Newspoll Tuesday – Graphdump edition and new Pollytrack methodology
Posted by Possum Comitatus on August 26, 2008
Another Tuesday, another Newspoll of psephy goodness via The Oz, this time showing the ALP down 2 on the primary and the Coalition steady for a 45/38 split -washing into the two party preferred for an ALP lead of 56/44, down from the 57/43 a fortnight ago.
Today we unveil a new Pollytrack methodology to plug these numbers into.
Previously we were using just phone polls, but over the last few weeks I’ve been experimenting around with different polling aggregations and comparing them to election results to find a more accurate way to do it (and a big thanks goes out to a few folk that gave me access to unpublished polling data to help with the calibration of this thing).
Instead of just using phone polls, we are going to do the obvious and use every piece of polling data we can find to give us rolling super samples calculated weekly. Each Pollytrack calculation is made up of a rolling all pollster average weighted by sample size, that operates over a 5 week maximum window, and where each pollster has only their latest poll contributing to the super sample at any given time.
As a pollster releases a new poll, their old poll result drops out of the rolling window, and gets replaced by their new polling data.
The 5 week window works best as ACNielsen (a monthly pollster) occasionally releases polls 5 weeks apart rather than the usual 4, simply because of timing issues. Using a 5 week window allows us to always have an ACN poll in the Pollytrack mix.
The current Pollytrack mix is made up of Newspoll, ACNielsen, Morgan face-to-face and Essential Research (EMC). Morgan phone polls will also be added when they pop up, but the last phone poll from Morgan was taken over 5 weeks ago, so it drops out of our Pollytrack window.
The beauty of this is that it gives us super samples. Our current sample is 6039, but it has been as high as 7450 in early July, while the lowest it’s been all year was 4158 in early May. That gives us minimum MoEs on this baby of between 1.1 and 1.5%.
On the primary and TPP Pollytrack charts, the minimum MoEs have also been added to the chart itself, giving us a much better handle on the current state of play that includes pure sampling error uncertainty.
Currently, Pollytrack has the primary vote on 45.7/37.2 to the ALP, with the two party preferred running at 56.6/43.4 to Labor. The minimum MoE for this week’s Pollytrack is plus or minus 1.3%.
That now makes Pollytrack by far and away the most accurate tracking poll in the country bar none. Nothing else comes within a bull’s roar of this baby.
(As always, just click the charts to blow them up)
The estimated seat changes are calculated using an assumption of a uniform swing on the national pendulum.
Essential Media have also been added to our Loess regression charts, which now give us:
Currently our local regression series has the ALP leading 45.6/36.4 on the primary vote for a TPP of 57.0/43.0.
Next up, we’ll take a squiz at the Newspoll qualitative metrics.
Finally a squiz at how all the pollsters have been traveling, including the new EMC polls we’ve added to the mix. There were a few numbers out of place in this series previously for the Morgan polls – I had a bit of a brainfart, but that’s all been fixed up.
Things might get a little bit all over the shop round these parts for the next few weeks as I prepare for the site move, so I might run a day or so late on stuff.
But good psephing to all, I’ve got to scurry off and do the latest US Intrade data that should have been completed yesterday.