Sunday, September 8, 2013

2013 Federal Election Late Counting - House Of Reps

This post will follow late counting in the House of Representatives. The Current State of Play will always appear at the top of the screen, followed by Fairfax updates, some general comments from the early history of this article, and the newest non-Fairfax updates at the top of the section below that.

Current State Of Play (This will be changed when status changes)

Last updated 31 Oct (Fairfax)

Election result is 90 Coalition 55 Labor 5 Other (2 Ind 1 Green 1 KAP 1 PUP)

Fairfax: What Happens Now, and Past Close Finishes

31 Oct: It's over! (At least to the point the bookies pay out at; it remains to be seen if the LNP want to risk antagonising the PUP leader by taking this to Disputed Returns.) Clive Palmer will be declared the winner of Fairfax by 53 votes at the end of a massive struggle over an electorate that was not deemed worthy of a single published seat poll.

That officially concludes this thread, which has been the most visited thread on the site so far.  Thankyou all for your interest.



29 Oct: Palmer now up by 58.

28 Oct: The first real reversal in the recount, but is it too late? Palmer's margin is 57.

25 Oct: Palmer's margin is 67.

24 Oct: Palmer's margin is 64.

23 Oct: Palmer is well ahead now.  He leads by 66.

22 Oct: Palmer's lead is now 50.

21 Oct: It's remarkable how much this continues to flow in the same direction.  Palmer has gained another five and O'Brien has lost six, so Clive Palmer now leads by 52.

18 Oct: Today Palmer lost three votes but O'Brien lost five, so Clive Palmer now leads by 41.

17 Oct: Three in a row for Palmer; today he gained five and O'Brien lost five so Clive Palmer now leads by 39.

16 Oct: Another day of losses for O'Brien who has shed another eight with Palmer gaining one.  Clive Palmer now leads by 29.

15 Oct: There is movement at the station with a shift of 13 votes in Palmer's favour (he has gained four and O'Brien has lost nine.)  Palmer now leads by 20.

14 Oct: The process continues (slowly!) and the patterns are much the same as last time.  Have not yet updated the graph.  Still no change to the overall standing of the candidates.

10 Oct: I have updated the graph below.  The challenge patterns discussed below continue but there have been a couple of booths in which the authenticity-challenge rate is (for whatever reason) much lower than in previous booths.  I've seen unverified comments that the authenticity challenge issue relates largely to the question of initialling of ballots and Palmer's beef that in theory anyone could initial a ballot.

9 Oct: Updated figures here confirm that every vote challenged for authenticity at three booths has been referred upwards.  The number of other challenged votes being referred upwards in these booths is 7.4% of the total vote cast.  I have not yet updated the graph below as I think the pattern is clear enough. :)  There is an ABC report on the recount here with background on the Palmer authenticity challenges, but a fair few of the figures in it are incorrect.

8 Oct: The recount continues with the Palmer crew evidently continuing to claim the majority of opposing ballots are fake and continuing to appeal such challenges when turned down to the state AEO.  I've updated the graph below.

4 Oct:  The recount is underway with details posted here.  The recount is being slowed down by the making of an unusually large number of challenges to the authenticity of various ballot papers, and these challenges seem to be then being referred on to the state AEO in huge numbers.  In fact, there is no evidence in the current count that any district-level decisions about these authenticity challenges are being accepted by the scrutineers making those challenges.  The beauty of the level of detail released by the AEC is that it enables us to get a handle on which side might be behind this behaviour by looking at the rates at which it occurs by booth:

Updated based on fresh data 10 October
There is an extremely strong correlation (explaining 67% of variation) between the two-party preferred result for the LNP against PUP, and the percentage of ballots that are challenged for authenticity.  It would seem therefore that the Palmer team are claiming that ballots cast for or benefiting their opponent are in many if not most cases potentially fake.  The LNP might also be making some challenges of this kind but there is no clear evidence of this in the data released.

Either the LNP are also challenging the formality (or assignment to a given candidate) of some ballots that benefit PUP (as would be expected) or PUP are challenging some ballots that benefit them (one never knows ...)   This can be determined because at most booths the total number of challenged ballots exceeds the LNP 2PP vote, in cases by up to 14 points.  And there is at least one booth so far, Buderim North, at which the number of votes referred upwards from district level exceeds the LNP vote.  The data are compatible with the PUP team referring all their challenges upward from district level and the LNP team referring some, though the former is not necessarily the case.

But I'd suggest we have a tactic from PUP, not normally seen in such recounts, that is blowing out the number of challenges considerably, adding to the time required and very likely the cost for the recount.  In the 2007 McEwen recount only about 650 ballots were challenged above district level.  In this recount 2945 ballots have been sent to the top already - and that's out of less than a quarter of the total vote!

1 Oct 5:00 pm:  The AEC confirms that only the votes used in the distribution of preferences have been rechecked thus far.  That does not include primary votes for Palmer and O'Brien, all of which will need to be rechecked in the recount, presumably with further decisions about the formality of specific votes.  Based on what we have seen so far, Palmer's seven-vote lead is very fragile.

We should not be surprised if there is considerable back-and-forth progress in the recount, depending on the order in which the votes are checked.

1 Oct 4:45 pm:  A rather odd headline to this report by Lania Scarr.  The report says Palmer "has fallen significantly behind" but at no point has the official count during the distribution of preferences shown this.  He's just been less ahead than he was.  The report does however confirm that AEC "scrutineers" (meaning official staff, not party scrutineers) have deemed ballots informal and that this is the cause of changes.  Palmer himself continues making noises about the count, suggesting that candidates may have been wrongly elected in the past.  But in fact the margin has shifted thus far by 29 votes, and any seat in which even three times such a shift could have made a difference in the past would have been subject to a similarly close rechecking of votes, of the sort not yet finalised here, with many slight changes as the count continues.   Furthermore had any candidate lost by, say, 150 votes, with their scrutineers noticing many borderline formality calls, they could have taken the matter to court after declaration.   It seems that if staff do not do their job thoroughly Palmer will complain, and if they do do their job thoroughly Palmer will also complain.

1 Oct 4:30 pm: Distribution of preference stage is finished and the margin is 7.  Full recount expected to start on Thursday.

1 Oct 11:50 am: Palmer lead is now 7; two votes have switched from O'Brien to Palmer.

5:48 pm: Palmer lead down to 3; more losses to informal.  It may well be that a lot of these losses are valid votes being treated as informal out of caution, but we will need to wait and see on that score.

5:40 pm: And more losses for Palmer; he's down to a lead of 9 (and this time, O'Brien has even gained a vote).

30 Sep 5:20 pm: Back to 13 again.  David Barry has observed on Twitter that the lead changes now consist of vote losses from Palmer's pile, without any gain by O'Brien, so presumably some votes previously ruled formal are being ruled informal.  I don't know whether this is the overwhelming cause of the change in the count, or whether O'Brien himself has gained votes from Palmer and lost an equivalent number to the informal pile.

We are not yet even up to the full formal recount stage but assuming we get there, at that stage decisions on votes that are made at district level and disputed by scrutineers will go to the Electoral Officer for the state, whose call on them will then be final at the declaration stage.  Then the candidates can ask the Court of Disputed Returns to have another go; the McEwen case from 2007 shows that in that count many ballots considered informal by the state AEO were accepted by the courts as valid.

