Saturday, August 28
The AEC has published its provisional distribution of preferences which makes it very clear that Andrew Wilkie will surpass the Liberal candidate in very fine style, recording 20,430 votes to the Liberals’ 15,695 after distribution of Greens and Socialist Alliance preferences, and then comfortably winning the seat on Liberal preferences.
Friday, August 27
The Australian Electoral Commission announces it will conduct a “provisional” distribution of preferences in Denison to ascertain whether the Liberals are likely to be excluded from the count before Andrew Wilkie, a necessary precondition for the latter winning the seat.
Tuesday, August 24
6pm. Indicative preference count finished for real now, with pre-polls and hospital booths added, and Wilkie’s lead has risen to 1.2 per cent (1375 votes).
3pm. The indicative preference count for ordinary votes has been completed, and it puts Andrew Wilkie 1091 votes (1.0 per cent) clear of Labor. That’s a big hurdle for Labor to clear on absents and postals, but there are too many imponderables to say it can’t happen.
Monday, August 23
11pm. “Only one seat now in doubt as Wilkie loses bid for Denison”, reports the Sydney Morning Herald, and it’s probably not alone. This misapprehension is based on the ABC computer’s projection of the Labor-versus-Wilkie indicative preference count, which assumes the 20 booths that haven’t been counted will follow the preference pattern of the 26 that have. There is a three-sided problem here: Labor’s share of the preferences is not as high in areas where they are weak generally; the booths are being counted in alphabetical order; and the strongest Liberal booths begin with an S. Antony Green’s modelling to account for this turns the projected 0.6 per cent Labor lead into a 1.1 per cent deficit (subject to a margin of error), a view shared by PB commenters who know their way around a linear regression. However, Labor is likely to at least close that a little on postal votes.
6pm. Labor might appear to have the advantage superficially at present, but Sykesie in comments has produced a model accounting for the association between Labor’s primary vote in booths that have reported and the share of their preferences in them. The upshot is that as counts are added for booths less preferable to Labor are added, their share of the preferences will come down, Sykesie projects them to finish on 48.4 per cent with an error margin of only 1.3 per cent. However, that doesn’t factor in the likelihood that Labor’s position will improve as postal votes come in. That still makes it too close to call, but Wilkie would probably be favoured.
2.30pm. The Electoral Commission is conducting a thrilling indicative preference count between Wilkie and Labor to ascertain what will happen if they are indeed the final candidates. Wilkie currently looks to be just slightly under the share of preferences he needs, but it’s been back and forth as booths have been progressively added in alphabetical order.
Sunday, August 22
Accomplished Tasmanian psephologist Kevin Bonham, who closely observed the behaviour of Greens preferences in relation to Wilkie when the latter ran at the March state elections, disagrees with Possum’s assessment that Greens preferences will not necessarily put Wilkie ahead of the Liberals, and thinks a greater threat to Wilkie would be that he might be overtaken by the Greens, who will have run a better resourced postal vote campaign. If he’s right, the surprises in Denison might not be over. It is mostly being taken for granted that Liberal preferences will allow Wilkie to ride home over Labor if he finishes ahead of them, but a WA Labor source advises caution on this count based on the precedent of Kwinana at the September 2008 state election. It was widely thought after election night the seat had been won by independent Carol Adams, but victory slipped away from her due to the surprisingly high number of Liberal voters who had Labor second.
Saturday, August 21
This post will be used to follow the late count in Denison, where independent Andrew Wilkie superficially looks well placed to win a Labor seat vacated by Duncan Kerr and contested for them by Jonathan Jackson. At issue is the distribution of preferences from the fourth-placed Greens, who polled 19.01 per cent. Wilkie needs them to close what at present is only a 0.1 per cent deficit over the Liberals, but Possum at least believes the fact preferences are splitting three ways between Labor as well as Liberal and Wilkie will land him short, especially after factoring in a likely weakening of his position as postal votes come in. However, the ABC reports “Labor scrutineers are predicting a desperately close result as preferences are distributed”.
