Simon Jackman at The Bullring has done as I did in an idle moment a few weeks ago, deriving various measures of spread from an election’s worth of swing results at national and state level. If Jackman’s interpretation is correct, a state-level poll is little better than a national one as a pointer to a given seat result:
We see that there is less variation in swing within states than there is overall. But not that much less. The measures of spead of the swings (range, standard deviations, the mean absolute deviation around the mean) for each state are still quite large relative to corresponding national figure … There is still an awful lot of variability in swings out there, and I’d be reluctant to start applying uniform swing models within states.
However, he does add just one caveat to all of this:
It could well be that when average swings are large (or dare I say massive), there is greater uniformity or even less uniformity than when average swings are relatively small. I haven’t looked at data from previous election to know the answer to that, but it be helpful to know the answer to that.
Which is easily done if you have a spreadsheet full of swing figures, like I do. These three tables replicate Jackman’s for the 1996, 1998 and 2001 elections, two of which saw heavy traffic from one party to the other.
And what do you know. The last time there was a big swing and a change of government, the gap between measures of spread at state and national level was significantly higher than in 2004. However, this was not true of the 1998 election, which saw a substantial swing to Labor but no change of government. That might be due to the effect of One Nation in polarising the cities and the regions, most evidently in Queensland. The even messier picture from 2001 provides support for Jackman’s suggestion that a relative lack of state-level uniformity might be a phenomenon of status quo elections.
UPDATE: Geoff Lambert, who knows way more about these things than I do, offers a well-made point about the leptokurticity (here, use my hankie) of swing distributions in comments.