30 Sep 3:30 pm: Palmer leads by 15.

30 Sep: A big hit to Palmer's lead in figures currently showing on the AEC site; he now leads by 13.  The distribution of preferences is expected to be finished tomorrow and this will if necessary - and it looks like it will be necessary - be followed by a full recount.

28 Sep: Clive Palmer's lead is now 42.

26 Sep: Clive Palmer's lead continues to grow, now at 43.

24 Sep: Clive Palmer's lead has increased by three votes so far to 39.

21 Sep: I'll be moving further Fairfax updates to this position in the article from now.

Clive Palmer has provisionally won Fairfax by 36 votes, but firstly preferences must be distributed, and then if the margin is still below 100 (which it almost certainly will be) there will be a recount.  The first part might sound a bit odd - haven't they already done that?  But no; Palmer's primary vote was only 8.24% above the Labor candidate, and while there is no real chance of the Labor candidate catching Palmer on the 13.95% of preferences for minor candidates (especially given that the LNP candidate can also receive these preferences) it has to be checked anyway.  In the process, some errors (inevitable in a count of this nature that a few votes will be wrong) are bound to be caught and fixed.

Then comes the recount (which continues until the DRO is satisfied the total is correct).  Then the result will be declared and the loser may choose to take matters to the Court of Disputed Returns to attempt to get the margin changed or the election overturned.

There are various reports about the margin being historically close but those making them have very short memories for the McEwen contest in 2007 (initially won by Labor by 6, then declared for Fran Bailey by 12, which blew out to 31 as a result of a re-examination of contested ballots following a court case brought by Labor).

Furthermore, I'm indebted to Dr Adam Carr on Pollbludger for much of the following: the closest declared margins in House of Reps seats were 1 vote in Ballarat 1919 and 5 votes in Riverina 1903; both were declared void.  However a 7-vote margin in Werriwa 1914 survived and there was also a final 12-vote margin in Stirling in 1974 (after an 11-vote shift in recounting) and a 14-vote margin to Chris Gallus in Hawker in 1990.

At state level an election for the Victorian upper house seat of Nunawading was famously tied in 1985, with the returning officer under the then-prevailing rules exercising a casting vote (which was done by drawing a name out of a hat.)  That election too was declared void by court and re-run.  There have also been other single-digit margins at state level.

So while Palmer's win is apparently extremely close, at this stage, it is not historically so.



The National 2PP (c. 11 Sep)

Those who have been following the national two-party preferred vote on the AEC website would have noticed that it has been tracking upwards in recent days and is now up to 53.28.  This is common in late counting because late counting - especially postals - tends to favour the Coalition.

What some people may not realise is that that 2PP count is based on 140 of the 150 seats.  It doesn't include the so-called "non-classic seats" (those that are not Labor vs Coalition).  Eventually, 2PP counts in those seats are conducted and added, but it takes a while; it's not exactly the highest priority.

Those seats include: Melbourne (Green vs ALP), Batman (ALP vs Green), Denison (Ind vs Labor), Fairfax and Fisher (both LNP vs PUP), Kennedy (KAP vs LNP), Indi (Ind vs Lib) and Mallee, Durack and O'Connor (all Lib vs Nat).   Of these seats, three are Labor-leaning vs the Coalition, and seven are Coalition-leaning vs Labor.  Based on the swings in the states where these seats are located (and other factors like getting rid of Peter Slipper), it's likely overall that they've leaned Coalition by something like 56.7:43.3.  Their inclusion is likely to increase the national 2PP by around 0.2% and, with post-counting not even started yet in 19 electorates, there are probably more increases to come.  The 2PP should finish over 53.5, perhaps even around 53.7, though I have not attempted to model its end point exactly.

On election night there was widespread rubbishing of Newspoll in particular on the grounds that its 54% 2PP was seen as excessive and the 40-seat victory had not eventuated.  Newspoll was indeed wrong about the size of the Queensland swing and its aggregate polls will probably scrub up badly when I eventually review their performance.  However it also gave a dire projection for Queensland in 2010 and was accurate that time.  The 2PP is likely to be very close to Newspoll's and Nielsen's and the victory margin is probably 35 or 37 seats (with two dropped by the Coalition to third-parties, something that national polls can hardly predict).  On a national basis any error by the traditional pollsters was trivial and Newspoll, Nielsen and Galaxy have clearly all performed well in that field.  Those who danced on the grave of Newspoll's credibility on election night should have waited until it was actually dead.

2PP Update (16 Oct): Thanks to Adrian Beaumont for advising that the AEC does not intend releasing the final distribution of preferences and final national 2PP until the Fairfax and WA Senate recounts are finished. They expect this to occur in a few weeks time.

Some General Comments About Post-Counts and Postals (Thurs 12 Sep)

It is being noted in some seats that the distribution of post-count votes seems markedly different to 2010, with more postals included.  This is bad news for anyone not from the Coalition, which traditionally does very well on postals.  I am trying to get some figures on the proportion of postals that are typically received back by the end of the count (presumably it is nearly all). 

What I was interested in was this: is there any evidence that postals favour the Coalition less at this election, perhaps on account of the greater number of them changing the leanings of postal voters?  On postal votes so far the answer is no - in Victoria at least.  18 seats with some postals in in that state have the Coalition outperforming its ordinary vote by an average of 6.11% compared to 5.89% in 2010.  A statistically insignificant difference, but indicative evidence that close seats with a greater proportion of postals to last time will be troublesome for Labor.

It may even be worth checking some generally "called" Labor seats with 2PPs just above 51 to see if any might be vulnerable based on modelling off the 2010 results.

General Comment About Absent Votes

A point made by William Bowe on Poll Bludger is that absent votes tend to come from areas bordering an electorate and hence can be highly variable as they come in.  For this reason I've revised a number of comments about absent votes below and have just assumed they will replicate the seat swing by the time all are counted, until the number of absent votes in is reasonably large.

General Comment About Recounts

A recount for a House of Reps seat occurs automatically if the final margin is less than 100 votes.  Otherwise the defeated candidate needs to ask for a recount and needs to give convincing reasons why a recount of some or all votes will occur.  If a candidate asks for a recount and gives no reasons the request must be rejected. 

Updates

1 Oct:

New England:  Barnaby Joyce finished first in this seat with an outright majority but it seems that an independent may have finished second, as the AEC has moved the seat into the list of seats where a leading candidate is supposedly not determined. 

28 Sep:

Wills:  This seat is being recounted as ALP vs Green because the Greens have finished second, but Labor will win easily.  That is why a seat has disappeared off Labor's tally on the AEC website.

26 Sep:

Fisher: Labor has finished second in this seat, not PUP as for much of the count expected, but Bill Gissane recently posted on Facebook that Mal Brough has reached an outright majority.

