Having run the new quarterly Newspoll figures through the simulation on the old boundaries, it might be time to run them with the new post-redistribution boundaries (thanks Antony Green for the data!) to see the most likely result were an election held in the last three months on our new boundaries for 2010.
The results – the most likely outcome is the ALP winning 102 seats in Parliament, a 19 seat gain on their current position. The charts come in like this and can be read the same as last time.
The reason why there isn’t a great difference in the results between the simulation on the old boundaries and these new ones comes back to the size of the swings being seen, combined with just where on the pendulum the seats were that had their margins changed. Basically, the swings at the moment are large enough that a very similar number of seats would have fallen under both sets of boundaries. We dont really see the 5 to 6 seat headstart that the ALP received from the redistributions come into play until we get much smaller swings. With a national two party preferred of 55/45 (and the State swings that combine to produce that national swing) it doesnt actually make much difference, but 53/47 on the other hand, a much smaller swing, produces far more seats for the ALP under the new boundaries compared to the old.
Depending on how the state swings would pan out that made up a national two party preferred of 53/47, we’d expect to see that small 0.3% swing to Labor to deliver somewhere between 4 and 9 seats thereabouts. We can see the “curve” of the seats gained for a given swing under the new pendulum with a simple chart (or course!)
The vertical axis is a given ALP swing – positve is the swing to the ALP while negative is the swing away from the ALP. The bottom axis shows the nth numbered seat in a new Parliament that the ALP would get from a given swing. As you can see, the curve has effectively shifted to the right for the ALP, delivering more seats for a given result, yet it really plays out with swings 1.5% towards the ALP or less.
So the new boundaries certainly benefit the ALP, but the ALP benefit is relatively stronger with small swings than large ones. The larger the swing, the less benefit they get relative to what they would have achieved under the old boundaries.
UPDATE:
Thought it would be worth adding a couple of extra charts. The one above, but with the same curve done with the old boundaires, plus a zoomed out version with both as well to highlight how the big change in the curve occurs at the low swing values.
29 Comments
The new boundaries just shuffle around the sitting member factor. A bunch of Liberals who campaigned super hard right up to the last street on the edge of their electorate find that their seat is notionally lost to Labor once areas they didn’t campaign in get included. I think if you had a very small swing to Labor, the pendulum might be relatively unreliable because all the new margins in really close seats have the sitting member factor jumbled.
Have you every looked at the sitting member factor in terms of government vs. opposition members? Just wondering if the sitting member factor would be slightly larger for government members because of the additional derived electoral resources that often come with government?
It’s larger for government members. The only time I notice it directly is to look at how the declaration vote swing changes after the defeat of a sitting government member. At the next election, I’d bet that the Coalition will struggle in seats like McEwen, Paterson, Gilmore, Hughes, Bowman and Herbert because the sitting Liberal MPs will not have those pre-prepared direct mail letters from the government secretariat explaining the goodies the Coalition government is delivering for the electorate.
I am very skeptical about the effect of political direct mail.
I am certain political operatives will have a range of models predicting this outcome from this type of mail campaign, and that outcome from that type of campaign. But I am skeptical.
I have no experience with political campaigns, but plenty with commercial campaigns, and all my experience has told me, time, and time again, apart from very specific situations for very specific outcomes, which are easily measurable, direct mail is a mirage.
I’ve seen plenty of predictive models from agencies, and the only thing it predicted correctly was how much it was going to cost me to follow the suggested campaign!
Still that want stop the parties pouring plenty of resources down that black hole.
Political direct mail is highly targetted. Members have huge knowledge of people interested in specific issues. Information on voters interested in Veterans affairs, pensioner benefits, assistance for young familes is highly sought after. At various times, parties have rung everybody in the electorate after basic information, and whenever you move into a marginal alectorate, you will be contacted by your local MP. This is a huge advantage for the sitting MP of the governing party.
You really notice the impact when a sitting MP is defeated. Suddenly a seat that has refused to budge with the swing for years suddenly moves.
In recent years parties have been spending more and more money on untargetted direct mail. It has become a sort of mutually assured destruction, where both sides spend more becuase the other side is spending more. It’s why both parties are now talking about expenditure limits on campaigning.
I’ve no doubt Antony, they believe their direct mail is highly targetted, and that they make every effort to target effectively, and in small specific cases this can be very useful for the constituent and the member, but for making a real political difference, again I am skeptical.
Apart from what I noted above, I moved from Kingsford-Smith, into Eden-Monaro, before the 2001 election. I’ve now experienced 3 Federal Elections and the only difference I noticed (and remember I’m not only an engaged political type, but also a professional marketer – so I was actually very interested) was more generic direct mail.
No targetted efforts, just more effort, and wasted effort at that.
I was actually quite surprised at how amateurish the whole effort was.
What you didn’t say is if you had been contacted by the MP when you moved into the electorate, and if you were, did you nominate issues you were particularly interested in? Young families, pensioners, veterans etc. I’ve seen the targetted letters the parties produce on issues like that. It’s not generic stuff.
