Thursday, May 19, 2022

Modelling the 2022 Election - Closer than we first thought?

Recently in a conversation with Ethan from Armarium Interreta, he made the observation that the polls in the final week of a campaign were (on average) more accurate than the polls in the final two weeks of the campaign. This is critical because we are seeing a tightening in the most recent polls. [Note: things might not stay this way as more polls come in, but this is how it looks now].

Based on this observation, I have modified my model to look at the mean polls in the last 7 days of the campaign, and compared those with the average polls over the past seven days. This changes the regression I am using for calculating an election outcome based on the mean late polls. Nonetheless, we still have the phenomenon that the opinion polls are typically less accurate when they have Labor well ahead.

From an opinion poll average of 47.4 per cent two-party preferred (2pp) for the Coalition, I estimate the Coalition's 2pp vote will be between 46.3 per cent and 51.8 per cent (94 per cent HDI) , with a mean of 49.1 per cent. Labor will be between 48.2 and 53.6 per cent 2pp., with a mean of 50.9 per cent. This might be a closer election than many of us first thought.

Projecting this to seats won, I get:

Caveats: I am using a regression to predict the number of other party seats. However, this relationship is not strong, and the election of non-major party candidates is historically idiosyncratic. As a result, my intuition is that in this 2022 election, the model underestimates the cross-bench by up to 3 or 4 seats, and over-estimates the Coalition by the same amount.

The most recent tightening in the polls, and only focusing on polls in the past week gives me the following probabilities for an election outcome.

If you want to compare my work with other modellers, you should check out:

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 couple of days, and has not been rigorously tested nor tidied up. 

Finally, there are no guarantees with this model. So, don't blame me if you put bets based on this model and you lose your money.


  1. A few problems. IPSOS has a poll out tomorrow night and Newspoll on friday night so it is problematic.
    the other thing is we do not know if any are rogue polls until it is over!

  2. Fair points - I will do a further update Saturday morning.