21 Sep:

2:30 pm: AEC computer now showing margin as 36, meaning that the last prepolls must have split to Palmer 77-63, despite prepolls favouring O'Brien 58:42 prior to that; just the latest 1-in-1000 (or so) statistical freak in a count that has been riddled with them.  (It is possible they came from somewhere wildly unrepresentative of course.)  Anyway, assuming that is all (the remaining votes showing are rejected votes) then there will be a recount next week and possibly a court case after that.  In the case of McEwen 2007 the net change in favour of the Coalition from the combined impact of the recount and the court case was 37.  But the chances that a margin of 36 will at least survive the recount if not the court case must be good.

And again 2:10 pm:  A 77-63 split on what I assume to be all the remaining valid prepolls has been reported on Twitter meaning that if there are no more to count Palmer has provisionally won by eight votes, pending next week's recount. A different report is indicating the margin at 36, possibly with the same vote split the other way around.

That one again 12:50 pm: Almost all the postals have been thrown and Palmer did well on them to only lose 23 of his margin with seven postals left to throw.  He leads by 22.  I don't know if the seven postals remaining are valid but if they're not, Palmer now projects to win by one (give or take ten and a recount and a court case ...), and if they are, he projects to lose by one.  Let's split the balance and project it as tied.  ;)

You know where 11:20 am:  The latest update of this thriller sees that it is O'Brien's turn to pull something out of the hat - out of a relatively small number of absents he has something like a 70:30 split to cut the margin down to 45 for Palmer.  The race now projects as so close that there is really not much point posting a projection because whatever it is could well be altered by 30-40 votes in the recount next week and the court case to come!  But postals would be expected to knock the margin down by 35 if they go as they have gone through the count, and with prepolls closing the gap by an average .156 votes/vote, not very many more have to be admitted for Palmer to lose the lead.  If yesterday's information on number of votes remaining is correct (allowing for postals to be added) then O'Brien now projects to win the normal count by 10, although either could still win it.  Then there's the recount, then the court case ...

20 Sep:



Fairfax 5:10 pm: We now have it from AEC that there are 450 to count, plus late postals, which knocks Palmer's projected margin down to 34 assuming the rejections are all EPPVs.

Fairfax 1:20 pm: All remaining absent votes have been thrown and Palmer leads by 111.  However a whopping 232 absents were disallowed.  If the disallowance rate on prepolls is much less then this might still be very close; for instance, my projection has Palmer winning by 18 assuming all votes are thrown. However, it's being reported that there are only 350 to throw, which I assume means a similar number of EPPVs are to be binned as well.  That being the case my projection has Palmer winning by 48.  The count will be completed tomorrow.

19 Sep:

And This Makes O'Brien's Task Much Harder:  Poster Julian T on Pollbludger has observed that there will invitably be some votes rejected in the categories of absent and early-vote prepolls, but that are not showing as rejected yet (they are apparently still listed as awaiting processing).  For instance in Franklin (Tas), where there are zero absents awaiting processing, 101 absents were rejected.  This explains cases in which the number of votes shown as remaining exceeds the number reported as remaining in some reports of the counts (eg Indi). 

Clive Palmer On Provisionals: Off The Scale! I thought it was worth taking a very close look at Clive Palmer's performance on provisional voting today.  Generally these were being treated as an unknowable and hence not included in projections, but provisionals can be modelled like anything else, and had I modelled them I may have projected Palmer's final results as about 30 votes better than I have (this does not include today's projection of a win by 44, now 46).  But there was no past evidence on which to anticipate what actually happened. 

In 2010, nationally Labor tended to outperform its ordinary vote on provisionals by 4 points primary, and the Coalition to underperform by 8.77.  In Queensland the ALP outperformed by just over 5 points and the Coalition underperformed by 10.5.  So, 2PP differences between ordinary and provisional results on the order of 8 points against the Coalition were common.  Furthermore, looking at the crossbench seats, these kinds of differences happened in two of the three crossbench vs Coalition finishes (Lyne and New England, but not Kennedy) but failed to happen to such an extent in any of the four Labor vs crossbench finishes (Labor vs Wilkie and three cases of Labor vs Greens).

There were fifteen seats in the country in 2010 where the gap between ordinary and provisional votes exceeded 12 points 2PP in favour of Labor - Blaxland, Cowper and Eden-Monaro (NSW), Capricornia, Fairfax (!), Hinkler, Longman, Maranoa, Ryan and Wide Bay (Qld), Ballarat and Isaacs (Vic), Braddon and Lyons (Tas) and Wakefield (SA).  The list was headed by Ryan (17.4), Braddon (16.63), Maranoa (16.03), Ballarat (15.99) and Longman (15.56).  (Fairfax scored 13.85.)  An interesting thing here is that some of the Queensland electorates where the Coalition does exceptionally well in post-counting generally are among those where it does the worst on provisional voting. 

Palmer, in this instance, has pulled a provisionals-to-ordinaries gap of 22.59%.  This beat the gap achieved in every seat in the country in 2010 and did so by more than five points!  It is completely off the scale.  Furthermore, while there's evidence that PUP does well on provisionals nationally, it is nowhere near as strong in general as in this seat. There is not much point in trying too hard to explain a spectacular break on a sample of little more than 200; Palmer may have been very lucky.  But it could also be that a combination of underlying trends on provisional voting in Queensland and something relating to this particular contest has led to this remarkable breakdown, that looks like winning Palmer the seat in a circumstance where even a routine advantage on provisionals may not have been enough.

McEwen 6 pm:  I've been ignoring this seat because the margin has been bouncing up and down a little but there's absolutely no evidence the Liberals should win it.  The fairly large number of votes remaining and the smallness of the margin are the only things preventing me from calling it, as a matter of probably unnecessary caution.

Fairfax 4:30 pm:  Remaining provisionals have been processed and mostly rejected and Palmer now leads by 98, with 339 absents, 424 EVPPs and 148 postals still shown as awaiting dispersal. (But some will not be admitted - see above.)

Fairfax: A huge bonus to Palmer as a batch of 220 admitted provisionals breaks enormously (73:27 no less!) in his favour, pushing him to a lead of 96.  Even if none of the remaining provisionals are admitted he is now projected to hold on by 44, but if the remaining 232 are accepted it will be more than that. 

18 Sep:

Fairfax: What If It's A Tie?  It probably won't be, but what if it's a tie on the final count?  Why recounts of course.  And if it's still a tie after every recount?  Then the Electoral Commission is required to file a petition that the seat be declared void (S 274(9C), S357).  The Court of Disputed Returns then has the power to declare a candidate elected or declare the election void, causing a by-election (S367A).  Most likely the court would do the latter unless its own examination of the votes convinced it that a candidate had won.  As we saw in McEwen in 2007, the court re-examines disputed votes and makes its own rulings on them, and can override rulings already made.  I believe that if the court could find no means to split the tie, it would trigger a by-election rather than appoint a winner arbitrarily.

Fairfax 2:15 pm: Trouble for Palmer again as a bunch of absents breaks less strongly for him than previous lots.  He leads by three but the projected margin is now 38 to his opponent.  Still close enough that anything can happen.

Fairfax 12:15 pm: O'Brien is in the lead now by 18 votes. The remaining votes are expected to be slightly favourable to Palmer and if they follow past trends and no more postals arrive then the projection now has O'Brien winning by 2, which is again utterly meaningless in terms of splitting it.  (I am assuming the few provisionals accepted will split 50-50, but they could even decide this!)

