Tuesday, April 2, 2013

Weekly aggregation

This week we have a new Morgan poll and a new Essential poll. Both polls see movement away from the government.

We start our analysis with the localised regression of all the polls (next chart). I would not pay too much attention to the end-point in this chart. In my opinion, LOESS is overly responive to the end-points in a series, and I suspect that is what we are seeing here.

Notwithstanding the end-point problem, this chart suggests that there has been two periods of decline since January this year. The first period followed the announcement of the election date and the second period can be associated with the recent challenge (or should I say the non-challenge) for the Labor leadership.

A similar pattern can be seen in the next chart from the Bayesian aggregation, with a decline in the first fortnight of February and a decline in the last fortnight of March. I suspect the end point in this chart is closer to the current state of play.

My national model is now predicting that Labor would win less than 50 seats, and the Coalition around 100 seats, were an election held at the moment.

Caveat: I am still nervous about using the Morgan multi-mode poll in my analysis. While I have been assured that the methodology will be consistent, I note that the sample size has been reducing with each poll. The sample size series for the five multi-mode polls to date has been: 9101 - 4627 - 3982 - 3494 - 2953.

No comments:

Post a Comment