Labor over Sex Party: Slight increase to 365 (update: later became 408).  As discussed above this will very likely come down and may reverse when early votes are added, but it definitely understates Labor's real lead.

* Liberal Democrats over PUP: Slight increase to 1039.  As discussed above this understates PUP's real position and indeed is presently not likely to be enough to exclude them, but it is likely to increase.

The bad news for PUP here is the margin behind the Lib Dems still increasing even despite the inclusion of a vote type on which both they and Labor did well.  A part of the reason for this is that just about everyone except the Liberals does well on absent voting (so the thirteen parties in the LDP's collection outperformed their ordinary votes).  But the main reason seems to be that with further checking PUP's share of the total ordinary vote has come down again!

In the last day and a half, PUP just haven't taken a trick.  They still have a serious chance because of below the line votes, but the question of whether the BTLs are enough to beat the Liberal Democrats lead or not is much more serious than at first.  As it becomes more serious, the Liberals' chances of bagging their third seat grow.

Update midnightish 15-16 Sep:  A very small number of votes were added sometime on Sunday.  With 88.54% counted the calculator has Sex Party clearing Family First by 811 (probably an overestimate), Labor clearing Sex Party by 345 (certainly an underestimate) and the LDP clearing PUP by 1004 (still not yet enough to stop PUP winning.)

Midday 16 Sep: Another half a percent is in the can and with 89% counted the calculator has the Sex Party clearing FF by 833, Labor clearing Sex Party by 256 and the LDP clearing PUP by 1146.  Still the case that if the button was pressed right now PUP would win, but we are still to see any early pre-poll votes counted and they will probably change the last two figures dramatically.

3:30 pm 16 Sep: We are starting to see the first Early (Prepoll) votes counted but with a very large percentage for the Greens and round figures for all of Liberal, ALP, Green (200, 180, 160) I greatly doubt these will be representative of this vote type and would advise against projecting anything off them until they settle down a lot.  Anyway for what it's worth the calculator has margins of 836, 235, 1105.  The Sex Party may well take the calculator lead very soon, but remember that that in and of itself means nothing because they will lose votes because of below-the-lines.

4:30 pm 16 Sep: 89.41% counted according to the ABC (the AEC web updating is probably lagging) and at that point the calculator margins are now 839, 222, 1127.  Just bunkering down for the pandemonium that will break out when that 222 becomes zero!

5:15 pm 16 Sep: 89.51% counted and the calculator margins are 830, 210 and 1118.  So far Labor are performing less badly on prepolls than I expected but Palmer United are performing worse; these two things are close to cancelling each other out in terms of PUP's net chances.  I have posted the calculator formulae for the LDP vs PUP stage at the top of the page so it can be followed when (as now seems extremely likely) the Sex Party takes the calculator lead.

8:00 pm 16 Sep: 89.79% counted and the switch is not far away: 863, 187, 1126.  Antony Green makes a useful point here about the influence of candidates without ATL votes (Roberts and Australian Republicans) on the calculator.  The correct quota figure advantages the Sex Party relative to Labor by close to 100 votes, and the LDP relative to PUP by close to 150, on Antony's calculations.  This partly cancels out the disadvantage on below-the-lines each group experiences.

10:30 pm 16 Sep: Nearly 90% counted and excitement builds with the Sex Party a mere 81 votes from taking the "lead".

The early-vote prepolls counted so far are unrepresentative.  They come from Denison and Franklin only (slightly more from the former.)  If all early-vote prepolls obeyed the pattern of the Denison and Franklin sample then PUP would win easily, but as it is things are less clear.  Labor and the Greens are PUP's biggest feeders and they will both go down in further prepoll counting - which is now most of the remaining votes.

Midday 17 Sep: 90.43% counted.  The Sex Party's calculator lead over Family First is increasing, out to 894 now, so the Family First scenario is becoming even more remote.  ALP over Sex Party has increased slightly to 95 votes; LDP over PUP is at 1155. No prepolls have been added yet today.

1 pm 17 Sep: 90.63% counted.  More prepolls added, this time from Denison and Braddon.  It is the prepolls from Bass and Lyons, yet to be included at all, that are likely to most hurt PUP. The ALP over Sex Party calculator lead is now down to 81 votes while LDP over PUP is 1140.

2 pm 17 Sep: Glitch Returns - for at least the third time in the count, the postal vote tally has returned to an earlier level and hence the total showing has gone backwards and is out of date.  In this case postals from Lyons are affected.  Whenever the calculator and/or the AEC page show a %counted well below the latest I have listed that usually means the figures currently showing are wrong.

3 pm 17 Sep: More votes are being added but about 4000 votes previously included remain missing from the current count.  I will be offline for a few hours this afternoon and if necessary will synthesise figures later, but hopefully the problem will be fixed before then.

5:30 pm 17 Sep: Looks like we are back to normal with 91.32% counted.  And it has happened - the ABC calculator shows the Sex Party with an absolutely meaningless 1-vote lead.  As noted above the Sex Party needs a hefty lead to cover for below the lines:


The Family First - Sex Party gap is still not closing so it seems FF are just about out of it now.  As for the critical LNP-PUP gap, you can't see it on the calculator anymore; I have it at about 1381 using real quotas (which would show on the calculator, if it did show, as 1244).  It's growing steadily, but it's still far from insurmountable.