17 Sep:

Fairfax 6:30 pm:  A big recovery for Palmer as the most recent batch of postals failed to break as strongly to O'Brien as the previous batches.  He now leads by 64 votes and if no more postals are received but the few remaining prepolls are, then I project the margin as 3 votes to Palmer, which means it is back to lineball territory and again very likely to go to a recount.

Fairfax 1:40 pm: 1500 prepolls added and Palmer's lead has been slashed to 122 votes.  Worse, the proportion of prepolls not going Palmer's way has increased.  My projection now has him losing by 131, outside the automatic recount margin. This does not mean he will necessarily lose, but his opponent's position is the better one now.

Fairfax 12:45 pm: Clive Palmer's lead has increased slightly to 364 votes but my projection now has him at -70 even assuming no more postals (there are probably not that many left to come). 

McEwen:Rob Mitchell's excellent performance on prepolls continues - actually it becomes even more excellent - so despite a big scare on the night I don't see much doubt that he has retained the seat.  This means Labor has saved one of the three that were lineball on the night, and lost all in which the Coalition were leading.

Indi: It's so over people are doing Twitter sweeps on the margin.

16 Sep:

Fairfax midnightish: The AEC are confident the Coolum mystery has been solved - it seems via the simple incorrect entry of online data to the wrong polling place, affecting House of Reps data and not Senate.  The AEC's explanation is a little confusing compared to what seems to have now actually happened, which is the complete transposition of votes from the two PVCCs in online data display.   That being the case, Palmer is now on his own in a mildly difficult position.

Parramatta 10:30 pm: Labor's lead in this seat has blown out to 849 and Julie Owens has clearly retained. 

Fairfax 5:30 pm: More absents in Clive Palmer's favour pushes his lead out to 362, but this is actually bad news for him as his share of them has fallen; the projected margin is now 67 (no more postals) to 114 (200 more postals) in O'Brien's favour.  If the decline in Palmer's position continues he might even miss the 100 vote cutoff for an automatic recount.

McEwen 4:00 pm: Finally some prepoll action in this seat and the first batch has actually bounced in incumbent Rob Mitchell's favour.  He now leads by 199 votes and if the remaining votes carry on this way could win by 500 or so.  I would not be surprised to see his advantage drop on early prepolls drop or perhaps even reverse, but that first favourable bundle has made things much more difficult for his opponent. 

Fairfax 3:40 pm: The Coolum Beach PPVC mystery has finally broken in the mainstream press with Clive Palmer saying he has been informed that the discrepancy in Senate ballot papers is balanced by a discrepancy in Buderim (presumably Buderim PPVC, which has still not published Senate figures). 

Indi 2:45 pm: A further batch of early prepolls has again dropped McGowan's lead, this time to 405, but again Mirabella's share has fallen where it needed to rise.  The hostile votes are all but gone and there should be little change to the margin from here. 

Indi 2 pm: Another update has seen McGowan's lead fall to 449 but the great news for her is that a lot of the early-vote prepolls are gone and Mirabella's share of those has fallen where it needed to rise.  McGowan's projected margin has risen sharply to 304 and there is not very much left to throw.  The slim but realistic chance Mirabella appeared to have of retaining in the past few days has faded away.

McEwen 12:45 pm:  Mitchell's lead is now out to 193.  But we are still to see any early pre-poll votes, the behaviour of which may settle the seat one way or the other.

Fairfax 12:40 pm:  O'Brien continues to close in on Palmer with the gap now down to a flimsy 209 votes.  My projections now have O'Brien the winner by 47 votes assuming no more postals and no resolution friendly to Palmer of the Coolum PPVC issue.  Of course it's well within the margin of error of such a projection that Palmer would win, but there is a high chance that this seat is headed for a recount.

Indi 12:30 pm: Significant respite for McGowan from the Mirabella surge with some absent votes not only breaking in her favour but doing so more heavily than previous absents.  This takes McGowan's lead out to 592.  There are potentially 2498 early vote pre-polls to throw (I'm assuming all arrive though unsure if they will necessarily do so), plus 521 absents, 54 known postals, and whatever postals may still trickle in.  I now project McGowan's margin as 207 assuming 300 remaining postals (and there may not be that many).  Assuming the rate on absents stays constant, Mirabella needs over 62% on the remaining early vote pre-polls compared to her (remarkably high) 58% so far. 

15 Sep:

Fairfax Discrepancy Update 10 pm:  There is a lot of discussion about the Fairfax Coolum Beach PPVC discrepancy on the Poll Bludger thread here.  The issue is both that this PPVC booth has a mismatched number of House and Senate votes (at this stage) and that it has a massive discrepancy between the LNP vote in the House and in the Senate, as compared to other booths.  This could be partly explained if it turns out that the Senate count is not complete at that booth (indeed, some Senate booths in various electorates have still not reported after a week); for instance, there may be Senate votes still to be added to the LNP's tally.  But it is also possible that the issue of the discrepancy in count and the discrepancy in result are unrelated.  Hopefully this PPVC booth will be rechecked thoroughly and matching Senate/House figures (at least in vote totals) available soon. 

McEwen 4:20 pm: Rob Mitchell (ALP) has resumed the lead by 97 votes.  Absent votes, of which there are potentially 2673 remaining, will stretch his lead, but the big issue here is the 5782 early vote prepolls, none of which have been thrown.  On current patterns a c. 53.4% split will give the Liberals the win here, but on 2010 patterns they would only get 51%.  We have no sample of how these are going yet so the seat is very much up in the air.

Indi 4 pm: More postal and early votes have been added.  The early vote picture has got even better for Mirabella (these are now splitting 58:42 in her favour - even better than postals, which is amazing) but the postal picture has got slightly worse.  McGowan now leads by 515, but there are still potentially 2464 early votes to throw, plus a dribble of postals and over 1000 absents. I now project McGowan's final lead as down to 180 assuming 300 remaining postals.  The projected lead keeps coming down, but so does the number of remaining votes and hence the likely error in the estimates.  It's still not over.

14 Sep:

Indi 6:30 pm: Some more absent votes have been added.  They broke more weakly to McGowan in 2PP terms than those already included but a higher proportion were formal. The net effect of these votes on my projection was one vote in McGowan's favour.  If a quarter of the outstanding postals arrive then the current projection is McGowan winning by 240.  If none arrive it is 284 and if all arrive, 109.  The main point requiring caution is that so few Early votes have been sorted.  At present Mirabella has 56.5% on those that have, but that is off a sample size of just 676, which has a margin of error of 3.4% (based on an assumed total of 4546 early votes). All else being equal she will need 59.7% if a quarter of remaining postals arrive (which may be an overestimate).  Her chance is slim, but it exists.  Maybe around 5% on current figures.

(For another example of modelling Indi, check out Michael McCarthy's blog here.   Readers may be interested to know that Michael and I have co-authored a paper or two in the distant past - on the modelling of snail ecology!)

