Pollytrack Saturday

Pollytrack comes in late this week as we were waiting for a special Morgan Phone poll to be published that was taken after the Turnbull leadership change. Due to leadership changes being a rather influential event on the polling metrics, we’ll have to do a few unusual things for this week’s Pollytrack observation.

Firstly, we’ll remove the legacy Morgan Face to Face poll from our Pollytrack window as it was taken before the Coalition leadership change – but we can replace that with the new Morgan Phone Poll. Secondly, as the Essential Report poll contained a two week rolling average of the voteshare and where one week of that average consisted of data gathered before Truffles took the helm, we’ll be using an estimate of just last weeks component of that poll. Unfortunately that reduces out sample size, but it’s better having a smaller sample of 4000 that actually measures reality than a sample of 6500 that only pretends to.

First up the raw data and how it plays out graphically:

The interesting thing here is the Primary vote chart. Notice how the ALP was starting to lose support the week before Truffles took the leadership, yet that loss of primary voteshare didn’t translate into a Coalition vote increase? Yet this week the ALP continued to lose primary voteshare (actually placing them below their 2007 election result for the first time) and this ran parallel to an actual increase in the Coalition primary vote.

Moving on to the swing and sample charts:

You can see this week’s sample size is a little smaller than usual because of our need to junk 900 people from the EMC poll and over a thousand from Morgan Face to Face – but this will start to build back up from Monday when EMC have the Turnbull leadership fully built into their two week rolling window.

If an election were held today with these results the ALP would gain an additional 13 seats on a uniform swing.

Finally we have our All Polls series with a LOESS regression run through it to get a line of best fit. I’ve changes these a little so we’ll now run two regression lines through the data, a low sensitivity line which ignores most poll to poll variation and a higher sensitivity line which takes into account far more of the polling variation.