Doing a multiple linear regression of the primary votes and trying to use them to predict the 2PP outcome for labor on a booth-by-booth basis gives some interesting results:
a) using labor and green primaries gives an error of about 1.36%
b) using labor and wilkie primaries gives an error of about 2.77%
c) using labor and liberal primaries gives an error of about 4.58%
(after 11 booths)
If I apply the best model (ie green/labor primaries) to the whole electorate primary numbers, I end up with a labor 2PP of 48.57% ….
Am I correct in assuming that pre-poll/absentee/postal votes will probably favour the ALP both in primaries and preferences?
And that these have not been counted yet?
[This is all very interesting but we seem to be forgettoing how the count actually works. Preferences can’t be distributed until all votes are in. Once that happens counting and elimination occurs at the electorate level, not the booth level (it is not the same thing).
The lowest candidate is eliminated, and then the next lowest, at the electorate level. It is clear that Barnes and Couser will be eliminated, on the current figures, but it is far from clear who will be eliminated next – especially if Couser’s votes flow strongly to Jackson. There may not be enough votes left for Wilkie to get ahead of Simpkins. In which case, game over.]
Yes, there are three questions, as I’ve been saying all along:
1. Does Couser catch Wilkie on postals and SA preferences?
2. Does Wilkie catch Simpkins on Couser preferences?
3. Does Wilkie catch Jackson after everyone’s preferences?
Today’s exercise answers only (3). If the answer to (3) is no, Wilkie loses. If the answer to (3) is yes, then we go to (1) and (2).
[11 booths in
average pref flow 66% to Wilkie. Not enough.]
Bit short at the moment but the booths are not representative. Will go up in my view, but will it go up enough to not only give Wilkie a lead but give him a lead that doesn’t die on postals?
[The lowest candidate is eliminated, and then the next lowest, at the electorate level. It is clear that Barnes and Couser will be eliminated, on the current figures, but it is far from clear who will be eliminated next – especially if Couser’s votes flow strongly to Jackson. ]
I am actually not yet totally certain Couser will be eliminated before Wilkie. He gets the SA prefs and may also improve his position on postals. Looks like too big a gap to quite bridge but we’ll see.
We know from the state election that Greens voters like Wilkie more than they like Liberals, by a lot. In the very Green booth of Cascades, where the Greens outpolled the Libs 2.5:1, it’s clear that the Green voters, at least in that booth, are preferencing Wilkie more heavily than Jackson on a 2CP basis (otherwise it doesn’t add up even if you assume nearly every Lib voter preferences Wilkie). I think it’s completely safe to assume that the Green voters like Wilkie more than they like either Labor or Liberal individually on a head-to-head basis. The tricky bit is that we don’t know how that translates as a three-candidate split.
Just an example of the very simple model mentioned in post 50 (calibrated for booths 1-11) and predicting for creek road and cross street (which weren’t in the training set for the model):
Creek road: Predicted labor 2PP (45.88%); Actual labor 2PP (47.15%)
Cross street: Predicted labor 2PP (46.44%); Actual labor 2PP (46.72%)
[Am I correct in assuming that pre-poll/absentee/postal votes will probably favour the ALP both in primaries and preferences?
And that these have not been counted yet?]
Prepolls cast in the electorate were counted on the night and are included in the current primary. Out-of-electorate stuff and postals are not in yet.
As for whether the postals and out-of-electorate votes will favour Labor, I expect them to do so as independents normally suffer in this situation. Wilkie would want to lead substantially after this 2CP exercise is done. But we don’t know how much of a difference they’ll make, because as the figures posted by Laocoon in #4 indicate, Labor doesn’t seem to have made much of an effort getting party-postals in this electorate.
The flaw in my simple model is of course that it doesnt take into account the order in which candidates are eliminated from the preference count 🙂
[Dr Good,
That’s correct but where do Wilkie’s preferences flow (as well as the reallocation of the others) ? If he goes out before Simpkins, will he elect Simpkins ?]