What goes on when the Greens' surplus is distributed at that point showcases one of the odious quirks of the undemocratic Inclusive Gregory system used for Senate preference distribution.  At that point, the Pirate Party's votes form part of the Greens' surplus along with those of the Greens and Labor.  But because surpluses are distributed by ballot papers rather than by the previous value of those ballot papers, the Pirate Party's votes get shrunk in the distribution, being effectively worth only 1.31% of the Green surplus instead of a fair 3.52%.  This disadvantages the Liberal Democrats, hence advantaging Palmer United in their fight with the Liberal Party, although the Pirate Party ticket actually preferences the Liberals ahead of PUP.  This is, of course, completely unfair and wrong, but Inclusive Gregory is not likely to become a headline electoral issue any time soon.  More's the pity.

======================================================================

Below The Lines Take Shape: Things Get Even Messier!

(18 Sep 1 am)

This bit is even tougher slogging than all the stuff above - wonk factor five here folks.  It doesn't change that much so feel free to skip it.  But it does give some further insight into the bizarre tactical complexities of a system in which the Liberal Party's best hope of winning a seat is that enough of its voters voted Liberal Democrat by mistake, and where Pirate Party voters best carry out their party's actual preferencing intentions by leaking around them than by following them - among other strangenesses.

 In the current count of 322,232 formal votes, a total of 291672 (90.5%) are ticket votes.  Given that we have heard that the below-the-line rate is about 10.5-11% I surmise that it is actually slightly less than that and that virtually all remaining Unapportioned votes (presently just over 10,000) are below-the-lines, additional to the nearly 20,000 already allocated to a candidate below-the-line.

On that basis I get the following estimates for numbers of below the lines (with percentage BTL  in brackets)

Roberts 290 (100), Aus Rep 31 (100), SPP 132 (39), SOL 35 (38.7), Pirate 490 (26.1), Green 9427 (25.6), Country 184 (21.3), KAP 242 (18.9), Aus Christians 184 (21.3), STG 219 (15.9), RUA 151 (15.7), FF 629 (15.2), HEMP 226 (14.1), NCTCS 22 (11), S+F 376 (10.9), Sex 505 (10.9), PUP 1613 (7.56), Smokers 57 (7.4), Labor 7730 (7.3), AFLP 49 (7.3), DLP 179 (7.2), Aus Inds 169 (7.1), Liberal 7299 (6.0), Lib Dem 249 (3.3)

These might be a bit out but should be a fair indication.  We can see that ALP, Liberal and PUP have low rates of below the line voting and the Greens and most of the micros are higher.  An exception is the Liberal Democrats, because they get most of their votes from people mistaking them for the Liberals.

I calculate the number of below the line votes in the parties whose preferences flow to the Liberal Democrats, excluding the Pirate Party, at about 2755. For each of these votes, two things matter: who it goes to (assuming it doesn't exhaust) out of the LDP, the Liberals and PUP; and whether it goes through the Greens first.  Each vote that goes to the Liberals without going via the Greens weakens the LDP's lead by one vote compared to its calculator position.  A vote that goes to PUP without going via the Greens is a loss of two.  Any vote that goes through the Greens reduces its own value and increases the total value of PUP-friendly Green preferences by one, so it's a loss of something approaching two as well.   So there's a potential here for the LDP to not only drop votes equivalent to its 2755 supposedly incoming BTLs here, but to lose more than one vote of its calculator lead per vote.

The calculation for PUP's below-the-line vulnerability is even harder.  It's necessary to think very carefully about what an ALP, Green or Pirate vote does. 

When an ALP below-the-line vote is distributed on the exclusion of Lin Thorp, it can go to any of PUP, the Greens, the Liberals or the Liberal Democrats.  If it was an above the line that either came from Bilyk or came from Brown via Bilyk, it's worth about .116 of a vote.  If it's an above the line that came from Brown without going via Bilyk, it's worth about .54 of a vote.  But if it's an above the line that is either Thorp's own or Dowling's own, it's worth a full vote, and these ones are 39% of Labor's below the line total.  So that's potentially about 3000 votes that can leak at full value, and maybe another six or seven hundred votes worth of leakables at this stage.

However each of these votes has the choice PUP, Green, Liberal, LDP, and it's only a leak if it goes to one of the latter two ahead of both the former.  It costs PUP's position relative to the LDP a vote if it goes to the Liberals, and two if it goes to the LDP.  I expect not that many Labor below the lines would do either, but that those that would would be more likely to go to the LDP (eg a Labor voter who disliked all of PUP, Green and Liberal).  

When we come to the Green vote there will be the Greens' 9427 below the lines flying about, plus below the lines that came from Labor (thus matching Labor's above the lines).  There are also those that came from other parties, but nearly all of these either favour the LDP over PUP or are neutral, and were therefore covered above.  The important issue here is that the Inclusive Gregory system is interested in ballot papers not values, and the Greens' tally at this point includes over 126,000 Labor and Green above-the-lines.  The ALP/Green below-the-lines are only going to be about 9% if that of an ALP surplus worth about 7000 (plus whatever comes from other parties to the Greens).  So while the Greens have a massive below the line vote, PUP's vulnerability to them is unlikely to be more than about 700 votes' worth.  These are mostly Green votes that have a choice between PUP, Liberals or Liberal Democrats.  Not too clear which way they will break since diehard Greens voters may well regard all three alternatives as crazy, or may have a slight preference for PUP because of its position on asylum seekers.

Lastly, there is the Pirate Party.  This is an odd one, because their above-the-line preferences (error fixed, this previously said below) go to the Greens and then to the LDP.  But in doing so they increase the value of the Green surplus, which consists mostly of PUP-friendly preferences, and hence they inadvertently benefit PUP, to the tune of nearly a vote per vote.  This is also true of any Pirate below-the-line that goes to the Greens before the LDP or Liberals.  The most damaging thing a Pirate below-the-line can do to PUP is to go to the LDP or Liberals without going via the Greens first. (This hurts PUP compared to the calculator by nearly two votes in the first case and nearly one in the second).  On the other hand, for every Pirate vote that goes to PUP without going to the Greens, PUP gains fractionally.   My guess is that the Liberals will see barely a thing from the Pirates by this method, but the LDP might be a different story as the party shares some civil liberties territory with the Pirates. Anyway, the below-the-line Pirate vote of 490 is not to be sneezed at, so here is an opportunity for leakage to cause harm to PUP.