People like Jackie Kelly, Trish Draper and Danna Vale built profiles through their direct campiagn efforts. I can list successful Labor MPs in the Hawke and Keating governments doing the same thing.
Oddly, in a seat like Eden-Monaro, direct mail is less important. IN a city seat, you need direct mail to get your name out to voters because local MPs get no radio or television coverage. In a regional seat, television advertising is much cheaper than in the city. Because of networking, regional stations have all these advertising spots to fill in networked programs. Bob Baldwin has retained Paterson at recent elections by flooding the airwaves with ads. TV advertising is the killer method in regional seats but it doesn’t work in the cities.
Antony, here in Paterson, not only did Baldwin saturate the television, but the Labour Party played dead. Strange behaviour considering Rudd was saying that every seat counted. I think at the very end, there was one 30 second slot featuring the Labor candidate (heck I’ve even forgotten his name) and that was it.
There’s only one Liberal in the Hunter, so the Liberals spend everything on that one seat. Labor holds four seats. Labor would have had to match the Liberal Party’s spend just to establish Arneman’s name where the Liberals were just re-enforcing Baldwin’s local work over nearly a decade as MP. I’ll guess Labor looked at their polling in the seat and decided it wasn’t worth the spend. Arneman had already lost the state seat of Port Stephens earlier in the year. Labor head office had already used up all its brownie points in the Hunter in imposing Jodi McKay on Newcastle and Greg Combet on Charlton. The margin was 6.3% before the election, and that’s a big swing to get against a sitting MP with name recognition in a regional seat. There was better low hanging fruit elsewhere.
I bet Labor picks its candidate carefully and runs hard this time.
Direct mail is known to have an impact! in 1998 Bruce Billson conducted a mail out on the Thursday before polling day in which he practically announced a series of policies called Dunkley Destiny.
While i suspect Bruce Billson would have held Dunkley in 1998 but still it was seen by many political watchers to have been a masterstroke for it was too late for the ALP to respond
“Dunkley Destiny” no less! Christ on a stick, where do they come up with these names!
So Possum,
Trying to understand these simulations a bit better.
You take the current polls with their errors and do a monte carlo The polls are national only. Therefore, you must assume a uniform swing (in this case to labor) with an error. Right?
However, I think I remember that in on of your recent posts that this is not what was happening. Voters are swinging more to labor in the city and less in regional and country? (maybe I’m totally wrong. There was that post about age and swings being different for age… even if you didn’t it seems plausible).
Anyway – what I’m getting at is that if you’re assuming a a uniform swing that doesn’t seem to represent reality. In the seats that changed hands in the last election the swing was as little as 0.19% and as big as 13.20% (without a retinerment)
http://en.wikipedia.org/wiki/Australian_federal_election,_2007#Seats_changing_hands
Maybe it does a good enough job for this senario. Also maybe the change in swings just cancel eachother out and all the non-uniform swings would do is slightly change the bell curve.
But still, maybe it has a huge bearing. Would there be a way of taking some characteristics of an electorate (demographics, region, state?) and predicting non-uniform swings?
A new maxim, it seems:
Death, Taxes and Labor to win >75 seats.
Poss! it was a booklet of several pages, with Bruce’s big smiling face on the front. at the 2001 election he recycled it with ticks next to what he had delivered and named that Dunkley Destiny 2
Baldwin lost nearly all of his majority even as it was. Something went seriously wrong in the Labor organisation I suspect. It didn’t just spread its funding out – it just played dead. IIRC, the TV ad for Arneman was actually paid for by the unions. It was a well produced ad btw, which made it more bizarre that it only got 30 seconds of air.
No, Antony, I’ve never been contacted by anyone, not when Gary Nairn was the member or since Mike Kelly’s been the member.
As I mentioned I was suprised by how amateurish all three campaigns were.
Direct mail, like a range of techniques used by political parties: advertising, PR, research, strategy, branding, was learnt from us (ie the marketing profession) starting mainly in the 1980’s during Hawke’s period.
We taught them about demographic and psychographic segmentation, target marketing and perceptual positioning, and most involved still don’t fully understand them. Let alone, research, PR and advertising.
None of the techniques have been implemented as well as they can be, and some, very poorly, because most of the political operatives don’t quite understand the underlying reasoning. And they have no interest in understanding the philosophical principles from which the techniques are derived, which is a key part of the problem.
Specifically on direct mail. It has two main benefits:
1. It allows you an additional way to maintain a relationship with your customers (voters)
2. It is directly measurable, as you can immediately measure your response rates, as opposed to other forms of communication.
Both benefits, however, require the maintenance of a data base. The only way a useful database can be developed is with information from your customers (voters). This they have to be willing to provide to you. Once they do this, then yes, you can provide targetted information to match their needs, but unless they give you that information, you’ve got nothing useful, except an address.