No, we already know this cannot happen because the original 2PP showed Jackson thrashing Simpkins if they were the last two in. Simpkins absolutely cannot win.
Kevin Bonham @ 56
ta
[Doing a multiple linear regression of the primary votes and trying to use them to predict the 2PP outcome for labor on a booth-by-booth basis gives some interesting results:
a) using labor and green primaries gives an error of about 1.36%
b) using labor and wilkie primaries gives an error of about 2.77%
c) using labor and liberal primaries gives an error of about 4.58%
(after 11 booths)
If I apply the best model (ie green/labor primaries) to the whole electorate primary numbers, I end up with a labor 2PP of 48.57% ….]
Fascinating! Will be interesting to see how this pans out with more data.
I note that in Creek Rd and Cross St, Jackson’s primary leads were only 9 and 7.7 but the differences between the Jackson primary and the Jackson 2CP were around Jackson’s overall primary lead. I view that as a bad sign for Wilkie.
[I know the media need something to feast on during this uncertain time, but it’s interesting to see them declare “Wilkie’s lost” and “it’s 73-73? as if it was fact.]
It’s possible they have access to scrutineering projections (which I don’t) but I do feel sorry for the poor dears trying to get their heads around this one.
On media blurtings: they are all trying to trump each other with scoops on things they don’t understand and/or are totally unscrupulous about.
Model applied to Dynnyrne:
Predicted ALP 2PP: 38.03%, Actual: 38.25%.
Applying to Elizabeth street:
Predicted ALP 2PP: 43.89%, Actual: TBA …
Keep the good work up KB/sykesie.
Gee if you wanna know whats going on politically in this country whether its philosophy, policy, regional or statistical, blogs like this are the place to go.
Forget the mainstream media.
There have been some revisions to outright primary totals. I now have the target figure for Wilkie at 66.88% of all preferences and the dodgy spreasheet of doom has him back in a wafer-thin lead now – which may blow out as the booth mix counted so far is still very pro-Jackson.
If anyone else wants to play around with the model here it is
(updated the model to include all 15 booths released so far):
ALP 2PP% = 1.144*(ALP primary) + 0.398*(Green primary)
error in coefficients: (1.144+/- 0.011); (0.398+/- 0.022)
standard error of prediction: 1.18%
Prediction: Labor 2PP of 48.7%.
Interesting sykesie
what do those coefficients mean in terms of how many
prefs are going to Wilkie from the
Libs, Greens and SA respectively?
I think the model says the following:
Green preferences seem to be the biggest determining factor in predicting the labor 2PP vote. (besides the labor primary vote which is obviously the biggest single factor).
I’m not sure I can answer your question directly regarding the libs/green/SA split as only the greens and labor are included in the model.
Basically green preferences are deciding who will be elected. Presumably the others are going closer to 50-50 and playing no overall role in predicting the outcome.
No you are right. Your model does not point
to a relatively uniform preference flow from
the other parties to Wilke.
With a third of the votes counted and 2PP at 52.35%, there is going to have to be some big booths for Wilkie if he is to get back in the game.
Does anyone Know which booths would favour Wilkie very strongly, over Jackson ?
It’s actually quite odd that a model for ALP 2PP ends up including a positive term for the Green vote when the Green vote is actually very likely to be favouring Wilkie over Jackson on a head to head basis. I wonder if that is because the Green vote is acting in the equation as some kind of reverse proxy for both the Wilkie and Liberal votes that is more explanative than either by themselves.
Note that the Lib preferences are actually more likely to be strongly pro-Wilkie because the Libs preferenced him on their HTV card. But it looks like where Wilkie is weak, the Greens are preferencing Jackson more heavily than where he is strong.
Anyway it does look like Wilkie will lead the 2CP over Jackson tonight, the question being by how much and will it be enough for him to be safe for that bit?
runim, at a guess, booths like Sandy Bay and Sandy Bay Beach?
I would say any booth on the city side of Newtown would be more favourable to Wilkie…that includes south, west and north hobarts, mt nelson, taroona, sandy bay, battery point, dynnyrne…there are still a number of those booths not yet reporting a 2CP.