Some rough estimates then:

LDP losses from micro-parties relative to calculator position: 3500
PUP losses from ALP below-the-lines: maybe 900
PUP losses from Green or ALP-Green below the lines: 600
PUP losses from Pirate Party below the lines: 250

That's putting the break-even point at about a 1750 vote calculator lead for the LDP, which is about where it's currently looking like it might end up, but with a great amount of uncertainty in these estimates.  But there is one thing I don't consider in this analysis and that is ballot order design.  Some voters will fill out parties they know at one or both ends of their ballot and then donkey those in the middle.  That may lead to the LDP, who are second on the ballot, doing better than my modelling predicts. 

In terms of the Sex Party - ALP situation, there number of BTLs going to the Sex Party is very much as I expected, so nothing has changed.  They probably need a lead over 1000 and I don't think they are going to get it.

In conclusion, looking closely at below-the-line votes has not greatly changed my assessment that the race between PUP and Liberal for the final seat remains open.  It has just become even more difficult to model it accurately, and don't be too surprised if there is still not a clear favourite leading into the press of The Button. 
======================================================================

18 September

Latest update has 92.4% counted.  Sex Party over FF is out to 946, and Sex Party "over" Labor is 117 (which as has been explained is not nearly enough to win).  I make the LDP-PUP gap at what is most likely the most critical point now 1591 (using real quotas not calculator quotas). We have still seen no early prepolls from the electorate of Bass and I expect that these to further increase the gap.  It's possible they'll do so to a point where the Liberals can be said to be slight favourites.

3:30 pm 92.78% counted.  Sex Party over FF is 952. Sex Party "over" Labor is 158, up slightly but still nowhere near enough.  LDP-PUP gap is now 1498.  A slight gain for PUP but still no Bass prepolls.

6 pm Note on Labor within-ticket leakage: Reader Chris has done excellent analysis of the possible impact of within-ticket leakage from the ALP ticket (mainly from Carol Brown's substantial ATL vote) which he estimates at 200-250 votes.  This not only affects the ALP in their race with the Sex Party (but not by enough on current figures to cause the ALP to get excluded - though it might approach that level with further counting) but also affects Labor as a preference feeder for PUP when Lin Thorp is excluded.  I've reduced my estimate of the calculator gap needed for the LDP to exclude PUP by 100 votes, but the estimate is extremely rubbery.

6:40 pm Family First Apparently Out Of It Now: There was an error in my figures analysing the BTLs for FF vs ASP last night; with that fixed, the BTL liabilities of Family First and the Sex Party are not that different.  As Family First is continuing to make no progress on the Sex Party's lead over them it now appears extremely unlikely that they can clear the gap and catch their train to the top. 

19 September

Midday: 93.05% and we have seen the first early-vote prepolls from Bass included, though at this stage they are under-represented in the count with 750 added.  These were unfavourable to Labor as expected, but the Sex Party - ALP gap on the calculator isn't moving much; it's currently at 141 which is nowhere near enough.  The LDP-PUP calculator gap has surprisingly closed to 1247 votes, boosting PUP's chances of winning.

5:30 pm: Please excuse the slow pace of updates today as I was distracted by a state poll.  There is not far to go now with 93.81% counted (in 2010 it made it to the low 95s) and the Sex Party's calculator lead is up slightly to 219 (by the ABC calculator) which becomes 317 using real quotas, which I will try to do from now on.  Still not nearly enough but if they can pull out a few hundred more by the end of the count then they will still have chances.

The LDP-PUP gap is at 1450.  It's increasingly likely that even as it goes "to the button", no-one will know who will win!

8 pm: 94.1% counted and very little left to throw.  Antony Green has also been tracking the totals and has the Sex Party's calculator lead at 392 and the LDP-PUP calculator lead at 1542.   The Sex Party continue to gain but are probably not going to get enough of a margin to hold off the BTL problem.  The other point remains way too close to call.

8:30 pm: We've got another update and it's 94.24% in the can.  By my calculations there is a maximum of 1.16% left to throw.   The Sex Party's lead over Labor has taken another jump to 480 (using real quotas) as Labor continues to struggle in late counting.  Considering the Sex Party have jumped about 250 votes (using real quotas) over the last one and a bit percent, if they can do that from the remainder they may go into the button press with a realistic chance.  However, the votes added during that percent and a bit included a lot from Bass which were bad for Labor. 

As for the crucial LDP-PUP margin, I have it at 1569.

Leakage From Second Candidates:  An aspect I haven't discussed so far is the possibility of leakage from the second PUP candidate, who has 276 votes.  I noticed that Antony Green in his analysis is treating only 50% of the second candidate's vote as locked-in for their party, and this surprised me, because I thought it would be much more than that (based on my experience with state-level Hare-Clark).

When I looked at actual results from 2007 and 2010 for Tasmania, I found that the BTL hold rate was sometimes around only half, but that generally applied to candidates with very low votes.

The x-axis here is logarithmic, so 1 = 10 votes for second candidate, 2 = 100 votes, 3 = 1000 votes (etc).

Based on this graph the LDP (at about 1.6 on the x-axis since their second candidate currently has 40 votes) would expect to keep just over half their second candidate's votes.  But PUP (at a bit over 2.4; their second candidate has 276 votes) should hold over 80%.  PUP will leak more from their second candidate, but maybe only by a few dozen votes. 

