Ever wonder how ‘at risk’ your local member might be?
Now that we have the full spectrum of quarterly polling data from Newspoll and Nielsen, not only can we run election simulations, but we can also use those simulation results to take a closer look at which seats would have been most likely to fall were an election held sometime over the last 3 months and the result was consistent with the polling.
We can split these seats up into 3 categories:
Extreme Risk – being those Coalition held seats that would almost certainly have fallen to the ALP. It would have taken an historical anomaly for any of these seats to have been retained by the Coalition.
High Risk – being those seats that would probably have fallen to the ALP. A large majority of these seats would have changed hands.
Moderate Risk – being those seats which would have been in some danger of falling to Labor. For nearly every High Risk or Extreme Risk seat which did not fall, there would most likely have been a seat in this Moderate Risk group that would have taken its place.
What’s worth noting is that there are a fairly sizable number of blue ribbon Liberal seats in the Moderate Risk group as a consequence of the size of the various swings going on at the moment in each state. So for every seat like Goldstein or Higgins that would be really unlikely to fall – those rusted on Liberal Party base seats – the risk to those other seats in the Moderate Risk category increases to take the probability weight. This suggests that seats like Casey would have been more at risk of falling than their nominal position on the electoral pendulum would ordinarily suggest. It’s also worth noting that every election throws up an odd bolter of a seat – some electorate that wouldn’t ordinarily be classified as being at risk by the simulation, yet which still ends up falling with a relatively huge and unusual swing.
This is what the list looks like (over the fold), including seat name, the state, sitting Coalition member and the mean simulated two party preferred result for the ALP. The aggregated polling suggests that there are no ALP seats currently in danger of falling to the Coalition, yet that necessarily ignores any major issues that might be happening at the local level which could cause an ALP seat to fall – such as with Belinda Neal in Robertson.
All these results are based on the new electoral boundaries for 2010.
27 Comments
Thanks Poss, much appreciated.
You often remind me of a swan, not a marsupial – glide along as though it was all so easy, but you are paddling like hell underneath.
I take it the simulation result is the probability of a particular result, expressed as a %?
Top stuff, thanks very much.
Dumb question (I’ve got a million of ‘em!) – is it of any use to compare this to the classic election pendulum? Are they comparable in any way?
At the moment I see how yours works, and how the election pendulum works, but I can’t reconcile the two.
As always, nice analysis.
For the benefit of Possum’s perusers I also undertook an analysis of the quarterly aggregates (Newspoll & Nielsen) and one Galaxy (QLD) and one Westpoll (WA), and found the same 26 seats, that Possum notes as at extreme or high risk.
It might be reassuring to readers to see that we both came to the same conclusion, independently.
It’s here if anyone’s interested:
http://www.ozforums.com.au/viewtopic.php?id=6699
Don – there is a strong relationship between this and a normal election pendulum.
Whereas the ordinary pendulum is just a single list that we apply a uniform swing to, this is actually 6 separate pendulums – one each for NSW, Vic, Qld, SA,WA and the final one I create for Tassie+the territories. I use a state based approach because the biggest determinant of how any single electorate will go in an election isnt the national swing, but the state based swing.
The other difference is that the simulation allows us to accommodate more accurately the uncertainty involved in some seats swinging larger or smaller than the uniform swing. The ordinary pendulum assumes that seats will fall equally above and beyond the national swing, but here we attach probabilities to that happening, guided by the state swings, historical variance *and* accommodate the inherent uncertainty in the polling numbers from sampling error.
Sometimes seats aren’t actually more likely to fall equally above and below a uniform swing because they might be clustered. There might be, say, 5 seats sitting half a percentage point below the national swing. A pendulum approach assumes none of those will fall above the swing unless some other seat that should swing more actually swings less (the pendulum needs replacements). A simulation on the other hand will allow 2 of those 5 seats to fall above the national swing on the basis of probability alone.
Ordinarily that might not make a great deal of difference – but it will often be the extra 10% in accuracy that we need to get as close as possible as we can to a likely result.
When I re-do these sims for previous elections using the final polls of each pollster in the campaign and compare the simulation results to the actual election results, the simulation is always at least as accurate as the standard pendulum, but often more accurate than the standard pendulum approach.
Does the Newspoll for South Australia cover the NT in its sampling?
Is the ACT covered in NSW sampling?
Speculatively the seat of Solomon that was won very narrowly by the ALP in 2007 might be one at risk for the government – due to the personal difficulties of the current ALP member, but then again politics in Darwin is pretty wild and woolly anyway and it might not make any difference at all.
