With an election nearly 12 months away we have added a new seat projection feature to the Poliquant Australia Poll Average on the Australia page. The projection is similar to what can be found at threehundredeight.com for Canada, UK Polling Report for the United Kingdom and FiveThirtyEight for US presidential elections. However the seat projection does use its own distinct method.
Seat Projection – Introduction
The purpose of the seat projection is to forecast the outcome of an election based on the Poliquant Australia Poll Average. The seat projection does not forecast the outcome of individual seats. The reason for this is swings are not uniform at elections and there cannot be used conclusively to determine whether the seat may change hands.
As we have previously discussed on this website, electorate swings whilst they are not uniform, do differ in a predictable way. Electorate swings at elections are generally, normally distributed. For non-statistical types this is commonly termed as the “bell curve”. Variables which are normally distributed are mostly found (95.6%) within 2 standard deviations of the mean. Breaking this down for our exercise; electorate swings are mostly within 2 standard deviations of the overall swing.
So what is the standard deviation for electorate TPP swings? When we broached this topic for the Queensland election we used 3% as the standard deviation. Prior to starting the seat projection we had a look at the standard deviation of TPP swings at the last 7 federal elections on a nation-wide and state-wide basis the table is below.
As you can see nation-wide the standard deviation averages at 3.24%. However, the state based standard deviations are generally smaller at around 2.5%. Ever since federation, state behave differently at federal elections. Any seat projection model needs to take this into account and thus a state based model is preferable.
With the information about standard deviations of electorate TPP swings and the assumption of normal distribution of such swings we can now build our seat projection model.
Seat Projection Model – Methodology
The methodology for the seat projection model is as follows.
- For each seat we calculate a probability figure between 0 and 1 for the seat to be won by either major party, given a specific state-wide swing from the poll average and a standard deviation of 2.5%. For the statistically minded, what we are doing is calculating the cumulative probability function for a TPP swing against an incumbent major party to exceed that seat’s TPP margin.
- For seats in Tasmania and the Territories, the national TPP swing from the poll average is used in place of the state-wide swing.
- Seats which are expected not to be won by either Labor or the Coalition are assigned to Others.
- For each mainland state the seat probabilities for the major parties are added together and rounded to the nearest whole number to calculate the seats won a major party in a state. If any seat in a given state is expected to be won by a non-major party candidate, the major party victory probabilities for that seat are not included in the state summation.
- Step 4 is done for Tasmania and the Territories as a group
- The rounded seat distributions for each mainland state and “Tasmania and the Territories” are summed to created a seat projection.
To give you better idea about how the how the probabilities work, lets use a hypothetical example. Take the seat of Richmond which has TPP margin of 7% for Labor and assume a NSW state wide TPP swing of 8% to the Coalition. For most observers, they would assume that the Coalition would win Richmond. Our model uses the cumulative distribution function (assuming a normal distribution) to calculate the probability of the Coalition winning the seat, being 0.655. In other terms given the above variable applying to Richmond, the Coalition has a 65.5% chance to win the seat.
As after the distribution of preferences it is usually Labor and the Coalition which are the TCP candidates. As the TCP vote always adds to 100%, the Labor probability is 100% less the Coalition probability. In the case of the above Richmond example the ALP probability to win the seat would be 34.5% or 0.345.
When Australian political observers use the Mackerras Pendulum to estimate elections seat outcome they are doing the same as the above process, however, they assign probabilities of 1 and 0 instead of fractions as the seat projection model does. In this aspect and the usage of state swings, the seat projection model is more sophisticated than extrapolating a result from a Mackerras Pendulum using a uniform national swing.
Now we turn to niggly bits of the model; assumptions about non-major party seats and the Tasmanian polling blind spot.
Minor Party Assumptions
In assuming the chances of non-major party candidates we first asses the re-election chances of those minor party and independent MPs currently sitting in parliament. We have been lucky to have some degree of public electorate polling for these MPs and it is the polling, where available, which forms the basis of our assessment on these MPs re-election chances. Poliquant ignores internal party polling that is reported in the press in its calculations. We look at each crossbench MP in turn.
Tony Windsor – New England
There have been 2 polls undertaken that I am aware of and they are summarised below. While both polls have small samples of approximately 500, they are the only polls we have. As Tony Windsor did not lead in either, we will assign this seat back to the LNP-ALP contest where it is a safe Coalition seat.