There are good reasons for this.  The less a Senate candidate has polled, the greater the proportion of their vote that is likely to be from (i) their close personal friends voting for them as a person and then across party lines (ii) voters voting for them for more or less random, whimsical or strategic reasons. And also the more a Senate candidate has polled, the more other parties will have been excluded by the time they themselves are excluded.  This relationship applies even though those second candidates excluded later are more likely to have non-primary votes from others in their pile.

=======================================================================

20 Sept

11:30 am: A few more trickles (not sure where these are from) and we are at 94.3% with, as far as I can tell, not more than 1% to go.  The Sex Party's calculator lead over Labor is down slightly to 461 (using real quotas).  (This bit noted the LDP/PUP gap at below 1400 but I suspect that was an error on my part at the time and that it was actually 1561.)

12:10 pm: 94.35% counted on the ABC site but it appears to be ahead of the AEC's updates so I don't have the exact quota.  Sex Party over ALP  by about 473.  LDP over PUP by about 1571

2:10 pm: 94.4% counted, Sex Party over ALP 488, LDP over PUP 1581.

Are The Sex Party Getting Enough?  Not quite yet (I think!), but it's becoming an increasingly interesting question as their calculator lead grows in post-counting (now 495 with LDP over PUP 1573).  See the comment on this issue from reader Chris (below).   The point Chris makes about Pirate preferences "leaking" to the Sex Party is, I think, especially important. 

At the point where the ALP/Sex Party contest is resolved, there could be around 3800 BTL preferences from 18 excluded groups (two of which are BTL-only) floating around.  I'm assuming unapportioned are all BTL, and I'm ignoring who the ATL preferences of these groups are for.  All of these votes must find a home out of Labor, Liberal, Sex Party, Green, LDP and PUP, or exhaust in the case of voter error.

1100 of these expected votes belong to left or at least leftish micros - Pirate Party (the biggest contributor in this category), HEMP, Aus Ind, SOL, Stable Population Party.  Actually I am not sure what sort of audience Australian Independents appeal to; possibly similar to Democrat voters, so I am not sure their preferences will be that leftish, while SOL's BTL vote is negligible.  But Pirate and HEMP BTL voters are very likely to preference the Sex Party much more heavily than the ALP, although this will be muted by their tendency to also preference the Greens.

The remaining votes belong to three classes of parties: Christian Right, likely to preference the Liberals (c. 1300), rural/trad user right (c. 1000) and anti-green/libertarian right (a few hundred).  Labor will probably do better than Sex (yes I know, that's a Hunter S Thompson book title) out of the middle category and perhaps out of all of these, but this will be watered down by competition from the Liberals and PUP.

To see what might happen when some of the left-libertarian micros are excluded, in 2010 in South Australia the Secular Party was excluded with 373 BTLs, nearly all its own.  Of these 119 went to the Sex Party, 93 Green, 26 Labor, 17 Liberal, 14 Liberal Democrat and 104 to a large list of parties not in the Tasmanian count (about half of these to the Democrats or SOL).  So it's quite conceivable the Sex Party will get at least 30% of HEMP or Pirate BTLs.

I think it's plausible that the Sex Party will outperform Labor across all these ATLs considered collectively.  Not by much if so, maybe 100-200 votes, but throw in the prospect of Labor losing another 300 to leakage from their own ticket, and a "locked in" gap of 600-700 (as per Antony's model, the assumptions of which concerning Labor ticket leakage can be argued various ways) is looking a little bit fragile, especially if the Sex Party can get another hundred or so in late counting.  They are certainly not out of this yet.

23 Sep: 

Counting of the few remaining dribs and drabs has resumed and we are up to 94.62% with less than 1% to go.  Sex Party over Labor is now at 523.  LDP over PUP is now 1597.  Both figures are continually increasing and the question now is how high they can get based on the few votes left to count.

The Forcett Anomaly: The Truth Seeker blog in comments here notes an oddity in current figures involving the Rise Up Australia party, which I have investigated further.  Rise Up Australia has 47 unapportioned (votes that are not necessarily either above or below the line, but at this stage most likely the latter) votes in the whole of Tasmania.  This includes 1 unapportioned vote in Bass, 7 in Braddon, 1 in Denison, 2 in Franklin and 36 in Lyons.  Of those 36 in Lyons, 30 are in a single relatively small polling booth of Forcett, giving Forcett a Senate RUA vote that is completely out of whack with the House one.  Furthermore, the Forcett booth has 29 more Senate votes than House of Reps votes.  No it's not the residents of Forcett suddenly deciding that Danny Nalliah is right about the causes of bushfires, and nor do the RUA candidates have family in Forcett, and nor did a busload of evangelicals happen to vote in that booth.  Rather it appears to be a simple clerical error; someone has typed 30 instead of 0.  If this is true, then the impact on the overall count is that 30 RUA votes do not exist, and the quota is at various times either 4 or 5 votes too high, which means that the Sex Party lead over Labor should be adjusted downwards by 8-10 votes, and the LDP-PUP gap by 12-15.  At present the quota difference is 5, so SP vs ALP is really 513 and LDP vs PUP really 1582.  Obviously this will be sorted out at some stage when an attempt is made to allocate those votes only to find they don't exist. 

3:30 pm 94.74% counted now and Sex Party's calculator lead is 530 (adjusting for the Forcett issue) and LDP-PUP is down (!) to 1520 (ditto). 

4:50 pm 94.76% counted and the 30 non-existent Forcett votes have been removed!  SP lead is 526 and LDP-PUP has dropped again, now down to 1493

8:15 pm:

We are now at 95.24% which must be very very close to the end, and all the BTLs are allocated.  What has happened in the last half a percent - wherever it has come from - is that the Sex Party's position has improved hugely to a calculator lead of 795.  Is this enough?  It's hard to say, but they now have a serious chance, and not just an outside one anymore.  Later tonight I will crunch some of the BTL numbers and see if I can shed further light on whether this is enough to win or not.

