I’ve knocked together a linear regression model to fit the results of the 2008 Western Australian state election in the vague hope it might shed some insight on where Labor and Liberal overperformed and underperformed.
I’ve limited this to metropolitan electorates, because the relationship between voting and the most potent explanatory variable — income — breaks down beyond the city limits. I’ve also excluded electorates where independents reached the final count in the preference distribution (like Churchlands, Alfred Cove and Kwinana).
On my first run I simply looked at income, and two not unexpected factors stood out: parties overperformed where they had a sitting member defending their seat, and Labor overperformed in the inner city and Fremantle. So I added two extra variables to the model: the Greens primary vote, to serve as a proxy for inner city-ness, and another variable that assigned a value of -1 to seats with a Liberal incumbent, 1 to seats with a Labor incumbent, and 0 to seats vacant seats. From these variables, Labor’s two-party vote can be modelled as 0.681 -0.419a +0.781b +0.056c, where “a” is a measure of median family income, “b” is the Greens primary vote and “c” is the incumbency variable. This explains 86% of the variation in the two-party results.
The data created — see the blog for the full table — ranks seats according to how Labor performed relative to the expectation of the model, from highest to lowest. The columns show median family income from the 2011 census, actual Labor two-party preferred, Labor two-party preferred according to the model and the candidate situation at this election. “Sophomore” refers to situations where a party has gone from not having a sitting member at the 2008 election to having one in 2013, hence excludes situations where the sitting member has changed as a result of byelections. “Vacancy” refers to the opposite situation, meaning a retiring sitting member in four cases and a mid-term byelection loss in the fifth (Fremantle).
As a basis for analysis this is far from foolproof, as in every case the result would have been influenced by multiple factors outside the model. Nonetheless, it hopefully offers a few hints where parties do or don’t have room to improve.
Labor’s six “best” results were achieved in seats where the Liberals did very little campaigning as they recognised them as safe for Labor, so it might then be thought notable that they are putting considerable effort into two of them (West Swan and Perth) this time. There’s no corresponding pattern at the other end of the table, although Labor did abandon Kingsley very early in the 2008 campaign after recognising it as a lost cause.
The two seats which would have intuitively been expected to have been near the bottom of the table, Mount Lawley and especially Morley, are in fact not far from the middle — though I hesitate to read too much into that.