This page contains some fun little bits and pieces that I encountered while playing with the various plots. All images on this page are static screenshots.
Early political geography
(Link.) I'm not much of a student of history, and knew little of the political landscape across Australia in 1901. My hope was that, when I threw all of the electoral boundaries and data together, something would show itself to me in the results. And it did! Here is a map of seat winners in 1901:
The Free Traders' stronghold was in Sydney; the Protectionists were dominant in Victoria, parts of regional NSW, and south-east Queensland. This isn't news to anyone who's studied this part of Australia's history, but it was fun to learn it from the map like this.
Swings in 1931
(Link.) In the first election after the start of the Great Depression, the governing Labor Party was very badly beaten, with enormous swings against it. In New South Wales, Jack Lang's rebel party won much much support than the federal party. Even allowing for this, and treating Lang Labor as Labor for two-party-preferred purposes (as I do in my estimates), the TPP swing against Labor in NSW was about 16 percentage points, similar to the national average. The one state to go against this trend – by a lot – was Queensland, where the six seats for which a TPP figure is known or can be estimated all swung to Labor.
Did the Depression not hit Queensland? Did the Labor campaign manage to blame it on the state government? I have no idea.
(Link.) Staying with the 1931 election, let's have a look at the results in the seat of Cook (Sydney metro). Cook was one of the safest seats (maybe the safest seat? I haven't searched systematically) in Australian history, with Labor having won 84% of the vote against a Nationalist candidate in 1929.
In 1931, it was effectively a contest between Lang Labor and federal Labor; the Lang candidate won 47.9% of the primary vote, with 31.0% for federal Labor, 19.8% for UAP, and 1.3% for the Communists.
After the Communist preferences were distributed, it was 48.7%, 31.2%, 20.1%. In those days, party names weren't written on the ballot paper, so I would have guessed that preference flows were generally weaker than they are today. In fact it seems that the parties' campaigns and the voters generally made sure they knew the names of the candidates and the recommended preferencing order. In Cook, an astonishing 97.9% of the UAP preferences went to the federal Labor candidate, enough to give him the seat, with 50.9% of the two-candidate-preferred vote! (By contrast, the strongest Greens preference flow to Labor in 2013 was 91.1% in Grayndler.)
Preference flows over time
(Link.) The introduction of preferential voting in 1919 was at first mostly useful for the conservative side of politics, since voters were now able to support both parties in the Coalition (it also often happened that a seat had multiple candidates from one party). After a brief period of relatively even preference flows overall in the 1930's, in part due to the two Labor parties co-existing, preferences generally resumed their pattern of going to Coalition parties, driven either by intra-Coalition preferencing or the DLP.
The formation of the Democrats, and their immediate assumption as the main third party in 1977, marks a very sharp jump in the overall preference flow time series. Democrat voters were perhaps becoming more left-wing relative to the major parties before the Democrat vote collapsed in 2004, although the overall rate of preference flow from the Democrats to Labor in 2001 was about the same as in 1990.
It's interesting to me that the preference flow from the Greens to Labor has generally become stronger over their short history.
Correlations over time: TPP
Electoral results from one election to the next tend to be well-correlated – swings aren't perfectly uniform across different seats, but it's close enough that safe Coalition seats remain with the Coalition, and safe Labor seats remain with Labor. The correlation between seats' two-party-preferred vote in 2010 and 2013 was 0.97: (link)
How far back in time can we go and see similar results to today? A word of caution: correlations of the TPP vote across seats in different elections are affected by redistributions, which may dilute the correlations (if the redistributions are random and the character of some electorates change drastically) or make the results seem artificially continuous (if redistributions follow particular parts of the community as suburbs grow). I don't know if the latter is important; the way to do this study properly would be to use the division boundaries (which I have) along with polling place or subdivision results (which I don't have in digital form prior to 1993).
Be that as it may, here is a graph showing the correlation in the TPP vote between the election on the x-axis and 2013, so that the graph reaches a perfect correlation of 1.0 in 2013: (link)
I don't see any grand truths in the above curve, merely a measure of the persistence of Australian electoral geography, I'd guess driven in part by class boundaries (at least spatially: richer and poorer suburbs generally staying that way) and in part by the Country/National Party always being strongest in rural areas. The correlation between the TPP in 2013 and in 1949, when parliament expanded to 74 to 121 seats, is 0.65: (link)
Things get more interesting if we calculate the correlations by state: (link)
(Colours are the states' usual sporting colours.) Tasmania's electoral landscape has totally flipped since 1966! Of course, it only has five seats, so if such a thing were to happen in any state, it would be in Tasmania. By contrast, Victoria's political evolution has been the most boring of all the states, with little change in the distribution of the TPP vote since the introduction of preferential voting in 1919. (The sample size for that correlation is only 12 seats, and I haven't done any proper statistical testing to account for multiple comparisons, but it's at least a surprising coincidence.) Here's a Victorian scatter plot, ρ = 0.86: (link)
And here's Tasmania, ρ = -0.97: (link)
Correlations over time: Turnout
Sometimes a peculiarity of the data is responsible for an apparently large change in overall statistics. Here is the correlation of the turnout across seats with the turnout in 2013: (link)
Much (though surprisingly, not all) of the big jump in 2001 of the correlation with 2013 is caused by the division of the Northern Territory into Solomon and Lingiari. Lingiari consistently has a much lower turnout than any other federal electorate, and its inclusion in the correlations starting in 2001 increases the spread of the data and hence the correlation. Excluding Lingiari from the 2001-2013 comparison reduces the correlation from 0.92 to 0.82.