Um, why on earth are they doing it in alphabetical order? That seems to undermine the little I thought I knew about the counting! Are the ballot papers now all pooled in some central location?
Well places like Tolosa & Lutana seem to offset any possibility from Sandy Bay, etc. Most other booths appear to be pretty close or favoring Jackson and Wilkie is a long way behind.
[runim, at a guess, booths like Sandy Bay and Sandy Bay Beach?]
Among others. Although the current 2CP is 52.3:47.7, that is coming from booths where Jackson polled 39.3% of primaries compared to his electorate average of only 36%, and also from booths where Couser polled better relative to Simpkins than in the electorate overall.
@51 & 65 sykesie
“if anyone else wants to play around with the model here it is
(updated the model to include all 15 booths released so far):
ALP 2PP% = 1.144*(ALP primary) + 0.398*(Green primary)”
Wouldn’t it make more sense to set up the model as
ALP 2PP% = 1*(ALP primary) + x*(Green primary) + y*(LIberal primary) + z*(SA primary?
You’d have enough observations now to do it (or at least without the SA term).
Hi Lukas – I did originally try it with labor/liberal/green/wilkie, but I found that predictability was best with just labor/green. Adding liberal/wilkie as well just increased the standard error. The regression coefficients for wilkie/liberal had around 100% error in the 4 variable multiple regression. I will try it again and see what happens now.
[Well places like Tolosa & Lutana seem to offset any possibility from Sandy Bay, etc. Most other booths appear to be pretty close or favoring Jackson and Wilkie is a long way behind.]
What we are seeing here is what is known locally as the “flanellette curtain”. It occurs at Creek Road and divides the Hobart suburbs from the Glenorchy suburbs.
So far we have had the following Hobart booths:
Albeura, Battery Point, Cascades, Cross Street, Dynnyrne, Elizabeth Street, Fern Tree
Wilkie has won the 2CP in all of these.
We have had the following Glenorchy booths:
Abbotsfield, Austins Ferry, Bowen Road, Brent Street, Chigwell, Claremont, Collinsvale, Cosgrove
Jackson has won the 2CP in all of those.
We have had the booth on the line, Creek Road. Wilkie won it.
Still to come:
North of the FC:
Glenorchy, Goodwood, Lutana, Montrose, Roseneath, Rosetta, Windermere.
South of the FC:
Hobart, Hobart Central, Hobart South, Hobart West, Kingston, Landsdowne, Lenah Valley, Lower Sandy Bay, Mt Nelson, Mt Stuart, Newdegate, Sandfly, Sandy Bay, Sandy Bay Beach, St Peters, Swan, Taroona, Tolosa, Waimea
More or less borderline: Moonah, Moonah North
Jackson’s 2CP is headed for a big bad demographic crunch.
Re-trying the model after 20 booths, there is still little benefit to adding the liberal term to the equation (error in the regression coefficient is around 50%). Its better than it was, but adds little to the model.
Right now (i.e. after 20 booths including the large Glenorchy Booth) the model is predicting 48.38% as the final labor 2PP.
Overall error of prediction 1.31%.
This model appears to be quite robust – it doesn’t look too good for labor.
Ta.
Obviously, I am an outsider and don’t know the area at all.
It just seemed odd, statistically (i.e. that with 40% of the vote counted, we would have expected a representative selection of the vote) , that there could still be hope for Wilkie.
Can we at least say the vote will be closer than the projected 57.5/42.5 that the ABC computers gave it on election night?
Well Glenorchy and Goodwood are in and ALP is 54.67%. The Hobart booths were all won by the Greens with ~30% pv. We will have a much better idea when we get to the Ks.
Given the widely mixed messages in the recent posts, is it possible to say anything definite at all about Labor’s prospects of winning the seat?
Time for predictions – at the end of the current 2pp count Labour will be on 48.5% but the incumbent vs independent advantage in the postals will give Labor a narrow victory
Oakeshott Country makes a valid point – the postals may well be the deciding factor – but on the raw 2pp count it looks increasingly difficult for labor to be in front.
