Monday, August 27, 2018

The first poll of the Morrison period

The first Newspoll of the Morrison Government was a shocker for the Coalition. According to that poll, if an election were held last weekend, Labor would have won 56 per cent of the two party preferred (TPP) vote. If we apply the cube rule, as a rough approximation, this would see Labor winning something like 101 seats of the 150 seats in the House of Representatives: a thumping majority. Winning 56 per cent or more of the TPP vote is a rare occurrence in Australian politics, which last occurred in 1966 and before that in 1931. The majority vote share in the 1975 election was 55.7 per cent.

The poll gave me a chance to test drive my updated aggregation models, following the leadership change and associated discontinuity. Obviously, with only one poll following the change of Prime Minister, this is not a meaningful aggregation, and I would want to see a few more polls before coming to a judgement on the impact of the change of leader.

One thing to note, I have further modified the TPP model in acknowledgement that the priors for the hidden voting intention model are weakly informative. To ensure they are not influencing the model excessively, I have set the value of these priors to reflect the polls immediately following the date of these priors. I have updated the code at the link above.

Let's look at the charts that come from the model.

Some of the moving averages are starting to turn, but will need more polls for this to settle. Unlike the Bayesian model, the moving averages cannot incorporate discontinuities.

Turning to the primary voting intention model (with a discontinuity), we can see.

And the implied TPP from the primary vote intention model follows.

Again, I would urge caution into these initial charts. Once we have a month or so of polling data, we will be better placed to assess the impact of the change of leadership on voting intention.

No comments:

Post a Comment