Good stuff Poss – unfortunately I don’t see Mackeller on the list
I know the Newspoll is geared towards House of Reps, but using the demographic info, is there anyway to apply something, that would give a good indication of the senate, in a half senate election?
Avalon Dave
Doug,
NT isnt included in the SA figures with Newspoll, but is with Nielsen. When I aggregate the figures from the two pollsters, I widen the uncertainty for SA to accommodate the Nielsen component that includes the NT in the South Oz results. Historically it doesn’t actually make any difference, but I just to it that way to be cautious.
ACT figures are included in the NSW figures for Newspoll but not for Nielsen – so I do the same thing there, just the opposite way around, but again, historically and thankfully, it doesn’t make any real difference.
Avalon Dave,
Things would really have to get politically gruesome for Mackeller to ever make such a list!
On the Senate – because the last spot or so in every State is usually decided by preference deals, it makes it hard to do something like this for the Senate before those deals are actually done. Even after the preference deals are completed, the uncertainty involved is still so large as to make it a dubious exercise.
So saying, I’d like to have something ready for the Senate along these lines when the election campaign starts. But at this stage it’s still a bit of a wishful thinking work in progress
Possum
thanks for the prompt response
Greatly enjoy your work
Possum – I live in Aston, and our local member (Chris Pearce) is retiring, with the Liberals already having pre-selected Alan Tudge, a former Liberal staffer. Would I be correct in thinking that the seat is at an even higher risk with Tudge – who does not live in the seat – compared to having Pearce, who I see every couple of months at my local shopping centre and has held the seat since 2001?
Love your work, cheers.
Max, you’d be right.
Being a sitting member is worth around 1.5% on average to the two party preferred vote.
Possum: Happy New Year to you, and thanks for your latest insightful piece of analysis.
Alas Berowra will never be at risk of falling to Labor, but I live in hope.
It really is a pity to see Joanna Gash, my local, in the High Risk zone.
I despised much of what the Howard government stood for and was glad to see it ousted yet still voted for Joanna Gash from the time I moved to the South Coast of NSW. She’s by far the best local member I’ve had and no matter what their ideology a good local member should be kept in.
On the other hand, it’s a real pity to see Cowper only in the moderate risk group. Labor would have won it in 2007 if they’d bothered campaigning – as it was, they got far more press for their preselection squabbling than the actual campaign, which was a non-event of the highest order.
Much as I’m not impressed by having Hartsuyker as a local member, unless the ALP actually choose a decent candidate, and run a half-competent campaign, they probably don’t deserve Cowper, no matter how much it’s there for the taking.
I wonder if the Nats will throw as much at Cowper as last time. Given how marginal it is, I’d be guessing they will.
@Max Laughton
Good pickup on the ‘Tudge Factor’.
I know many Aston voters wont appreciate someone coming in from outside the seat.
I still think the Libs will hold onto the seat, but it will be interesting to see who the ALP put up for a run. Perhaps Gerry Raleigh for another run?
Dear Possum,
I’m just wondering about Wentworth. There are any number of reasons that it may not be on the list, but I know Labor did really believe (incorrectly) that it was winnable in 07. Have you written it off because of the lack of a swing last time? Or is there some other reason?
Nic Stuart
Possum, I’m interested by your figure of 1.5% as the advantage of an incumbent member. Is this a guess or do you have data to back it up? (I know its just an average and will vary a lot according to the quality of the member. Also, is it the same for federal and state? I’d have thought it should be larger for state seats, as they’re smaller and easier for the MP to meet a large portion of the constituents, but maybe that’s balanced by state MPs usually having fewer resources.
Nic at 15: probably because of the small swing to ALP in NSW, plus the OK sized margin it currently has (It was one of only four seats in the country to swing towards the Libs in 2007). Somebody posted a link on LP to Antony Green’s calculator with those state swings plugged in, and it spat out a 2.1% Lib margin for Wentworth. It’s done a little bit differently to Possum’s (just a flat swing, no fancy Monte Carlo stuff), but I guess it turns out similar.
Chris @ 14
I believe the preselection for Aston has already been completed.
Community and Public Sector Union deputy secretary Rupert Evans is the labor candidate. I am not sure who else contested other than Monash Mayor Paul Klisaris.
http://www.knoxjournal.com.au/news/local/news/general/unionist-up-for-aston/1654419.aspx
Poss,
Nice work. Is it possible from the figures to determine how at risk members are to Greens, i.e. Lindsay Tanner in Melbourne?
IN
Thanks Poss, an excellent analysis. This aggregated data hides the fact that the unpublished Newpolls MUST have been bad for the Liberals. They would have loved to publish a negative poll for the government before the Christmas break, but it was obviously the opposite result. News Corp in Australia now is simply a partisan right-wing organisation, as it is in the USA.