Does anyone know what time they stop counting in Denison tonight ?
[Can we at least say the vote will be closer than the projected 57.5/42.5 that the ABC computers gave it on election night?]
Absolutely, that projection was better than the Simpkins/Jackson 2CP but still rubbish. It was never going to be that wide.
[Given the widely mixed messages in the recent posts, is it possible to say anything definite at all about Labor’s prospects of winning the seat?]
If sykesie’s model holds up Labor will be too far behind on 2CP to catch on postals. That will remove hurdle 3 from the equation and Labor will need Wilkie to fall at one of the other two. So the way it is going today’s counting will improve Wilkie’s chances.
[Time for predictions – at the end of the current 2pp count Labour will be on 48.5% but the incumbent vs independent advantage in the postals will give Labor a narrow victory]
I don’t know about this, because it is not like a big-effort marginal where the incumbent has been trying really hard to get all the postals. It’s a seat taken for granted with a token party postal vote effort (see #4) – except perhaps by the Greens. Would postals save Jackson if he was 51.5:48.5 behind? I have my doubts about that. Are there any comparable cases?
The only positive thing I can say for labor at this moment is that my model under predicted the first two Hobart booths by about 1.6% for Hobart and 2.4% for the smaller Hobart Central booth.
In the Denison state election, Wilkie got 7.7% of postals, compared to 8.5% of all other vote types combined.
@79 sykesie
“Re-trying the model after 20 booths, there is still little benefit to adding the liberal term to the equation (error in the regression coefficient is around 50%). Its better than it was, but adds little to the model.
Right now (i.e. after 20 booths including the large Glenorchy Booth) the model is predicting 48.38% as the final labor 2PP.
Overall error of prediction 1.31%.
This model appears to be quite robust – it doesn’t look too good for labor.”
Sykesie, I take it that you’re now including Liberal in your model?
Can you tell me what your coefficients presently are?
If it was underperforming the simple labor+green model before, it would only have been because of the small N.
@87, 88 Kevin Bonham
Kevin, what do you reckon is the minimum 2PP that Labor must achieve from the booths in order to get up on postals?
I should add, thanks to both Sykesie and Kevin B for great contributions here.
————–HELP PLEASE————————-
The AEC is saying that Labor got a primary vote of 48% in Denison, which is no swing. However, Im looking at the 2CP count and in almost every booth theres a swing of about 10% against Labor in the primary vote. Whats going on?
[The AEC is saying that Labor got a primary vote of 48% in Denison, which is no swing. However, Im looking at the 2CP count and in almost every booth theres a swing of about 10% against Labor in the primary vote. Whats going on?]
A stuffup in the displays, that’s what. The primary figure showing for Jackson, both number and percentage, is completely wrong. It might be Duncan Kerr’s total from last time! Correct figures http://vtr.aec.gov.au/HouseDivisionFirstPrefsByVoteType-15508-194.htm
[Kevin, what do you reckon is the minimum 2PP that Labor must achieve from the booths in order to get up on postals?]
I don’t like their chances if below 49.0 but I have no really comparative situations to apply to that so I may not know what I am talking about.
Aaahh, iv found out whats up with the AEC’s numbers. According to the site, theres been 13,136 early votes cast in Denison, which is absolutley plausible. The thing is that the AEC have added ALL those votes to Labor’s primary vote, which of course is a mistake.
They have now counted 99.42% of votes according to the AEC.
Forget that and check this.
Jackson has 100% of the 13,136 pre-poll votes.
However it is remarkable that the number of pre-polls is exactly the same as the current 2pp for Labor. The pre-polls for 4 of th5 centres have been counted in and they only come to about 3,500 – the divisional office must have been like Pitt St if they took the other 10,000 pre-polls
OK so I was a bit behind the other comments, cos I was interupted by a phone call and then posted the comment above.
Has the AEC stopped for dinner or just gone home?