Rob Oakeshott – Lyne
Three public polls have been conducted for this seat since the last election and they are summarised in the table below. All have the Coalition leading Rob Oakeshott. We will assign this seat back to the LNP-ALP contest where it is a safe Coalition seat.
Andrew Wilkie – Denison
There has been only one public poll for this seat which I am aware of, a Reachtel poll taken in June which had Andrew Wilkie well in front with a significant drop in the Labor vote. The details are in the below table.
As Andrew Wilkie won this seat at the last election with only 21.2% of the primary vote, preferences from the Greens and then the Liberals were vital to his victory. Peter Brent at Mumble provides an excellent analysis of the Denison situation here. As Mr Brent points out Greens preferences are the key as they will likely finish fourth in the primary vote. If Wilkie cannot increase his primary vote substantially he will likely need Greens preferences to leapfrog either the Liberal or the Labor candidate into the final count.
However if Wilkie can improve his primary vote as indicated in the Reachtel poll he will win absent a major party preference stitch-up. We will keep this seat in the others column.
Adam Bandt (Greens) – Melbourne
No polls for this seat. However the big development is the Liberals preferencing Labor above the Greens in the Victorian state election held 3 months after the last federal election. If repeated at the next federal election, Adam Bandt would need to improve his vote to win. We estimated the Greens situation in Melbourne (and other lower house seats) in this post. In short, if the Liberals preference Labor ahead of the Greens on their HTVs, the margin would effectively turn to 54.5-45.5 ALP. Until public polling emerges showing Adam Bandt in a winning position in Melbourne, we will assume the seat will return to Labor. We will assign this seat back to the LNP-ALP contest where it is a safe Labor seat.
Bob Katter (KAP) – Kennedy
No polls for this seat. Since the election Bob Katter has formed his own party, the Katter’s Australian Party. The KAP had great success for a new minor party at the Queensland state election in March winning 11% of first preferences and 2 seats. On this basis and the fact he has won the last 4 elections as an independent and in absence of any public polling to the contrary, we will presume Katter will win the seat and we will keep this seat in the others column.
Peter Slipper – Fisher and Craig Thomson – Dobell
Both MPs were members of major parties and won their seats at the last election as major party candidates. Unless public polling arises which puts either of the MPs in a winning position as an independent we will keep these seats in the LNP-ALP contest.
Any other candidate
As far as we are aware at the current time there are no other minor party or independent candidates with any chance of winning a seats at the next election. Public polling proof will be needed to rebut this presumption for any effect on the seat projection model.
Minor Party and Independent Candidate Summary
In short we have pegged others at 2 seats; Kennedy and Denison. These seats will be kept out the probability sums in the model calculating ALP and Coalition seats.
In the absence of public polling with breakdowns for Tasmania we can only assume to use the nationwide swing in the model. However there are good reasons to believe that Labor will do poorly in Tasmania.
Firstly, Tasmania has the worst unemployment in the country at 7.3%, which has recently risen from a range of 5-6% in the past year. We can presume the economy is not doing too well in Tasmania compared to the rest of the country.
Secondly, Tasmania has a long term state Labor Government which is unpopular at a mid to low 20’s range in local polling.
Thirdly, Labor received a 4% TPP swing state-wide at the last federal election in Tasmania. Considering that the nation as a whole swung to the Coalition by 2.8% at the last federal election there will have to be some rebound swing to the Coalition at the next federal election in Tasmania.
However, until we get some hard public polling data about Tasmania we have nothing to put into the model to reflect the state. The seat model needs hard data, not hunches. Currently, the model is giving 3 seats to Labor 1 to the Coalition and 1 to Wilkie. If you are convinced that Labor is facing a wipe-out in Tasmania, just take three seats off Labor and add the same number to the Coalition.
The Final Product
With the explanation out of the way, here is the seat projection graphic:
As you can see we have a schematic House of Representatives Chamber, because it is the done thing….. We also have a simple table indicating seat changes by state.
As per the poll average we will be updating the seat projection on the Australia page of this website every time new eligible polling becomes available. Make sure to keep an eye out for it. The Australia page can be accessed by clicking the Australia tab above.