As for the LDP-PUP gap, should it come to this, it's 1469.

BTLs entered:

Antony Green here has been tracking what he calls "locked in" votes for candidates Thorp and Swan.  I am going to again take a slightly different tack and identify exactly what votes each candidate has and what below the line votes will be thrown that are committed to no candidate.

Robbie Swan has:
* 4285 Sex Party above-the-line votes
* 547 below the line first preferences
* Above the line preferences from: SPP (215), CA (717), SOL (60), AI (2287), AFLP (664), NCTCS (185), HEMP (1444), S+F(3241).

Total 13645

Lin Thorp has:
* 2516 Lin Thorp below the line votes
* At bare minimum 8236 ALP above the line votes from the surplus of Catryna Bilyk. (This is the figure that Thorp would receive if all Brown #1s and all Bilyk #1s exited the ticket). 

Total 10752 votes

There is then the question of BTL votes within tickets.  Robbie Swan has access to 39 within ticket votes of which I expect him to retain at least half, although this is a negligible issue.

Lin Thorp has access to:

1055 John Dowling votes.  On past patterns (which may not be reliable) Thorp should retain 90% of these.

5333 below the line votes for Carol Brown and Catryna Bilyk.  If all are received as part of the surplus their extra value compared to the bare minimum scenario is (10771-8236)=2535, however their contribution to that value varies depending on their history. On past patterns (which may not be reliable) Thorp should retain 93.3% of these at their variously reduced values.

On past patterns Thorp gains 3314 from the rest of her ticket.

Assuming the above are correct then Thorp has an effective lead of 401 not considering below-the-lines.

Assuming that Labor's preferences flow more weakly to Thorp with holds of 85% from Dowling and 90% from the rest of the ticket (as a result of the lower BTL rate, Tasmanians' habit of cross-ticket voting and the unpopularity of Thorp) that lead drops to 265.

The following votes then affect the outcome if they go to one of Labor or the Sex Party:

* leakage from Liberal, Green, PUP, LDP minor candidate below the lines.   Of most importance is potential Green leakage which could be worth around 150 votes and could well go to either of these parties.  The rest is unlikely to have any real impact.

* Labor votes assumed to have leaked (276 in the as-normal scenario)

* Below the lines from, in order of numbers:

Family First 743, Pirate Party 502, Shooters + Fishers 455, Andrew Roberts 332,  Katters Australian Party 291, Aus Christians 290, HEMP 270, Country Alliance 230, Aus Inds 206, DLP 203, Stop The Greens 161, Rise Up Australia 160, Stable Population 157, Fishing and Lifestyle 65, Smokers Rights 65, Senator Online 44, Aus Republicans 34, No Carbon Tax 26. 

As discussed before there is every reason to think the Sex Party will make progress on the BTLs of the Pirate Party, HEMP, SPP, and possibly Aus Inds.  Whether Labor make progress depends on whether voters for the conservative parties preference Thorp below the line to significant degrees or just send their votes to the Liberals.

I did assess a 700-vote calculator lead as being likely to be about enough for the Sex Party but when I look at the above figures I'm not convinced they're likely to overhaul Labor based on these BTLs.  It is however close.

It is worth noting that if the Sex Party do eliminate Labor at this point, they are not quite home and hosed.  There is a decision point later in the count with PUP where the Sex Party has a calculator lead of 2195.  I am currently examining this.

Tues 24 Sep:  As readers may note I've cooled on the Sex Party's chances following close analysis of the figures - their chance is realistic, but not as high as I first thought when the surprisingly large 795 vote margin came out.  The main thing needed for them to win is a very high leakage rate inside the ALP ticket and I'm sceptical that it can be high enough, especially with the Inclusive Gregory system making the votes of leakers worth less than the votes of ticket-followers.  I now think PUP are the best placed of the three but there are so many uncertainties that I can't even assign them a 50+% chance of winning.

The decision point I mentioned above involves the Sex Party, PUP and the Liberals.  If the Sex Party beat Labor then there is a point where they are 2195 votes on the calculator ahead of PUP.  However assuming there is no leakage from Labor or the Green BTLs then this comes out as a lead of over 1400 on ticket votes.  Leakage from Labor or the Green BTLs will knock votes off this but PUP would then have to outperform the Sex Party by at least 1000 based on all the BTLs thrown to that point, in a situation in which they can also go to the Liberals.  I don't think this would happen but it is worth noting for the sake of completion that this decision point exists.

Final Result And Analysis Over Here.

=========================================================================
Who I voted for!

I like to declare my vote after every election.  In the House of Reps I voted for Andrew Wilkie.   In the Senate I voted for the Pirate Party, in my view the only non-illiberal party on the Tasmanian ballot to escape from this ridiculous Senate election with most or even any integrity intact. My preference strategy was designed to cause maximum possible damage to Family First.  With FF irrelevant, my vote (after first assisting the Sex Party) at one point helps the Liberals beat PUP, and at another helps PUP beat the Liberals, and I am uneasy about both these things.

============================================================

34 comments:

  1. Just wanting to thank you for your work here. Invaluable resource.

    ReplyDelete
  2. I noticed earlier today the fluctuation of the sixth Senate seat, but the calculator is a very imperfect tool with only GVTs to predict preference flows, especially in Tasmania or the ACT where in excess of 20% of formal votes were made below the line in 2010. Also, I noticed that for some reason there were small numerical differences between the data the ABC was getting from the AEC and the calculation of the provisional quota – probably insufficient to change the way the elimination process ensues, but I tweeted Antony with my observation and I presume from the lack of acknowledgement it isn’t significant enough to bother fixing. His article today was pretty good, and I hope people listen to the message it sends.