Indi 4:30 pm: A useful issue in modelling the rest of the count in this seat is turnout.  No division in the entire nation exceeded 96% turnout in 2010, though Indi was one of those at the high end at 94.8%.  Assuming the acceptance of all absent, early and prepaid envelopes awaiting processing takes the turnout in Indi this time to 95.2%.  (I am ignoring provisionals since most are rejected, and those accepted will likely be of voters not included in the enrolment figures).  Receipt of all the outstanding envelopes in other categories would push turnout to an improbably high 96.9% (that's my estimate, I've seen some higher) so likely that won't happen and at least half the remaining potential votes won't be arriving. That makes for a current projected margin of probably just over 200, but with a lot of votes still to throw this can change.  Something some people are not noticing in modelling the current figures is the issue of informal votes. They are higher for absent votes than prepolls and postals.

Indi 2 pm: Even with the 1000 votes to McGowan this seat is getting unexpectedly close again as Mirabella is performing extremely well on early prepolls.  But it's probably just not quite enough.  At present McGowan leads by 804 but on current trends from absents, earlies and postals will have that lead cut to a final margin of 108 assuming all postals arrive (it would be 283 if none arrived).  I'm a bit embarrassed about treating this seat as a done deal prematurely (something I try very hard to avoid) after the 1000-vote discovery; it is still not absolutely over.

Lingiari 2 pm: Worth mentioning that although the margin in this seat is low there are very few votes left to throw (see comments) so it shouldn't be closed down. Lingiari is a historically low turnout electorate.


Fairfax 2 pm:  Things have taken a turn for the worse for Palmer here.  With more early votes added his share of them is only 43%.  His lead is just 498 but if the votes awaiting processing continue their current patterns he will lose about 453 of that.  That is a very fragile position especially as that calculation does not include postals yet to be received.  There are potentially 1181 of those, not all of which will arrive, but not that many have to given that he is losing 1 vote of margin for every 4 postals. Unless the Coolum PPVC mystery is resolved in Palmer's favour this is going to the wire and probably beyond that to the courtrooms.

13 Sep:


McEwen 6:50 pm:  Absent votes that were about as favourable to Rob Mitchell as last time (accounting for swing) have slashed his opponent's lead.  For now he is on course to retain despite being behind but we will need to see more votes to be confident.

Fairfax 6:40 pm:  The postal plunge continues with Palmer's lead now down to 718.  With a lot of the postal pain gone Palmer is still tracking to win by 400 or so even if most remaining postals arrive, and that's without the Coolum irregularity rebounding in his favour, so his position is still strong.

Eden-Monaro 12:30:  Mike Kelly concedes the famous litmus seat is gone.

Parramatta midday:  Labor's position on early prepolls has improved and these are now breaking almost as well for them as predicted based on the last election.  They should now hold the seat.  

Capricornia 11 am: Labor has thrown in the towel.  Conceding defeat has no actual impact on the electoral process; it's just a nice thing to do; there are cases of candidates conceding only to find out they have later won.  But if it's good enough for them, it's good enough for me (and there is no reason to believe Labor can recover it).  Capricornia goes to the LNP.

12 Sep:


Barton 6:40 pm: The postal rout continues with Varvaris now 961 ahead.  Labor did gain more than that on earlies and absents last time but that was off a 2PP of 57:43.  I don't know if the fat lady is singing but Freddie Mercury certainly is.  Another one bites the dust.

Fairfax 6:40 pm: Early prepolls for Palmer are a little less bad than they might have been and he may have a small edge irrespective of the Coolum PPVC issue (which is yet to break)

Petrie 4:50 pm: Another seat where the margin has gone over 1000 votes thanks to postals and another seat I no longer doubt that the Coalition will win.

McEwen 4:45 pm: More postals added here and the margin is out to 389.  But the postals are nearly gone and Mitchell performed very well on absent votes, of which there are a lot in this division, in 2010.  If he can repeat that performance (accounting for the swing) he can still pull this back.

Capricornia 4:40 pm: Rinse and repeat, more nasty postals for Labor, margin already 630.

Barton 4:40 pm: Lead to Varvaris out to 781 and to the delight of my election model and the sorrow of the seat betting markets, it looks like Labor are toast in this one.  I'll keep it in the Likely pile for now.

Dobell 4:30 pm Lead to Coalition over 1000 so game over here as well, unless there are any missing votes anywhere!

Parramatta 4:20 pm The first batch of Early votes is in and they are roughly a 50:50 split, enough to knock another 100 votes off Owens' position in my projection.  The first group of Absent votes are also in and they have broken badly for Labor too but I don't know where they came from hence don't know if there is a problem here.  The raw margin is 389 and it will probably come down further from there. 

Fairfax 3 pm: Remember what happened in Indi yesterday?  It could be rinse and repeat as a suspected discrepancy has been found at the Coolum Beach PVCC - see Poll Bludger for the details.  If verified, the discrepancy is very likely to result in Palmer winning the seat.

Fairfax 2:45 pm The long-awaited first batch of postals has arrived and Palmer is being thwacked 63:37 on these by his LNP opponent, cutting his lead to 1050.  Having seen PUP's Tasmanian Senate performance, where their postal primary is about half their ordinaries performance, this sort of thing shouldn't be too surprising in a seat where the LNP performed well in post-counting last time.  There are enough remaining postals for this sort of split, should it continue, to wipe out Palmer's lead and send the seat to the murky waters of absents versus early votes.  Early votes won't be much fun for Clive; the seat remains unclear.

Parramatta 2:30 pm Early postals have shown Julie Owens not repeating her very strong performance on postals in 2010 and her lead has been slashed to 480.  Luckily for her there are not that many postals left.  She also gained on both absent and prepoll votes in 2010 so assuming a similar performance this time will still win.  But I would want to see some of the remaining prepolls first as my projected margin is down to around 400 and I am not convinced the 2010 performance will be repeated.  I have downgraded the seat from likely Labor win to unclear.

Capricornia 11 am: Post-counting finally got underway in this seat with a bunch of postal votes breaking 59:41 to the LNP.  This is entirely consistent with the pattern from 2010 in this seat in which postals broke to the Coalition by 11 points more than ordinary votes.  It's "close" at the moment (18 votes) but at this rate it will not stay that way. 

11 Sep:

Note re Indi 8 pm: Currently the AEC site shows Mirabella trailing by 501 votes.  This does not, however, include the 1003 missing votes.

Barton 6:30 pm More postals have gone in and the story gets even worse for Labor as the split in postals in this lot was even more severe than the first bunch.  The gap is now 643 votes and there are still around 2000 postals to be thrown. 

Barton 5:40 pm A horrible result on the first lot of postals for Labor puts them 309 votes behind.  Labor's slight advantage on early/absent votes combined would struggle to erase that if there were no remaining postals; there are 4000 of them.  I have upgraded this seat to a likely Coalition win.

Eden-Monaro 5 pm Early postal votes have favoured the Coalition more heavily than projected with the lead now out to 736 and extremely difficult to close.

Petrie 4:30 pm A small bunch of absent votes have broken slightly in Yvette D'Ath's favour but actually by less than projected.

Reid 4:20 pm The first batch of postals here was massively in the Coalition's favour pushing their candidate out to a lead of 928.  Time to stick a fork in this one I believe; I've assigned it to the Coalition's pile.

