This index represents whether the ALP would have won an election were it held at any time since December 1985.If the ALP score is above zero, it would have one, if it was below zero it would have lost.
The elections are shown on the graph, as is an assumed date of November for a 2007 election.
This index (which successfully accounts for every election result since the 1987 election) suggests that the ALP is better placed to win this election than any party has been for 20 years.
The index was constructed in two parts; a separate index that measured primary vote levels as determined by Newspoll and a second index that measured satisfaction differentials again, as determined by Newspoll
For the primary vote index, a series was constructed that measured how far away the ALP primary vote was from 40.It’s generally held that the ALP needs to get above 40 to win an election without some extraordinary circumstances such as that which prevailed in the 1990 election. So we’ll use 40 as our baseline.
This “above or below 40” series was then normalised by subtracting the mean from each observation of the series and dividing by the standard deviation of the series.
Next I took the Coalition primary vote series and did the same thing, except whereas I used 40 for the ALP, I used 43 for the Coalition. Because the conservative vote isn’t as split as much as the non-conservative vote with minor parties, unless there is a group around like One Nation that attracts a large share of the minor party vote, the Coalition cannot win if its primary vote is below 43%.
Next, a problem came up – the One Nation vote had distorted the historical Coalition primary vote series. To accommodate for this I tested the size of the vote that One Nation took away from the Coalitions primary vote using a number of regression models. They all suggested that One Nation reduced the Coalition primary vote between 5.5-6.2% during the period May 1997 through to just after the 1998 election.
With this info, I then adjusted the “Coalition above or below 43%” series for the period May 1997 through to the end of the 1998 election to accommodate for the One Nation effect on the Coalition. This final series was then normalised.
This gave me two modified series. A normalised ALP primary vote series with a zero value representing a primary vote of 40 and a normalised Coalition primary vote series, adjusted for the One Nation effect, with a value of zero representing a primary vote of 43%.
I then subtracted the Coalition series from the ALP series to give me a series called ALPtoWin1 which represents the relative strength of the ALP compared to the Coalition in terms of their primary vote.
Next up I used satisfaction and dissatisfaction levels as determined by Newspoll for each party going back to December of 1985.I subtracted the dissatisfaction levels from the satisfaction levels of each party to give me a Net Satisfaction rating for each party.
Those net satisfaction levels were both normalised, and the Coalition normalised series was subtracted from the ALP normalised series to give me a series called ALPtoWin2, which represents the relative satisfaction levels of the ALP compared to the Coalition.
I’ve been playing around over the last few weeks with finding the explanatory power of all manner of variables on primary voting movements. In order to keep this post from being a thesis length, basically the relative primary voting performance of the two sides of politics has twice the explanatory power that satisfaction differentials do when it comes to determining elections. As a result, to keep the components of this index realistic, the satisfaction index “ALPtoWin2” was divided by two to give it a realistic weighting.
Finally the simple bit – we just add the two components together to give us the ALP Victory Index.
How well does it work?
Here are the results for the elections since 1987.If the ALP gets a positive index number it should win the election, if it gets a negative number it should lose the election.
July 1987 = 1.54 ALP win
Mar 1990 = 0.76 ALP win
Mar 1993 = 1.51 ALP win
Mar 1996 = -1.60 Coalition win
Oct 1998 = -0.0016 Coalition win (just)
Nov 2001= -0.50 Coalition win
Oct 2004 = -0.60 Coalition win.
The October 1998 result is interesting as the ALP actually won a higher two party preferred vote than the Coalition, even though the Coalition retained government. I was surprised the index worked for that election.
The current value of the index has the ALP on 7.233.That is the highest value the ALP has ever had since the series starts on December 1985, and it is also higher than the absolute value of any Coalition score, even during the 1991 recession when the ALP vote and satisfaction ratings tanked.
Considering the nature of the previous movements of the index, it doesn’t look possible for the Coalition to turn that around in the time available. This too, like Election Prediction Model 1 posted earlier on the blog, points to a Rudd victory.
It was asked if a close-up of the 2001, 2004 and 2007 periods could be done to help gauge the magnitude of the turnaround Howard needs to make.
Here they are: