A quick and dirty Bayesian model of the 2022 Election

UPDATE: unfortunately there were glitches in my code from yesterday. I have now corrected these and updated the charts: here. Yesterday's post (with the errors) follows.

I have spent the past day cobbling together a very rough and untested model for the 2022 election, which is informed by the latest polls as well as the polling and election outcomes since 1983. Like all models, there are a ton of compromises where I either did not have sufficient data, and/or could not think of a sufficiently robust method to make an informed estimate/projection. The model is particularly weak in respect of predicting the number of other (minor) parties that will be elected to Parliament.

The model follows the historical transitions from:

1. the two-party preferred (2pp) polls before an election to the 2pp outcome after an election,
   
2. the 2pp election outcome to the number of seats won by the major parties, and
   
3. the other parties primary vote share to the number of seats won by other parties.
   

The model can be visualised as follows. 

The model projects the two-party preferred vote at the election to be as follow.

In terms of the seats won by each party, the model thinks the most likely outcome is 7 others, 59 Coalition and 86 Labor.

Labor has an 88.3 per cent probability of forming majority government. There is an 8.7 per cent probability of a hung parliament. The Coalition has a 2.9 per cent probability of forming majority government.

My model is more favourable to the Labor party than the other models out there:

• Buckley's and None sees Labor winning 79 seats, the Coalition winning 65 seats, and 7 MPs sitting on the crossbench resulting in a ALP government.
• The Australian Election Forecasts sees Labor on 81 seats, the Coalition on 59 seats, and 11 on the crossbench in the next parliament.
• Ethan and Rebekah at Armarium Interreta see Labor on 81 seats, the Coalition on 60 and 9 others in the next Parliament.

If you want to see how the sausage was made, the Jupyter Notebook is on GitHub. But please note this Notebook was written in a day, and has not been tidied up. There are no guarantees with this model. If you lose money because you placed bets based on this model, that's your problem. 

I will further refine and tidy things up between now and Friday.