McEwen 4:20 pm  Late-breaking ordinary votes have improved Rob Mitchell's position on ordinary votes from 50.1 to 50.19, cutting the deficit to 105 votes.  It's still a difficult position but close enough that I have moved it back to the unclear category.

Eden-Monaro 4 pm Labor's Mike Kelly has recovered some ground with an 85-vote gain on early votes to trail by 506.  Kelly did not gain on early votes in 2010 so those votes are a bonus. 

Fairfax 4 pm  A big bonus for Palmer as he records a 59:41 result on a batch of absent votes, about five points better than projected.  However I don't yet know whether they came from an unrepresentative area.

Indi 3:30 pm The story checks out as there clearly are 1000-odd missing votes in the Reps count for that booth and there is a gross discrepancy between McGowan's vote at that booth and the main Wangaratta booth that is not replicated in other booths with matching PPVCs.

The mountain McGowan had to climb on preferences, apparently needing 77.5% of all preferences to go her way, was a very steep one and projections last night were hinting she might not get there.  But it turned out this is an illusion and the mountain was a mere 74.6%, which makes all the difference.

Note that it has been reported that there are three uncounted booths, but two are interstate PPVCs and another is a BLV PPVC (blind and low vision.)  There are few if any ordinary votes to be added.

What an extraordinary way to win!

Indi 3 pm The Australian reports a sensational development: the story is that 1003 overlooked votes for McGowan have been found at the Wangaratta pre-poll centre.  If this is true as stated, Mirabella is totally busted from a position in which she looked like she might win (unless she finds a 1000 vote pile of her own!) I am awaiting official confirmation of details in case of any error in reporting.

2 pm Changes in seats processed so far today don't look significant. 

10 Sep:

Barton 10 pm: Barton is a seat of some interest to me as it was the only seat predicted as a win by my model but not by the seat betting markets.  Anyway, the Coalition has taken the lead from an odd source: a small parcel of absent votes split 55:45 to the Coalition.  Last time absent votes had about the same split as the ordinaries.  While this difference only worsens Labor's position by 56 votes, they are projected to lose by a few hundred if remaining postals and prepolls behave as they did (relative to party votes) in 2010.

Indi 6 pm:  The first batch of 2000 postals has knocked 288 votes out of McGowan's lead and she now leads by 764.  Even the c. 7000 postals awaiting processing will be sufficient to more than erase this lead at this rate but McGowan could pull back most of the difference on early and absent votes combined.  The remaining trickle of postals not yet received could also assist Mirabella.  McGowan needs to perform better on future tranches of postals or Mirabella could just retain.

McEwen 5:00 pm: More postals have seen Mitchell fall 322 behind, but worse news for him is that his share of postals is worsening significantly. Unless this changes it will be difficult for him to erase the already existing Liberal lead.

Fairfax 4:30 pm: Palmer's ordinary vote in Fairfax is 50.88% 2PP, or a lead of 1105. He has the very favourable Nambour PVCC to come which may add close to 300 votes to his lead pushing him out to 51.1%.   He has lost 52 votes of that in out-of-electorate prepolls already, and these favoured his opponent by about one and a half points more than 2013 projections.  Depending on the rate of return and formality of postal votes, my modelling off the 2010 results shows the seat as still too close to call with a possibility that the LNP will pull back the margin.  Flippant prediction: Palmer to lose by 20 votes and the AEC to spend the next 30 years in court.  Serious assessment: For the time being, tossup.

Fairfax:  No new figures today but Palmer has been projected to finish ordinaries at 50.9% (poster Dodgy Dave at Pollbludger).  In this seat last time the Coalition pulled back 0.60% on the post-count, one of their better efforts.  Throw in also the increased proportion of postals in the post-count and the unusual nature of the contest against a non-traditional candidate (albeit a very high profile and well-resourced one) and it is very difficult to say what size of lead after ordinaries would make for a guaranteed win, save that his current lead is nowhere near enough to call the seat as a certain gain, and in my view it is still unclear until we see some postals.

McEwen: (second update) A second batch of postals was less friendly for Rob Mitchell moving him 90 votes behind.  More importantly, his share of the postal vote has dropped from just below 50% to around 48%.  Although this is still positive compared to 2010, he has the same problem as McGowan in Indi - a very high proportion of the remaining votes are postals.  On the assumption that every postal is received I project this as very close at present; even if it is not, the current postal situation may worsen for Mitchell as the first batch might have not been representative.

Indi: (revised) Cathy McGowan leads Mirabella by 1015 votes at the end of assignment of ordinary votes (50.72% 2PP).  I originally thought this was likely to be enough given Mirabella's modest performance in previous post-counts.  However Poll Bludger member canberra boy points out that postal votes feature much more heavily in the remaining portion of non-ordinary votes than in 2010 and this greatly increases Mirabella's chances of drawing back the lead.  The seat is hence still very unclear and we will need to see how the postal vote is going later today.

McEwen: Almost 2000 postals have been added in this seat cutting incumbent Rob Mitchell's lead from 72 votes to 66.  Although Mitchell went backwards by six votes and although his lead is tiny, I take this as an excellent start to the post-count for him.  In 2010 in this seat the Coalition performed 8.22% (2PP) better on postals than on ordinary votes, but in the postals thrown so far this year the Coalition has performed a mere 0.2% better.  If these postals added so far are anywhere near representative then Rob Mitchell will hold the seat; even if they are not, they have significantly boosted his chances.

Fisher: Mal Brough will win this seat (CALLED).  In the process of the re-alignment of votes to the probably correct two candidates, Brough currently has 55.4% 2PP with about half the booths included.  The booths included do slightly favour Brough so far (he polled 45.8% primary in them compared with his overall 43.9%) but PUP would need to keep Brough below 45% 2PP in the remaining booths to equalise the ordinary vote count.

------------------------------------------------------------------------------------------------------------------
9 Sep:

Indi and Fairfax Counts:

Indi and Fairfax counts are in the process of being re-aligned to correct two-candidate preferred.  Note that as they are realigned, depending on what booths are included at each stage, votes will go up and down and candidates may seem to be winning by lots when they are actually not.  If good booths for a candidate are down the order in which booths are included, that candidate will go up.   Raw figures currently showing McGowan on 51.8% and Palmer on even more are deceptive; both seats are closer.  Several people wrongly called Denison for Labor in 2010 because they did not understand how this system works.

The process is blisteringly fast so far and because of a work commitment today I have not had the time to follow it I would have liked.    However as late booths are added Cathy McGowan is likely to finish the re-alignment process on about 50.3%-50.5% of two-party preferred according to modellers posting on the Poll Bludger thread

(UPDATE 10 pm New comments on the thread indicate McGowan's position as slightly better, and perhaps pointing closer to 50.7% from ordinary votes.)  

McGowan's currently shown lead is overstated because of bad prepoll voting centre (PPVC) booths for her that are not yet in the count.

It will then be a question of whether or not Mirabella can catch that lead in post-counting to come.  That post-counting favours the Coalition in Coalition vs Labor contests (see below) but there is not much data for Coalition vs Independent.

