Getting into the swing of things

After my attempt at a general formula to approximate seat-swing from marginality and national swing crashed and burnt, I started looking at the possibility of formulating a swing index. This would basically be a value for each seat. For example, a seat that generally experiences a smaller-than-average swing might have an index of 0.5, meaning in a national swing of, say, 7 percentage points that seat could be expected to have a swing of around 0.5 x 7 = 3.5 percentage points.

For this purpose, I looked at the last three federal elections (2010, 2007 and 2004) and calculated their swings from the post election pendulums (e.g. the 2004 swing is the difference between the Two-Party-prefered result in 2001 and 2004.) Note that the post-election pendulum (aka election results pendulum) is different to the pre-election pendulum for the following year, since corrections are made following by-elections, redistribution of seat boundaries etc.

The pendulums I used are as follows: 2010, 2007, 2004 and 2001. Some seats had to be excluded due to the nature of the calculations. Any seat that had elected an independent or minor party during those four years would confuse the data from the simple 2PP ALP v Coalition analysis I was conducting. That was goodbye to Calare (Independent 2001-2004), Calwell (Independent 2001), Cunningham (Greens 2004), Denison (Independent 2010), Kennedy (Independent 2001-2010), Lyne (Independent 2010), Melbourne (Greens 2010) and New England (Independent 2004-2010). Three more seats were excluded because the second party was a minor party (which would also mess with the 2PP data): Batman (ALP v Greens 2010), Grayndler (ALP v Greens 2010) and O'Connor 2010 (renegade Nationals v Liberal). Had Melbourne still been in play, it too would have been excluded after 2007's ALP v Greens. Finally, any seats that had only contested the 2010 election could not be assigned swings, so Durack, McMahon and Wright were also excluded. Other seats with short enough histories to not date back to 2001 (Bonner, founded in 2004, Flynn founded in 2007 and Gorton founded in 2004) have been included but lack some of the early data.

So, to jump right into things, here is a graph of each (usable) seat's swings from '04 to '10. (Negative swings are swings away from the incumbent.)

Now to appreciate how closely each seat reflects the national trend, we need to normalise the direction of the swings so that they are all swings to or from the same party. Or, if this were a science-fiction pollitical arena, it would be time to (selectively) reverse the pollarity. (See what I did there? POLLarity? Oh, never mind...)

Here are the same seats with there swings towards the party (or parties in tha case of 2004) that would go on to form government (Coalition 2004, ALP 2007 and 2010).

Notice that the green and blue swings are generally positive, while those from 2010 are generally negarive. This is a good sign, since the public vote swung towards the future government in '04 (Howard) and '07 (Rudd), but away from Gillard's Labor in 2010.

There is still work to do though. Some seats demonstrate large swings one year and small ones another. This is starting to look like random, but lets double check to be sure, by eliminating the variable of national swing magnitude. All swings might be expected to be exagerated in volatile years with large national swings, but quite small in the more stable campaigns.

Here is one last graph, reprisenting the swings as a proportion of the national swing, according to the formula (Seat Swing)/(National Swing) x 100%:

This was also the point at which I realised that a single tall bar graph makes more sense than chopping up a column graph.

Now almost all of the graph lies in the positive - which is unsurprising, given that if most of the seat level swings went against the national swing there would be some pretty odd outliers. That said, Wakefield's -1000% swing in 2004 is pretty nuts.

Other than that, there does not seem to be any indication of such things as big-swing and small-swing seats. Perhaps a longer view of the history will prove otherwise, ignoring some recent outlier elections. Then again, perhaps such paterns belong to a bygone age in that case.

Either way, we still do not have a reliable means of estimating the size of a swing in a given seat. I'll give it all a bit more thought, but I cannot promise any progress on this front.