The 2014 Victorian Election is now one hundred days away (29 November). The outcome of this election is being keenly watched by many as the current seat balance between the Liberal/National Coalition and Labor hangs on a knife edge. On the back of the success of our model for the 2013 Federal Election, we have turned our analytical skills to the upcoming Victorian Election.
Background to the upcoming election
At present, the Liberal/National Coalition government holds 44 seats, whilst the Australian Labor Party (ALP) holds 43 seats. With 88 seats in the lower house of the Victorian Parliament, 45 are needed to govern outright. The Coalition won 45 seats at the last election but now only holds 44 as a result of the resignation of Geoff Shaw from the Liberal party.
Federal Election 2013 predictions
For last year’s Federal Election, we analysed more than 300 polls from a number of polling houses to track the evolution in public sentiment over time. In our election eve publication, we predicted that the most likely outcome for Labor was 54 seats and that the most likely outcome for the Coalition was 95 seats. The actual outcome was 55 seats for Labor and 90 for the Coalition. The discrepancy is made up in part by a significantly larger than expected number of minor parties and independents. Four months previously, when pollsters and pundits were talking about a much more extreme result, our original prediction had been that the most likely outcome was 56 seats for Labor (they ended up with 55).
In terms of the two party preferred vote (2PP), we predicted that the Coalition would get 54.7% (95% confidence interval: 53.3% to 56.1%), and that the ALP would get 45.3% (95% confidence interval: 43.9% to 46.7%). This compared with the actual outcome of 53.49% for the Coalition and 46.51% for the ALP, placing the actual outcome well within our model’s uncertainty – our model performed well.
Victorian Election 2014 approach
We have followed a similar approach to that used previously. However, in our federal election analysis we incorporated odds from betting market data as an additional source of information beyond published opinion polls. We do not have a record of historic betting market data for the Victorian election, and so are unable to do this here.
In a similar fashion to our federal election analysis, we have used previous election results to place a constraint on the range of possible outcomes on election day.
We have used the results from the last Victorian state election but updated these as a result of the by-elections that have occurred in the interim period.
Our model currently predicts an 80% chance of an ALP win, an 18% chance of a Coalition win, and a 2% chance of a hung parliament. This is based on a predicted 2PP outcome on the election day of 51.6% for the ALP (95% CI: +/- 5.6%) and 48.4% for the Coalition (95% CI: +/- 5.6%).
The results are unfortunately imprecise, but this reflects a combination of: infrequent polls (eg every two months from Newspoll); not many polls in the field from other polling houses; and relatively small sample sizes on many polls.
We expect all of these factors to improve as the election draws nearer, and so we anticipate a significant improvement in our model’s precision closer to that date.
As an aside: if data is plentiful then statistics may not be needed. However, the use of advanced statistical method is useful precisely in situations such as this one – where the polling data is infrequent, then careful use of more advanced techniques can best extract the ‘hidden meaning’ from the data that exists.