A similar situation exists in Fairfax where Clive Palmer has what looks like a winning lead but will  lose most of his lead when the PPVCs are added to the re-alignment.  He may finish the re-alignment  with about 50.7% 2PP.  That sort of lead is tough to overturn but not something I'd call just at that stage.

My thanks to PB posters sykesie and DB, and to David Barry, for their modelling data.

8 Sep:

Non-Classic Seats

Of the two non-classic divisions, if only the current votes were thrown, Cathy McGowan (Indi) would need 75.96% of the available preferences to go to her over Sophie Mirabella.  About half the available preferences are Labor, and the rest are of a gaggle of candidates who will not be seeing their deposits again, including the Greens, who were bombed down to 3.16% by McGowan's run.  Rise Up Australia and Family First have a few percent between them, although RUA's tally includes donkey votes, which will flow to McGowan.

Clive Palmer (Fairfax) currently needs 71.26% of available preferences to go to him against the LNP.  57% of those preferences are Labor and 26% Green, and both those parties have preferenced Palmer on their cards.

I suspect the asking rate for both challengers will become a little steeper in late counting.  By way of comparison Andrew Wilkie (Denison, 2010) eventually needed 66.9% and got 69.7%, albeit against a Labor candidate.

There is a third likely non-classic division, Fisher, where PUP can get ahead of Labor on Green and other preferences and then try to knock off Mal Brough on Labor's vote.  But the asking rate there is extremely steep at 85.9% because of the low PUP primary vote, so I don't think it's a realistic chance.

Of the classic divisions, at this stage the Coalition has the better of the close seat hands with Reid, Petrie and Dobell all having gaps of at least 600 votes while Eden-Monaro is at 245.

Labor has a solid lead in Parramatta but the other three it leads in are on the wire with leads below 150. 

Here are some figures about late counting from 2010.  The addition of late-counting votes (absent, provisional and postal, not counting absent votes counted on the night) improved the Coalition's standing by an average of 0.28% compared to the two-party figure for all "ordinary votes" (which include the prepoll-centre prepolls that are counted on the night).  So, for instance, if the Coalition had 60% of the two-party vote after all ordinary votes were in, then on average it would have improved to 60.28% at the end of the count.  (I haven't tested if these correlate with 2PP yet but might do so later). 

Labor did outperform the Coalition on non-ordinary votes compared to ordinary votes in 33 of the 142 classic divisions in 2010.  The distribution of the differences (below) makes for one of the oddest graphs I have run on here:

The least we can say of this is that it is not a normal distribution!  The Coalition side has a very long tail caused by huge post-counting differences in the seats of Flynn (Qld), Maranoa (Qld) and Melbourne Ports (Vic).  But Labor did not improve by more than half a point in any electorate; its best being Lindsay (NSW, 0.49 points).

Looking at the seats the Coalition leads in now, Dobell (currently 50.61 Coalition) requires an improvement for Labor exceeding any they recorded in 2010.   (The amazing thing to some may be that Labor is even this close, but I've always felt the damage of Craig Thomson to the new Labor candidate's vote was being overrated.)

Petrie and Reid require improvements that Labor exceeded in just one seat in 2010.  Labor did notch a slight improvement (0.04) in Petrie in 2010 and went only slightly backwards in Reid (-0.04).

Eden-Monaro looks much more gettable but here too the news is bad for Labor: three Coalition-friendly booths have not been included in the 2PP.   When they are included I expect Peter Hendy to gain about 300 votes pushing his 2PP out to about 50.36%.  Labor gained enough to overturn such a gap in just five seats in the country in 2010.

I should caution here that given the larger volume of pre-poll voting in 2013 it is to be expected that there will be slightly larger changes from ordinary voting to the final tally in given seats.  There is also the possibility that the voters voting early now are less conservatively inclined than in 2010.  I wouldn't read too much into either of these issues though  -Labor's position in all these seats is very difficult. Possibly the dynamic of this election for post-counting generally will be different but I rather suspect not much.

Of the four shaky Labor-ahead seats Parramatta (50.75) is the only one with much of a cushion.  The Coalition made gains of that size or larger in only 11 seats in 2010, none of them near Western Sydney.  In fact, Labor made post-counting gains here in 2010.

Capricornia (50.1) is Labor's third-smallest lead.  The Coalition gained by more than 0.1 in 70% of post-counts in 2010 including a massive gain of 0.91 in this seat itself (their seventh-best post-count). McEwen (50.05) and Barton (50.05) are on margins that would have been erased by the Coalition in 73% of 2010 post-counts.  McEwen saw an average Coalition gain of 0.28 points in 2010 and Barton a small one of 0.10.   The signs are not great for Labor in these three seats collectively but I would still give them a fair chance of saving one of them.

All up, possible outcomes for Labor range from 53-61 seats but most likely 54-57.  The Coalition's possible outcomes range from 84-94 seats but most likely 88-93. 

I will update this thread with more post-counting over coming days. 




29 comments:

  1. Thinking this through - to get a significant post count gain/loss, you need both a large number of votes in post count, and a decidedly different skew to the votes.

    Are these predictable or random?

    ReplyDelete
    Replies
    1. I think there would be things that are predictable here, eg some large rural seats will have high rates of postal voting, coming from conservative communities. Haven't looked closely but one thing that comes up is that if the three monster outliers are kicked out, there is a weak and shallow inverse relationship between Liberal 2PP and the Liberal performance on post-count, on average. Only 5% of variation explained there though. Not visually a complete splatter-plot but very weak. Probably proxy for a bunch of other causes.

      Lots of urban Sydney deep blue high in the list of comparatively good post-polls for Labor last time, also western Sydney marginals pretty high up.

      If I get time at some stage I'll have a look at factors that predict these things. It's useful stuff to know about.

      Delete
  2. Is there likely to be a relationship between a sitting member? i.e. what if the x-axis was instead swing to incumbent? A sitting member will usually have more time, resources, infrastructure to conduct a rigorous postal vote campaign.

    Thanks for your summary. Don't have much time to read the papers, but this is a quantitatively based independent analysis - well done.

    ReplyDelete
    Replies
    1. Ta, I'll have a look at that as it's an obvious one, and post some results a bit later.

      Delete
  3. Had a look at this and got the following (and then realised why):

    For a recontesting Coalition incumbent, as the swing to incumbent goes up, the post-count swing to Coalition compared to ordinary goes down.

    For a recontesting ALP incumbent, as the swing to incumbent goes up, the post-count swing to Coalition compared to ordinary goes up.

    Not strong relationships, 9% and 8% explained respectively after removal of three outliers. And the relationship is so mild it takes an 11% swing to a Coalition incumbent, or a 17% swing against a Labor incumbent, before there's a projected zero net post-count swing (ordinary votes to final) to Coalition.

    Thought "that's odd", then realised it probably isn't. For instance, if there's a big swing in either direction in an electorate, then it may have built up over the campaign, and hence the post-count (containing early voting) will be milder for the Coalition in the case of a swing to Coalition, or better for the Coalition in the case of a swing to Labor.

    Maybe here a whiff of a hint that Labor may perform better in terms of post-count swing in McEwen than if there'd been nothing going on in that seat.