Looking back over that totality of polling in the past year, we seem on track for a substantial further swing to Rudd in the next election, including no doubt the end of Fielding/Mr X balance of power in the Senate. A robust sample indicating over 100 seats suggests the result could be historic, taking perhaps two more terms to reverse, assuming the Liberals reform. The recent old guard shadow cabinet selection suggests the opposite. They really should be thinking long term in their pre-selections, because on these numbers they won’t win government before 2015.
Nic,
Wentworth didnt make the generic list simply because the swing in NSW toward Labor is currently relatively small.
Feral Sparrowhawk,
That 1.5% figure is what I get as an average when I look back over the last 6 or so elections. Everyone else seems to get the same approximate figure as well – Antony’s election calculator for instance has that number built into it when a member retires. As for the size of the effect at the State level, I honestly couldnt tell you. Antony might know though – he’d probably tell you if you asked him on his blog.
Imaginative nickname,
Unfortunately not since I’m using state based two party preferred swings rather than primary votes.
Socrates,
The infamous unpublished Newspoll must have been around the 56-59 mark looking at the way the aggregates panned out, including the 52/48.Although, I still strongly believe that there was nothing nefarious behind the non-publishing.
I’ve done regression analysis on swings after various elections and stuck in dummy variables for retiring MPs and changes in donkey vote. You always get about 1.5% out as the retiring MP factor. Clearly this is an average, as some retiring MPs will be brilliant and others will be dickwads. A sitting member with a personal vote above 3% would be highly unusual. A retiring MP in a regional seat, like Labor’s Harry Quick in Franklin in 2007, is an example, which was why Labor reacted so strongly when Quick expressed disapproval of the first chosen replacement Labor candidate. It only ever matters in a handfull of electorates, so unless you’re talking about a government being elected by half-a-dozen seats, the importance of sitting member factor is overstated.
Once you account for state, you account for so much of the variability of swing from seat to seat. Factors like retiring MPs and donkey votes always pop up as a small constant. Factors for both of them are built into my ABC calculator just to randomise the result a bit from the application of a uniform swing.
I love all the maths in your Monte Carlo simulation, but it nearly always produces the same headline seat number as just applying a uniform swing. It works much better on US Presidential elections because the states have different electoral college numbers, and the Monte Carlo method works much better on this than any uniform swing method. But you do get nice analysis of the probability of individual seats.
I’d be amazed if Labor got about 55% 2PP at next year’s election, and I think 53-54% is more likely, so I hope your loyal readers understand that all your probabilites and classifications apply to a hypothetical election held in the last quarter of 2009 and are not a precition of the 2010 result. If only an opinion poll could come with statistics predicting the course of future opinion.
that ‘precition’ in the 2nd last line should be prediction.
Thanks for that Antony.
The monte carlo analysis – as you say – really comes into its own when you get large semi-autonomous blocks of votes to analyse like the electoral college in the US. In Oz, it’s always going to come to a result close to the uniform swing approach unless there’s a large cluster of seats sitting on the margins of the swing required to fall.
It has an extra benefit though of letting me extrapolate those – now regular – Auspoll exit poll results on the night of an election before you take over with the actual results themselves.I can use the monte carlo in conjuction with some regresion work to expand a state or national result from an exit poll of, say, the 20 most marginal seats, into to a state or national wide total result.
I hope the readers understand too that these things are rear view mirror stuff – an account of what would likely have happened were an election held and the results mirrored the polls.
Which brings me to the point – please folks, for the sake of Antony having to field phone calls from journos asking silly questions about this type of analysis – it’s an exercise in what *was*, not what *will be*. For the latter, you can call 1900 POSSUMSNAVEL !
Damn interesting and informative stuff PC, thanks.
However, Poss, as you say yourself, the trend is your friend.
It would be absolutely extraordinary for the result of a near future election to be too far from the long term-trend.
The only thing that really is in dispute is the size of the ALP majority and whether it increases or decreases. Will the ALP crack the 100 seat mark and thereby capture 2/3 of the seats available?
Just for clarification-
If an election was held during that quarter then the Libs would have:
1. Almost certainly lost their education and health shadows (Pyne and Dutton) + 4 parli secs
2. Likely to have lost their small business shadow (Billson) + another 4 parli secs
3. Be nervous about another 5 shadows (Robb, Smith, Hunt, Andrews, McFarlane).
Losing 2 shadows is unfortunate, losing up to 9 is just plain careless. Its like someone turned the lights on and all the shadows disappeared.
And this is a potential high point- If Abbotts approval dives like many other previous opposition leaders and takes their primary votes with it god knows where they will end up.