Pollytrack – November 25

With Newspoll and Essential Report having been released, Pollytrack comes in this week with the ALP Primary vote down 0.1 to 45.3, the Coalition primary vote down 0.5 to 37.7 washing into a two party preferred of 56.2/43.8 – which is an increase of 0.3 to the ALP over past 7 days. i.e. nothing much happened at all.

The usual charts and graphics can be seen over at the dedicated Pollytrack page.

Essential Report has now had two weeks in a row where it’s operated at approximately the same vote estimate levels as the phone polls rather than having a slight lean toward the ALP as it had historically done – that’s very interesting.

That Essential Report structural shift is responsible for our low and high sensitivity LOESS regression lines behaving strangely over the last two weeks, especially the primary vote trend lines.

As time goes on, most of that volatile behaviour will wash out as the algorithm recognises that it was noise and adjusts the trend line accordingly as new polling data comes into the system. Our low sensitivity line is suggesting about 2 points of movement toward the Coalition and a point away from the ALP over the last month on the primary vote estimates – however, about 3 quarters of that movement is as a result of Essential Report’s new vote estimate level more closely reflecting the phone polls.

Our LOESS regressions don’t particularly care if some polls lean more strongly towards one party than another, as their role is to determine longer term movements in political support toward or away from a party – the actual vote estimate level is not really that important at any given time, it’s the change in the level over time that these regressions help us to determine.

But these regressions have hissyfits when a poll changes the magnitude of it’s lean, at least during the transition of that change. If polls lean consistently toward a particular party relative to the other polls , then as a party’s support moves up or down, all of the polls generally move up or down together, keeping the size of their relative lean relatively consistent.

But when a poll changes the size of its lean, meaning that a poll shows a change in the level of political support by a party that isn’t reflected by the other polls, especially a weekly pollster like Essential that has lots of observations – then we get this sort of temporary volatility occurring until the new relative lean of Essential (if indeed it is a new lean) has had time to build up in the system values.

On something entirely different – a question.

Parts of this site get regularly updated in the background – sometimes up to three or four times a week – as polling data comes in or as I update various analysis. Would readers be interested in having a little text display in the sidebar that shows which bits have been updated as they happen?