    As for sitting member vs no sitting member, very little to see there (maybe makes a little bit of difference but not statistically significant for one election.) Ditto for ALP sitting member vs L-NP sitting member - difference in averages was that the post-count swing to Coalition was 0.06 points lower if there was a Labor incumbent, but it wasn't statistically significant. Even if it had been stat. sig. it would have said nothing because it could be that the kind of seats held by each party have a bearing on it.

    Seems that if being a sitting member helps you do well on postals it also helps you do well on the day and that the idea about postals favouring sitting members isn't really any use here.

    ReplyDelete
    Replies
    1. Okay, thanks. Sounds like "Myth busted"

      Delete
  4. Melbourne Ports has a large orthodox Jewish population (who tend conservative and don't vote on Saturdays) which may explain its behaviour.

    ReplyDelete
    Replies
    1. Thanks Blair. Yes, I also suspected the large Jewish vote in Melbourne Ports had something to do with that.

      Delete
  5. I assume the 2PP distributions on the ABC are rubbish in seats like Fairfax because there is no historical evidence for preference flows involving PUP?

    ReplyDelete
    Replies
    1. They're estimates only and I don't believe they're informed by scrutineering data. The ABC's estimates overstated the flow to Wilkie in a similar situation in 2010 (by a fair amount, I recall) and I suspect the same is true for Palmer in this case. But not necessarily enough to cost him the seat.

      Delete
  6. Probably not something of much interest to you Kevin, but is there any way to get a feel for voter turnout numbers at this stage, or will we only know about that when all votes are in?

    ReplyDelete
    Replies
    1. I haven't looked at this yet. I do know that voting before the day was up, presumably causing a decline in voting on the day, but I don't know where the overall picture is likely to be once both are accounted for.

      Delete
  7. I assume your latest update about postals throws extra doubt onto the Fairfax race as well? I've seen this called for Palmer in a few places already. At the least there doesn't seem to be such an acute focus on that seat around the traps anymore.

    ReplyDelete
    Replies
    1. Your assumption is correct. Additionally, Fairfax saw a strong performance on post-counting by the Coalition last time, where they gained by 0.60% 2PP. If they gained by that last time, and they are up against a candidate who surged on late campaigning this time, then given the increase in postals it's quite possible they'd pull back from over 51% 2PP. Palmer is currently projected to have 50.9% at the end of ordinaries. He could be projected to 51.5 and I would not be willing to call it for him as a certain win at that point. Not without seeing some postals.

      That said, we do not know whether the Coalition would outperform Palmer the way they outperform Labor on postals.

      Delete
    2. interesting that's it's viewed as a no contest elsewhere then. I guess people just want to believe...

      Delete
    3. Denison in 2010 was prematurely reported as a win for Labor in numerous media sources during a similar process.

      There has been an update in Fairfax, I will post new comments soon.

      Delete
  8. I'd guess (and it really is a guess) that Palmer would struggle with voters who were not so exposed to what I assume was a blitz of PUP advertising in Fairfax. Moreso because it seems as though the surge of support for Palmer only manifested itself in the last couple of weeks of the campaign.

    Could be very interesting...

    ReplyDelete
    Replies
    1. I will note Palmer probably did have a semi-decent base vote before the campaign there. Scott Steel said PUP were polling low double figures in that area even before any campaigning, and that comment was months ago.

      Delete
  9. One interesting thing to note - Palmer BTL propensity is similar to the Greens, not to Labor or Liberal.

    ReplyDelete
  10. Replies
    1. Unless Sophie has 1000 votes of her own somewhere.

      There are the ordinary votes, the early votes, the provisional votes, the postal votes and the absent votes. And then there are the *missing* votes. Bit hard to compete with a 100-0 split on those ones in a tight contest. ;)

      Delete
    2. Is that consistent with the missing vote count in Indi in previous elections?

      Delete
    3. :) Apparently it's a swing of 100%!

      The votes were not actually "missing" as such, but rather mislabelled. A box with slightly over 2000 McGowan votes in it had been marked as containing slightly over 1000. I can well imagine how this would happen as I have sometimes seen such counting errors myself - for instance when wrapped bundles of 1000 are wrongly assumed to be bundles of 500.

      What is amazing is that a discrepancy as large as 1000 votes got as far as the results being published on the AEC website without cross-checking to determine that the PPVC booth was 1000 votes short of target. Can't recall that ever happening before. AEC will need to conduct a review.

      Delete
  11. Hi Kevin,

    Is there any chance of Lingiari changing? The 'turnout' is at 74% and a small margin that seems to be narrowing.

    ReplyDelete
    Replies
    1. That's a good question to ask as I suspect a lot of people are wondering about this. The key with Lingiari (and the reason I've not paid it attention despite the small number of votes) is that there are very few votes remaining because a lot of voters haven't voted. Following the Declaration Vote Scrutiny for the electorate (http://vtr.aec.gov.au/HouseDivisionDecScrutinyProgress-17496-306.htm) it seems that there are at most 54 absent, 414 provisional, 802 early and 413 postals to go. That's even assuming most of the remaining postals ever arrive which some will not.

      The CLP would need a 71:29 split of all remaining votes assuming all were formal (which they won't be), all the postals arrived (which they won't) and all the remaining provisionals are accepted (which I assume also won't happen). They've been getting close to that split on postals but nothing near on other vote types (absents have been favouring Labor). So despite the closeness of the margin at present, there's just not enough to throw.

      Lingiari is historically a low turnout electorate. In 2010 its final turnout was 75.87%.

      Delete
  12. On those counts, assuming no significant counting errors, Palmer has to be close to home and hosed, no?

    ReplyDelete
    Replies
    1. If the remainder behave about as their vote types have so far (accounting only for random variation) then Palmer's chances are very high indeed. That's even though O'Brien needs 55.4% of the remaining 911 and is projected to get 52.9%, which seems pretty close.

      I am a bit more cautious about it than the implied probability because there are still lots of absents in the mix and we don't know where they're from, though I suspect they'd be a dribble from various places and hence break representatively.

      Delete
  13. Those booths with surprisingly low authenticity challenge rates have surprisingly high formality challenge rates. My assumption is that the PUP scrutineers are challenging on formality where it's sensible to do so, and challenging on authenticity otherwise.

    This is perhaps more modelling work than these challenges are worth, but here goes. Assume that the formality challenges are in proportion to the TCP vote in each booth. So, e.g., Bli Bli has the LNP on 42.3%, so guess that the PUP scrutineers are responsible for 42.3% * 302 of the formality challenges in Bli Bli.

    Add that estimate of the PUP formality challenges to the number of authenticity challenges, convert to percentage of ballot papers, plot "estimated total PUP challenges" against LNP TCP %. You get an R^2 of just over 80%, a noticeably better fit than when you just plot the authenticity challenge percentages.

    ReplyDelete
  14. That makes sense. The figures don't break down votes that are challenged both for formality and authenticity but I'd assume there would be some.

    Early on in the count I did notice there were a couple of booths where the total challenge rate was below the LNP 2PP, meaning that some LNP-preferencing ballots were going unchallenged (whether by accident or design) but I haven't had time to check how many more such were added since.

    ReplyDelete