Using Google Search Data To Predict The Australian Federal Election Result

Using Google Search Data To Predict The Australian Federal Election Result

Quick Points

  • Data was gathered from Google’s Keyword Planner Tool. (Not Google Trends)
  • Data shows the number of times each candidate’s name was searched for in Google leading up to the election.
  • Historical data predicted 95% of the 2019 popular vote.
  • This data is released monthly in arrears meaning April is the most accurate pre-election data available. 
  • Data is based on relevant state search volumes for each electorate.
  • High Interest Levels in Independents & other candidates.
  • Data suggests a Hung Parliament likely.

 

With opinion polls coming under heavy criticism for their inaccurate prediction of a sweeping Labor victory in the 2019 federal election, I thought I’d test whether Google search data could be a viable alternative for predicting the outcome of the 2022 election. 

Checking the search volume for each federal candidate’s name via Google’s keyword planner tool for the April 2019 period, revealed a strong correlation between the number of searches for a candidate and the eventual winner of the popular vote. In simpler terms; whoever’s name was searched for the most in Google generally went on to receive the most votes on election day.

On the raw numbers, Google search volume data predicted the popular vote winners of 137 out of 151 seats – however when combined with qualitative research to remove disambiguation of candidates who share the same name as other famous individuals and outlier results created by headline generating antics of controversial candidates (checked via Google News results for April 2019) – the data predicted 143 out of 151 seats – equating to a 95% level of accuracy – This could likely be improved upon by someone with a much stronger understanding of the political landscape than myself.

Whilst this level of accuracy isn’t high enough to predict the overall winner of the election – given the complexities involved with the two party preferred system – it can be used to gain valuable insights into the 2022 election race.

What the Data Says About The 2022 Election

Applying the same process to April 2022 search data shows a surge of interest around independent candidates with 15 recording a higher number of searches than their Coalition and Labor opponents – in many cases, by a significant margin.

In 2019, this method of using Google search data correctly predicted all 5 winning independent and other candidates.

If proven to be correct, the data suggests Australia could be heading for a Hung Parliament with the breakdown of seats based solely on the primary vote as follows: (doesn’t take into account preferences)

Below are all the predictions of who will receive the highest primary vote based on Google’s search volume data coupled with my own basic qualitative research – starting with the Key Seats as identified by the ABC, followed by a state by state break down of the results (best viewed on a desktop device).

If you’d like a raw copy of the data, feel free to email me at jared@jaredbennett.com.au

Key Seats Predictions

  • Red = Projected primary vote higher than incumbent
  • “April 22 SV” = Amount of Google searches in April 2022
  • “3 Month Change” = Change in search volume compared to previous 3 months
  • Where two candidates received identical reported searches, the 3 month change in volume has been used as a “tie-breaker”.
  • Scroll horizontally to view data for all candidates

Sam Birrell & Steve Brooks: Combined search volume equalled 720

Predictions By State

Candidates with qualitative research applied highlighted in red – explanation notes below data tables.

New South Wales

Victoria

Cameron Smith: name disambiguation
Belle Gibson: name disambiguation

Queensland

Milton Dick: Data withheld by Google
Nathan Buckley: Name disambiguation

South Australia

Mark Butler: Data withheld by Google – unsure why
Stephen Grant: Possible Name disambiguation

Western Australia

Tasmania

ACT

Northern Territory

2019 Search Volume Data

No Comments

Post A Comment