With the Essential Report and Nielsen out this week, the Pollytrend numbers have been updated. The All Pollster trend is coming in with a Labor two party preferred of 55.4%, while the phone pollster trend is running at 53.9%, the first time it has broke below the 54% barrier under the Rudd government.
Something we haven’t looked at for a while is the back stats that form the basis of the pollytrend measures. The all pollster trend is where we take a rolling all pollster pooled poll aggregate weighted by sample size (mouthful!), and we take it every time a poll is released. That provides the raw data that we then run a locally weighted polynomial regression line through to get our all pollster trend.
The Phone Poll trend on the other hand is a rolling all phone pollster pooled poll aggregate weighted partially by sample size, but mostly by the time (in days) since the last polling observation of each phone pollster was released. This is what the raw aggregates with their respective regression trends look like. The all pollster trend first, followed by the phone pollster trend.
The numbers on the bottom axis are the days since the series began, which was way back on the 28th January 2008. Even though the phone pollster observations look much more volatile than the tight all pollster numbers (having many more observations for the all pollster series certainly helps), the same trends emerge. That’s the power of these LOESS regressions we use for this sort of analysis – they can be very, very good at cutting through polling noise regardless of the density of the raw data we feed it.