    ReplyDelete
  3. PUP votes show similar levels of BTL to the Greens.

    ReplyDelete
  4. One unfortunate thing (for Australia's sake, though not for democracy's sake) is that too few voters are aware enough of the situations to vote tactically either above or below the line. For instance, the ALP has zero chance of a third seat and it's been well known that they had no chance of a third seat the whole campaign. If just 500 additional Labor supporters had been aware enough of this to preference Swan above Thorp (or to tactically vote Sex above the line), Tasmania would be getting a 4th progressive Senator. Instead it looks like they'll fall just short and end up with a third conservative. While that is far more representative of the wishes of the Tasmanian voters, it's not necessarily a good result for federal government policy.

    ReplyDelete
  5. Not to discount your contribution, but another likely reason for the Sex Party gaining during the last week of the campaign is their video getting a run on ABC TV's Gruen Nation during that time.

    ReplyDelete
    Replies
    1. Sorry for slow clearance of the above comment; the board software didn't send me the notification email (first time this has ever happened).

      Yes it turns out there's a national trend in the Sex Party performing much better in booth voting than postals. From 1.73 to 2.43 times better in the mainland states. Tasmania is currently running above even that with booth voting at 3.10 times postals, but the preferencing issue isn't necessarily the cause.

      Delete
  6. Good to see you in the ABC making informed projections rather than the usual guesswork! http://www.abc.net.au/news/2013-09-16/last-tasmanian-senate-spot-still-too-close-to-call/4959666

    ReplyDelete
  7. This is a bit of a text dump, but the following are ATL/BTL splits by party. Hopefully the formatting plays well with Wordpress.

    Party TIcket BTL Percent
    Country Alliance 627 200 24.2%
    Liberal Democrats 6737 555 7.6%
    Australian Labor Party 93294 11126 10.7%
    Rise Up Australia Party 7801 156 2.0%
    Liberal 106296 12908 10.8%
    Palmer United Party 1867 3 2300 11.0%
    Australian Christians 1198 300 20.0%
    DLP Democratic Labour 2195 284 11.5%
    Pirate Party 1308 522 28.5%
    Australian Greens 2569 8 10356 28.7%
    (HEMP) Party 1296 270 17.2%
    Shooters and Fishers 2865 531 15.6%
    Australian Independents 2117 230 9.8%
    Sex Party 3902 640 14.1%
    Senator Online 55 36 39.6%
    Katter's Australian Party 947 290 23.4%
    No Carbon Tax 161 30 15.7%
    Australian Fishing 588 72 10.9%
    Stable Population Party 190 133 41.2%
    Australian Republicans 0 29 100.0%
    Smokers Rights 666 85 11.3%
    Family First 3357 721 17.7%
    Stop The Greens 1090 248 18.5%
    Independent 0 276 100.0%
    Total 281061 42298 13.1%

    ReplyDelete
  8. Um. Slight! error in the above. Rise Up Australia only 768 votes ATL.

    That said, a little follow up analysis.

    At the elimination of Shooters & Fishers, the following can be noted:

    Sex Party - 8880 votes secure, 1134 votes off party below the line (insecure)
    Lib Dems - 11635 votes secure, 907 votes insecure
    Family First - 8244 votes secure, 824 votes insecure
    Greens - 27006 votes secure, 522 votes insecure.

    Family First should still lose that next vote - but not by as much as the ATL indicates.

    Sex Party is not likely to hold a lot of its insecure vote given the sources- maybe 50% (~750) - gives Labor a bit more room than it seems.

    Of the 6000+ effective votes flowing from Labor and Greens to Palmer, some 900 of those are insecure, but are the only insecure votes Palmer has. So this contest (lib Dems vs Palmer) has similar number of insecure votes on each side.





    ReplyDelete
    Replies
    1. Apologies for the delay clearing the above - I am still having the problem with not getting emails for all comments.

      Delete
  9. Has there been any detail on how the BTLs are breaking released so far? I'd be very interested to see if Lin Thorp suffered any preference leakage, given her defeat the last time she faced the electors and her seeming unpopularity as a minister. Is there any chance that Labor first prefs could leak away from Thorp and give the Sex Party the lead on the ALP-ASP break point? Or is there no way of knowing that until they press the button?

    ReplyDelete
    Replies
    1. My past experience is that the rate of within-party BTL leakage in Senate voting tends to be very low indeed, although it does happen to a very small degree. If there is some then Thorp is potentially at risk of leakage on the BTLs from Brown and Bilyk but those get reduced in value (greatly in the case of those that go through Bilyk). Likely to have <100 votes impact and probably much less.

      Delete
    2. Having gone back and looked at the data from 2010 (didn't have the time to analyze all the BTL ballots, but the general DoP trends), it's possible Labor may leak more than 100 votes.

      In 2010, Polley lost 6.6% of BTL papers to other parties, and Urquhart lost 6.8% of BTL papers (her own first preferences and 1 Polley 2 Urquhart BTL). The division of BTL votes is broadly similar this year, with about 60% of Labor BTL voters giving first preferences to Brown or Bilyk. With an expected transfer value of around 0.55 for Brown and 0.10 for Bilyk, and the same leakage rate as last year, Labor will lose around 176 votes to other parties or exhaustion.

