Over the weekend, Newspoll via The Oz brought us their quarterly demographic breakdowns which you can see in their entirety over here – although beware, that’s an 11MB pdf file. Those Newspoll results cover the period from October through to December, giving us a mega sample on the vote estimates of 8059. We can combine those results with the Nielsen polls over the same period – 3 polls with a total sample of 4000 – to give us a rather large pooled sample for the October-December period of 12059 respondents.
On the sidebar to the right, the new voting intention estimates for various demographics have been calculated using this new pooled Newspoll/Nielsen data – which probably makes it the most accurate voting intention data by demographic cross-tabs in the country. It’s certainly worth having a bit of a squiz at.
The other thing we do with the quarterly data is run a monte carlo simulation to give us a reading on what a new Parliament would most likely have looked like if an election were held over the Oct-Dec period and where the result was consistent with the polling.
The simulation works by using the state level breakdowns of the two party preferred voting estimates. We first calculate what the swing currently is in each state – so using Victoria as an example, the combined Newspoll/Nielsen data currently has Victoria sitting on an ALP two party preferred vote estimate of 59.4%, a 5.1% swing towards them since the 2007 election.
For every seat, we create a swing for that seat that is a probability distribution – where the mean is the state level swing (for example, 5.1 in Victoria) and give it a standard deviation that is a combination of both the margin of error of the pooled state sub samples and the historical standard deviation of federal election results in that state.
We then draw a random number out of that probability distribution and add it to the two party preferred result in that seat at the last election. When we do that once for all 150 seats, it gives us 1 simulated election where we add up the number of seats that are now sitting in the government’s column.
We then do that another 20,000 times, tally up the numbers, turn them into a probability distribution and we have our simulated election result.
If an election were held over the October-December period and the results were consistent with the polling, the most likely outcome would have been a new Parliament where the ALP had 107 seats, the Coalition had 40 seats and the Independents held 3 seats.
In those 20,000 simulated elections, the worst result for the government was 92 seats while the best was 121 seats. The median and mode results were 107 seats while the mean result was 107.2 seats. The distributions of the result come in like this.
The first chart is just a histogram of the results, while the second chart gives you the probability (left hand axis) of the ALP getting at least a given number of seats (bottom axis) in the new Parliament were an election held last quarter.