Thursday, May 16, 2019

A herd of new polls

The election was called on the 11th April, and every poll since has been in a two-point range: 48 or 49 per cent for the Coalition; and 51 or 52 per cent for the Coalition. Last night we had an Essential poll that split the difference 48.5 to 51.5; and an Ipsos (49/51).

Suspending disbelief for a moment (and ignoring the image of Jesus on the toast that just popped out of the toaster) we can pop these numbers into the Bayesian aggregator (unfortunately, designed for numbers that come from statistical collections and not some other processes) and we can estimate that the Coalition has 48.3 per cent of the national two-party preferred (TPP) vote share. Labor has 51.7 per cent.

For a serious discussion of pollster herding and the highly implausible under-dispersion in the Australian opinion polls immediately prior to the Australian 2019 election, see my earlier post. It is also worth having a look at 538, Peter Ellis and Kevin Bonham.

In making this estimate, we have applied the following adjustments to each pollster, to account for house biases.

We can compare the Bayesian aggregation with an ensemble of moving averages; all of which have the Coalition out-performing the Bayesian aggregation.

Turning to the primary votes, we can see:

1 comment:

  1. Mark,
    I am looking forward to hearing what you sat AFTER the election.

    It is strange