      This doesn't account for the fact that Labor couldn't leak much to the Greens in 2010 because Milne had already been elected, but this year can leak to Whish-Wilson. That could increase the number of Brown and Bilyk voters that leak to Greens. Labor's BTL rate is down from 15.3% to 7.4%, and my hunch would be that people who shifted from BTL to ATL would have been more likely to vote 1-2-3-4 Labor when they did vote BTL, so the remaining BTL voters may have a higher leak rate.

      I don't know what influence Labor having a 4th candidate will play. My prediction would be that Dowling's BTL vote is largely people who voted 4-3-2-1 Labor, and so almost all of it will flow to Thorp and not other parties when he's excluded.

      My guess is that all this put together means that Labor may leak somewhere around 200-250 votes. That's almost certainly not going to be enough for the Sex Party to pass them unless those votes are leaking directly to the Sex Party (doubtful). The final count may come down to the button, but from the looks of things Labor should hold on and Swan will be excluded.

      Delete
    3. Nice work. Yes, the inter-ticket leakage from Labor will be higher than I thought. There will also be some leakage from the Liberal ticket, which I haven't taken into account as the Liberals are not really in any specific close race (and looking at 2010 it was not worth much for Labor anyway.)

      Delete
  10. You wrote: "Lastly, there is the Pirate Party. This is an odd one, because their below-the-line preferences go to the Greens and then to the LDP."
    Did you mean "above-the-line" in this sentence? I don't understand how they can direct BTLs.

    ReplyDelete
    Replies
    1. Yep, this will be fixed. Lost count of the number of times I've said one when I meant the other!

      Delete
  11. I'm getting total BTL at 7.38% Unapportioned at 2.05% (Midday 19th). So your information on BTL percentages may be out, or may have been changed by the addition of postals etc.

    Suggests a maximum BTL of ~ 9.4%, not counting any votes yet to be added.



    ReplyDelete
    Replies
    1. Yes when I added it up two days ago in preparing the BTL analysis it came to slightly less than 10% as well (at that time slightly over 30K total of unapportioned and known BTL). It does seem that the early publicly-circulating estimates of 11% then 10.5% and 36,000 BTL entries were a bit high.

      If it goes down much further, ie if a significant portion of the remaining Unapportioneds are not BTLs, then I'll have to revise my estimates as concerns the LDP-PUP gap. Which seems to be really where the action is as the Sex Party isn't building enough of a lead and probably won't at this rate.

      Delete
  12. Just wondering if there is precedent or provision for a recount because a 'critical count' is very close? A few votes at an intermediate stage could mean 2 different people elected in WA or 1 of 2 other candidates in TAS, but the final result might not be close enough to indicate a recount could change things.

    ReplyDelete
    Replies
    1. For the House of Representatives there is an automatic threshhold - a margin under 100 votes is automatically recounted. Margins above that may be recounted depending if there is good reason. For the Senate there are general provisions (see http://www.aec.gov.au/elections/candidates/candidates-handbook/recounts.htm) but there is no specific cutoff that I am aware of. So there is certainly provision. I expect that if a <100 vote difference occurred between competing micros at a very early stage of a Senate count there would be reluctance to recount it (unless extremely close) given the reduced chance of an error where fewer ballot papers were involved.

      Delete
    2. Cheers for that. I asked Antony Green as well and here's what he added:

      "A re-count can be ordered if any critical cut-off would have made a difference. All the ATL votes would be brought to the central scrutiny centre and re-counted against the original tallies done in the offices of the Returning Officer. All informal votes rejected without being data entered would also be checked. The new totals would then be entered into the computer for each polling place and the preference distribution done again. If it is still close they can start checking the data entry of BTL votes as well. A close scrutiny of the tally sheets on the distribution of preferences would also be done to ensure there are no errors.

      It is a rare process but was undertaken for two Victorian Legislative Council Regions in 2006."

      Delete
  13. How is it possible for the ABC site to be ahead of the AEC's?

    ReplyDelete
    Replies
    1. I would expect they had some kind of CSV feed from the AEC and that in cases the update went through faster on their site than on the AEC's.

      Delete
  14. If the Labor preference flow below the line is similar to that of the last time Labor ran 4 candidates, in 2004, the Labor-Sex race could be even closer than it looks.

    Having a 4th candidate did not cause as much leakage in 2004 as I thought it would, as over 97% of votes that went to the 4th candidate (Wells) ended up back with the third candidate (Price) after Wells was excluded. I'd guess this is because most of those votes were 4-3-2-1 Labor. However, Polley lost nearly 6% of her BTL votes to other parties, and O'Brien lost over 9% of his (largely to Christine Milne). With the Greens #1 having less than a quota this time, we may see a similar leakage number this year.

    If the flows are the same as in 2004 (w/ O'Brien=Brown, Polley=Bilyk, etc), Labor will leak 256 BTL votes using the provisional quota. Thorp would be left with 14,189 votes, plus whatever she steals from other parties below the line. Singh stole 1689 BTL votes in 2010, but that was a good year for Labor, with Greens #1 already elected, and with 20% BTL votes. I would be surprised if Thorp steals more than a couple of hundred votes.

    At 14,189 for Thorp, Swan would need to hold just 45% of the BTL votes the calculator's crediting him. That might be tough (I doubt he'll get many BTL prefs from Shooters, F&L, or Country Alliance), but it's far more reachable than the 75-80% that Antony Green's most recent blog post suggests he needs (Green's model presupposes that Thorp will hold 95% of Brown and Bilyk BTL voters, and I doubt that very much).

    There is also the possibility of Swan stealing BTL votes that the calculator sends to different parties. For instance, I think Swan will see a substantial share of the 479 Pirates BTL votes rather than those going to Whish-Wilson (isn't that what your ballot did, Dr Bonham?). As an aside, 25% of Pirates voters went below the line, quite impressive given the <10% overall, and the fact that their ATL process is by far the most transparent and representative in the country.

