With today’s combined two week Newspoll breakdown released over at The Oz, we can aggregate the results with the last two Nielsens and the JWS Godzilla poll to run some simulations. Ordinarily we make a pooled super sample of all the polls,  weighted by sample size, to get aggregate state results from which to crunch the numbers. Yet, with the JWS monster and it’s 28,000 sample, it poses a problem since it would simply dominate the simulation results (not that it would actually matter – but more on that later) because its sample is so large. To get round this issue, we’ll pull the weight of the JWS poll down to 40% – letting the Newspoll and Nielsen take the remaining 60% of the pooled sample weight.

Here, we’re running our quasi-dependency Monte Carlo simulation method which treats individual seat results as neither dependent nor independent events, imitating the real world effect we see (or, at least,  we think we see) happen at elections, where seats “move together” at the state level of aggregation (not that it actually makes much difference at the moment!)

So crunching these numbers – taking about 30,000 simulated elections – this is what the probability distribution of the change in seats from the current Parliament looks like (click to expand these):

simhistaug18

While the more useful cumulative probability distribution comes in like this:

simcumulativeaug18

To read the chart, choose a number at the bottom of the chart – the left hand axis tells you the implied probability of the ALP winning at least that many seats were an election held over the last fortnight and the election result was consistent with the polling results.

As we can see, the ALP has an implied probability of 71.7% of winning at least 75 seats – which is slightly less than the implied probability of an ALP win given by the headline betting markets at the moment. However, the betting markets ask who will be the next government. If we look at the probability of the ALP winning at least 74 seats from the simulation (where Labor would be certain to operate in minority government), we get:

74seatsplus

Here it jumps to 79.7% – which is only a few points higher than the current betting market headline probability.

There’s been some debate lately over whether the polls and the markets are being consistent – the headline probabilities for an ALP by both the polls and the markets certainly suggest the two are holding hands around the same tight probability band, so they are definitely consistent.

Also worth noting is that the most likely result is 77 seats being won by the ALP.

On the issue of a hung Parliament, there are two possibilities – the ALP winning 73 or 74 seats and their being 3 Independents, or the ALP winning 72-74 seats and their being 3 Independents and a Green in the seat of Melbourne. Running a separate dual simulation for these results, the implied probabilities for both scenarios for a hung Parliament become:

hungparliaug18

So a possibility of a hung Parliament comes it at between around 16 and 22%, depending on the result in the seat of Melbourne (which, BTW, has been given to the Greens in the simulation)

The other interesting thing here is if we run the simulation again, but treat the JWS poll at face value and simply weight by sample size as usually happens. The results come in a bit tighter on the tails (as in the implied probabilities of Labor getting around 70 seats or the ALP getting around 86 seats become much less), but the headline results remains 77 seats as the most likely result, and the implied probability of Labor winning at least 75 seats comes in at a fraction higher – 73.9% – while the probability of Labor winning at least 74 seats (guaranteeing minority government) comes in at 84%.

Remember though folks, this simulation is old – in that it contains data up to two weeks ago when Labor was in a worse position than they currently are. Labor are probably a few more seats better off at the moment than these simulations suggest.

(Visited 30 times, 1 visits today)