We saw something in the SA election on Saturday night that you just don’t see very often in Australian elections – a major skew in the swing. Often what happens when a swing is on against a government is that there’s a slight skew, where safe government seats swing very slightly more, on average, than government held marginal seats.
The last Fed election is a good example: if we scatter the pre-election, ALP two party preferred of each government held seat against the swing to the ALP that each seat experienced, we get a big cloud with a very slight but statistically insignificant trend involved:
If we run the exact same chart for what happened on Saturday in SA, where we use the Liberal Party TPP in government held seats going into the election and scatter it against the swing to the Liberal party in each, the result is pretty amazing:
Not only is the trend statistically significant, but it’s very, very strong.
The safer the government seat, the much larger was the swing against the government. In fact, for every 3 additional points of two party preferred vote above 50 that a Labor held seat enjoyed going into the election, the swing against the government increased by a full percentage point on average. I can’t remember an election where this relationship happened to this strength.
These sorts of highly non-uniform swings play complete havoc with our simulations when we don’t have any seat by seat, or region by region, or marginal vs whole electorate polling to feed into them, particularly in the probability tails.
Our point estimates look to be about 3 seats out (assuming Rann gets 24 seats), but our probability tails got hammered with the non-uniformity, suggesting that the likelihood of Rann winning at least 24 seats was only 8% – even though the Newspoll it was based on got within the MoE of the election result.
Why this happened can only be partially explained by Bruce Hawker’s resource allocation of Labor campaign spending. Labor focused most of their resources into retaining marginal seats including the political messages tailored to those marginals, letting the really safe seats take the full blow of the swing running against them, knowing that their margins were large enough to withstand such a swing. The point of an election is to win seats, not margins.
Yet, while the safe seats had large swings and the marginals had very small ones, the group of seats between the marginal and safe seats experienced a swing against them that was between the size of the swing suffered by the other two groups of seats.
That seemed to be achieved more by accident than political design – but it’s interesting that those middling seats came in consistent with the regression line (even though they have a larger variance in their results), suggesting that there might have been something else at play apart from the political strategy that gave us this strong relationship across the full spectrum of Labor held seats.