    I think that the Sex-ALP count will ultimately be decided by fewer than a hundred votes either way, and it may be too close to call right up to the moment the button is pushed. At this stage, I think Thorp will beat Swan, but the Sex lead keeps growing on the calculator and as it gets to 500+ territory I think Swan's chances grow exponentially with each additional vote. If Labor's BTL leakage rate is much higher than 6%, this race tightens significantly, and if it's anything above 9%, I think the Sex Party has it.

    The fact that it's so close at a key break point and that it's hard to make even educated guesses about how BTL votes will transfer, let alone create an effective or accurate model, is what makes this scenario so interesting. I think even with less than 1% of the vote to count we still may be waiting for a week to see who wins.

    ReplyDelete
    Replies
    1. Thanks for the analysis. I have been watching this blog closely and both the work of Kevin and the commentators has made for an exciting time.

      Sex Party are going to sneak home

      Delete
  15. Excellent analysis Chris. I've biffed another hundred off my estimate of the minimum SXP need to win, and made lengthy comments in the main article.

    ReplyDelete
  16. You may have mentioned it in your article (sorry if i've missed it), but, is the 495 lead based on actual quota or calculator quota? Actual quota will be, of course 50 higher, which reduces the ALP vote by 100, and increases the SXP lead by 100.

    ReplyDelete
  17. Actual quota. I switched to using actual quota not long ago and said I'd try to keep figures in actual quota from now on.

    ReplyDelete
  18. Kevin, do you have a rough estimate for how far LDP would need to be in front of PUP for Chandler to feel comfortable?

    ReplyDelete
    Replies
    1. I don't think Chandler can feel really comfortable with any size of lead that's actually going to happen based on the few remaining votes to throw. There are so many uncertainties that even if it was, say, 2000, the Liberals would have reason to feel that they had the better chances, but not that they were by any means safe. I can't rule out that even 2500 could reasonably be overturned, although I do not think it's likely.

      Delete
  19. I'm surprised Sex made such a late run to make things interesting. Looking at the numbers, it's likely Labor will win by <100 votes, unless Labor leakage is higher than in the past.

    Based on 2004 preference flows, Thorp will get 14,088 votes from Labor voters (surpluses + Thorp and Dowling first preferences). She'd also pick up 8 votes from Liberal surpluses, giving 14,096, and leaving Swan 451 behind before the other BTLs are considered.

    I don't know if Swan can make up 451 votes with the BTLs available. That would be 38% of the BTLs from Sex 2, HEMP, Independents, SPP, and Pirates--if Labor gets no votes there whatsoever. If Labor gets 10% of those votes, Sex would need 48% of the BTLs from those 5 parties, and to at least tie Labor on BTLs from the other parties. I think Greens would pick up a sizable number, so I don't know if Sex can pull that off.

    I think the key number would be a 10% leak rate from Labor 1 BTLs. If Labor leak less than 10%, I think Labor wins; if they leak more, I think Sex pulls it off. I think it's unlikely, though certainly not unthinkable, that Labor leaks 10%, so I think Labor will pull it off by a margin of around 50 votes.

    ReplyDelete
    Replies
    1. It's the "and to at least tie Labor on BTLs from the other parties." that seems more likely than not to be the killer for me. Even if the gap on BTLs is at the lower end because of high Labor leakage, making it likely Swan closes the gap on the surplus of the left parties, I do not see how he can avoid dropping votes to Labor trickle by trickle on the preferences of the right parties. Family First preferences, for instance, will go to Labor to some small degree but virtually none will go to Swan. Even if the loss is, say, a hundred votes, I'm not convinced Swan can get enough from the sympathetic parties to take the lead by that much.

      I am a little surprised that with the gap as high as 795 he doesn't come out as more competitive but I guess that is down to the increase in the size of the count.

      Delete
  20. The right micros may not be as bad for the Sex Party as one might think.

    At the 2010 Victorian Senate election (the election with the most similar mix of parties to this election that I could find), the Sex Party gained 22.92% on Secular Party, 15.98% on Senator Online, 6.53% on Building Australia, 14.68% on Climate Sceptics, 7.68% on One Nation, 16.86% on Shooters, and 2.85% on LDP (though with no LDP candidate remaining). By those numbers, I mean that of all the non-ticket votes available, the Sex Party won that percentage more. For instance, the Sex Party got 1258 of the BTL votes held by the Shooters at their exclusion, and the ALP gained 572, of the 4070 total BTL votes, for a 16.86% gain by the Sex Party. If that happens with the Shooters ballots in this election, the Sex Party will gain 70 votes over Labor, and with Fishing and Lifestyle, a further 12 votes.

    Labor does better with the religious minor parties, but not significantly so. The ALP gained 1.71% on Family First and 3.46% on the Christian Democrats. The DLP candidate ended up elected so I don't know how it went.

    I'm still not confident the Sex Party will get enough to pass the ALP, but that's substantially better with the right/libertarian BTLs than I would have thought. Sex will still probably need to do very well on Pirate, Hemp, Burnet and Ann BTLs plus have a 8-9% leak from Labor, and I don't know if they'll get it, but I do think it's going to be exceptionally close.

    We could end up with the interesting situation of Labor scrutineers trying to get ALP ballots ruled informal in a recount so that the Sex Party can pass them, which would be pretty funny, if not exactly truly democratic.

    ReplyDelete
    Replies
    1. Those are very interesting figures. Sounds like there is some proportion of micro-party vote that just goes BTL to a big bunch of micros in order to demote the majors. SEC and SOL I would expect to be "left" but gains on Shooters and ON are surprising. Perhaps right-wing micro voters don't perceive the Sex Party as all that "left".

      Delete