With all the pollsters in, we can finally run our final election day mega-simulation – and boy is it a close one! Not only does this simulation use our usual quasi-dependency framework (where we don’t treat seats as independent events), but we’ve also anchored seats together within each state on the basis of capital city seats being linked and non-capital city seats being linked, as well as Western Sydney seats being linked. In the simulation, seat groups effectively “move together” as we see in real life, as well as accounting for the “Western Sydney” issue that the polling heavily suggests is happening.
The pollsters used are Newspoll, Nielsen and Morgan phone – giving us a pooled sample of 6419 – where we use their aggregated state breakdowns to derive state based swings as the key mechanism to calculate total seats (with the fancy shmancy quasi-dependency framework buolt into the swings)
The monte carlo simulation stabilised at around 130,000 iterations, with the ALP seat results coming in like this (click to expand):
The most likely result is the ALP winning 74 seats, the Coalition winning 72 seats, the Independents getting 3 and the Greens winning 1 (where I substituted betting market data for the seat of Melbourne in the surprising absence of any seat polling).
However, switching to the more useful cumulative probability view of the results, we get (click to expand):
Labor is coming in with a 50.2% probability of winning at least 75 seats. That jumps out to 59.6% for winning at least 74 seats and to 68.3% for winning at least 73 seats (which is the close to guaranteed ALP minority government result).
However, there is only a 24% implied probability of the Coalition winning at least 75 seats , jumping up to 32% for them winning at least 74 seats – where 74 Coalition seats would be nearly guaranteed to give the Coalition minority government.
If we run two simulations under two seperate assumptions – 1.where the Greens win Melbourne and there are 3 independents and 2. where the ALP wins Melbourne and there are 3 independents, the implied probability of a hung parliament is 34% with a Melbourne Green and 20% with only 3 members on the cross-benches.
The combined phone polls show it to be a close one, slightly favouring the ALP.
The big issue in the simulation was actually the Newspoll result for Qld. If the actual Qld result comes it at around 45% for the ALP rather than the 42% that Newspoll estimated it as (pulling down the Qld estimates in the simulation), those simulated implied probabilities start shifting to Labor pretty quickly because of the spread of the margins in Qld and the number of seats involved. The 45% TPP level for the ALP is a bit of a threshold in Qld (assuming the rest of the polling is approximately correct). If they get above it – it increases the likelihood of an ALP win substantially, below it and it increases the likelihood of an ALP loss just as quickly.
There’s a similar key threshold point for the ALP in NSW – sitting about the 51% mark. Above that, the chances of an ALP win start increasing dramatically, below that they start declining dramatically.
The final polling has it tight – very tight. With the relatively high undecideds some were pollsters were picking up on top – there’ll be a lot of nervous folks around today 😛
And don’t forget (disgraceful self promotion) – Possum needs a job!