Federal election prediction

The accuracy of predictions based on polls has been poor in recent years, not just in Australia.

The polls in Australia failed to predict a Liberal-National Coalition victory in the May 2019 election.  One of the betting companies paid out on a Labor victory the day before the election, so sure were they of a Labor victory only to have to also pay out on a Coalition victory after the election.  A subsequent review of the polls by the Association of Marketing and Social Research Organisations found that:

“All the national election polls published during the 2019 election campaign purported to show that Labor had the support of the majority of the Australian voters in terms of the two-party-preferred vote.  The Coalition went on to win the election with 51.5% of the vote compared to Labor with 48.5%, almost the mirror opposite of what the final polls found: all missing the result in the same direction and by a similar margin”.

The main reason, according to the inquiry, was that the samples were unrepresentative and inadequately adjusted.  The polls were likely to have been skewed towards the more politically engaged and better educated voters with this bias not corrected.  As a result, the polls over-represented Labor voters.

Similar errors were made in the British elections of 2015 and 2017 plus the 2016 Brexit vote.  At least in Australia voting is compulsory, unlike in Britain, so the Australian polls should tend to be more accurate.

Given the failure of the Australian polls and betting markets in 2019 foreseechange employed seven other indicators, in addition to the polls and betting markets, for predicting the result of the 2022 Australian federal election held on May 21.

These were:

  1. The relative success of the parties in previous elections, which is in part related to demographic change (track record).  This method predicted a Coalition victory.
  2. A survey question (Resolve Political) asking which party would win.
  3. A survey question (foreseechange) asking the likelihood of a Labor win on a 0 to 100 scale (Wisdom of the Masses).
  4. A survey question (foreseechange) asking whether it was time for a change of federal government, whether it was best to stick with the incumbent, and whether more independent voices were needed in Parliament.
  5. A survey question (foreseechange) asking about which parties (and independents) could be trusted to do the right thing for the country.  Labor and independents were most trusted.
  6. A model of past election results based on the rate of unemployment.
  7. A model based on the expected likelihood of factors such as an economic slowdown and imminent climate change (Wisdom of the Masses).

All indicators predicted a Labor victory, except for the relative success of the parties in previous elections.  In most cases a narrow victory was predicted.  Indicators 4 and 5 also suggested that independents would do well.  The polls and betting markets also indicated a Labor win.

The outcome in the House of Representatives, in which the government is formed, was a Labor win and an increase in the number of independents.  The 2022 and 2019 results were:

Labor 77, up from 68;

LNP Coalition 58, down from 77;

Independents 10, up from 3;

Greens 4, up from 1;

Other 2 (no change).

This case study demonstrates the importance of employing a combination of forecasting methodologies.  Each method has some unique information and is also subject to error.  By averaging the separate predictions, using weights based on expected accuracy, we use all available information and minimise errors.  Errors may be made up of bias (like the polls for the 2019 election) and random components (such as random sampling error) but averaging tends to reduce these errors.

Our models will be re-calibrated on the basis of the actual outcome of this election in addition to previous elections.  This is so that we will be able to make accurate predictions for the next federal election, which is due in three years.