You can only get this information by contacting the voter, either by phone, or by mail, and then you need them to be willing and interested enough to provide the data you want. By definition, these people are more highly engaged.
As I said, I’ve little doubt they are convinced it is effective, and will insist that is why they held this or that seat, and I’ve no doubt whatever targetting they have managed has not detracted from their position, but I think they are overstating its importance.
Hence, as a professional, I am skeptical.
Tristan,
I don’t actually use the national swing.
What I do is first, take the state two party preferred from the Newspoll quarterlies and then combine those with the state breakdowns of the Nielsen polls that occurred over the same relevent period – so I have two sets of state breakdowns for federal polling, the Newspoll State TPP’s and the Nielsen aggregated TPP’s.
Next I combine those two measures for each state to arrive at a single two party preferred estimate for NSW, QLD, Vic, SA and WA (by weighting each measure by sample size, so the larger Newspoll sample has a larger weight compared to the smaller Nielsen sample, for example)
From that aggregated State TPP figure, I change it into a swing – the difference the current polling is from the last election result.
That gives me the mean swing for each state.
Next up I create a standard deviation for that swing in each state which is a combination of both the historical standard deviation that each state has experienced over the last 5 federal elections, plus the standard deviation of the polling itself based on the margin of error that derives from the combined pooled sample size of each state.
Then I create a normal probability distribution of the swing expected in 147 electorates (I ignore the indies and assume they will hold their seats) where the mean of that distribution for each seat is the relevent State TPP figure and the standard deviation is as calculated above.
Next comes the actual simulation.
At simulation iteration number 1, each electorate has a random number generated from within their respective probability distributions. That number is then added to the prevailing two party preferred vote that the ALP received at the last election (or in the case of redistributions, the estimate of that TPP according to Antony’s hard work). A positive swing will create a higher TPP, a negative swing will create a lower TPP.
For the seats in Tasmania, the NT and the ACT I use the national swing and do the same thing as above.
After the first iteration is finished, all 147 electorates are then “called”. If the estimate for the ALP TPP is greater than 50% the seat goes to the ALP, if it’s less than 50% it goes to the Coalition.
I tally up the results and end up with 1 single iteration of a simulated election.
I then repeat that simulation 20,000 times (but it can be as small as 10,000 and as large as 100,000 – it depends on how long the simulation takes to stabilise) and built up a distribution of the simulated elections, the results of which you see above.
I dont add demographic variables (but I might closer to the election). What I use is just the estimated state swing, since the state swing is the best predictor of the outcome in any given seat, being far superior to the national swing.
Beemer went:
Oh good grief – fark me dead!
If I were a Liberal or National Party Member – then I would be insisting that Malcolm and co reject any negotiation around the ETS – thus ensuring a double dissolution and the glorious re-election of the mighty coalition to government – so we can resume our rightful place on the Treasury benches.
And then I would find a calculator or a slide rule and have my electoral staffer calculate the majority that we would enjoy as a result.
As we march into oblivion.
So I’m looking at the polly demongraphics…
Maybe you already do something like I mentioned in my above comment?
Poss – all up how many people do the quarterly results cover?
Jimmy, this quarter Newspoll had 6879 respondents. Nielsen had 2800 on top of that – so the state breakdowns came from a pooled sample of 9679.
Wow that’s a lot of number punching into a calculator
Geez, what did peope do before computers???
The keys are certainly worn out
Stochastic simulations were really something that came after world war 2 with pre-electronic computing, although there were some early forms of simulation – Mario Lazzarini attempted to solve the Buffon’s Needle Problem by throwing a needle on the floor over three thousand times and collating the results to estimate π
Some say he wasn’t the full quid.
Mario Lazzarini was one of my favourites on:
“World Championship Wrestling”.
The Buffon’s Needle? I’m sure that’s in there with the abdominal stretch and the sleeper hold. Mario was well talented at both!
Thanks for that Possum.
I don’t do monte carlo analysis myself, but really want to find a reason to use it.
Anyway, as a person who loves to see what different assumptions do to these models I would be very interested to see if you could include some demographics and more geographics into the model.
Yep, it seems reasonable that state would provide better prediction than national swings. But still that’s just breaking down to less uniform swings.
But is it better than region? (eg. urban, suburban, regional, semi-rural, rural)
How about those retirees moving to the gold coast?
What about the employment sector?
I’m not sure what sort of stuff there is enough data on to include (or worth going to the trouble including) and the more valiable you include the more difficult it would be to sort through interdependancies. But then, maybe possibly, you might start to see some more interesting stuff. However, it might be a pipe dream here…
Lets have a poll on preferred PM from the Labour Party ranks – K Rudd, J Gillard, L Tanner ?? hasn’t been done yet as far as I know. Would love to know if Kevin is really as popular as polls between he and MT suggest. I suspect there are quite a few out there that find his manner and behaviour, self-proclaiming speeches etc a bit disturbing. A poll like this would give a great perspective on how Australians see him.
Well – I would pick JG ahead of KR or LT