This diary relates to Australian politics so I’ve included a quick glossary to begin with.
Glossary
2PP/Two party preferred. Due to Australia’s use of preferential voting it is possible to determine which of the two major parties (or any two parties or individuals) each voter prefers.
AEC: Australian Electoral Commission
Coalition: Generally speaking a long term alliance between the centre-right Liberal and National parties, though technically not in all states.
Major parties: Labor and the Coalition
Lower House: House of Representatives
Upper House: Senate
A Model
In this diary I discuss a model I’ve created to produce a potentially more reliable predictor of seat vulnerability than the commonly used Mackerras pendulum.
Voting in Australia is mandatory and there is a very high compliance rate. The Lower House (House of Representatives) consists of 150 single member districts drawn by a non-partisan body. At the federal level (which is all I’ll be talking about today) Lower House elections use a compulsory preferential system where voters must list all candidates by their order of preference (better known as instant-runoff outside Australia). The Australian Electoral Commission (AEC) publishes a two-party preferred count after the election even if it is ultimately won by an independent or minor party. The Senate uses a rather different system that I won’t go into detail about today but in brief most voters simply nominate a single party and exceptionally few people identify with particular senatorial candidates.
At the most recent federal election in 2010 the governing Labor party lost a net of 11 seats (lost 13, won 2) and required the support of three independents and 1 Greens representative to form a minority government, 76 to 74. The opposition Coalition have the support of 1 Western Australian National representative and 1 Katter’s Australian “Party” representative. All 6 of these non-major party representatives come from seats that are notionally uncompetitive.
The Mackerras pendulum system of predicting election results produces a list (See below) of every Lower House seat in order of the percentage of the two-party preferred vote at the most recent election. By using polling to estimate the size of the swing from one of the major parties to the other a line can be drawn on the list which then estimates how many seats are likely to switch hands. For example at the last election the swing against the governing Labor party was actually 2.58% but the consensus estimate the day before the election could have been 2.5% (depending how much you wanted to believe which poll). So drawing a line on the chart below at 52.5 we can see 11 seats the Labor party held by a margin of less than 2.5.
http://en.wikipedia.org/...
Therefore the Mackerras pendulum predicts that assuming a 2.5% swing away from Labor is accurate the Labor party would lose 11 seats. As is eventuates that is indeed what transpired and Labor lost 11 seats to the Coalition (and also a net of 11 all up). However when you look at the particular seats lost by Labor the Mackerras pendulum does far worse as only 4 of the 11 seats the Mackerras pendulum predicted Labor would lose were actually lost.
The model I’ve created calculates an expected vote a generic Lower House candidate would receive in a given electorate based on the Upper House vote in the same electorate. Unlike a more cursory examination just comparing the primary vote for a candidate to the percentage of voters favouring the candidates party in the Senate this model compares the two party preferred percentage for a candidate to an approximation of a two party preferred percentage in the Senate. A summery table of what the model would predict for the 2010 election follows.
This graph compares the percentage Coalition vote predicted by the model against the actual 2PP vote received at the 2010 election. Data points in the upper left quadrant represent divisions where the Coalition won despite being predicted to lose while the data points in the lower right show the seats Labor was predicted to win but did not in actuality. Points in the Upper right quadrant are Coalition seats (at least nominally, there are 4 points held by non-major parties representatives. Two of which currently support Labor). Those points found in the lower left are Labor seats (again nominally, two of these points represent non-Labor representatives that nonetheless currently support the government).
The tendency for incumbents to perform better than the prediction can be visualised in this graph (which is the same as the previous graph except for differentiating between seats with and without incumbents). Incumbents averaged an over performance for their seat of nearly 3 percent (83% outperforming by at least 1%).
Breifly, an analysis of the seats either the AEC or the model consider marginal;
Coalition Marginals
Aston, Herbert, Hughes, and Macarthur: The Coalition held these seats despite retiring incumbents. They should all improve their performance relative to the senate vote at the next election provided the freshman incumbents stand again.
Bennelong, Bonner, Brisbane, Dawson, Forde, Hasluck, Longman, Macquarie, and Solomon: These seats were won from Labor by the Coalition in 2010. Provided their new representatives stand at the next election they should be less marginal for the Coalition then the AEC rankings would indicate.
Boothby, Dunkley, and Gilmore: Labor were predicted to win these seats by the model. Strong performances from Coalition incumbents were sufficient to hold them, however if their incumbents were to retire these seats would be favoured to switch hands even if there were no nation-wide swing.
Canning, Casey, Dickson, McMillan, Paterson, Stirling, Sturt, and Swan: The AEC considers these seats marginal for the Coalition despite their Coalition incumbents. Should any of these seats become open they would be particularly vulnerable at the next election in the event of a Labor swing.
Wannon: The model considers this seat marginally marginal but freshman Dan Tehan did very well for an open seat and won comfortably in a seat the Coalition has held for over fifty years.
Labor Marginals
Banks, Blair, Brand, Corangamite, Deakin, Dobell, Eden-Monaro, Fremantle, Greenway, Hindmarsh, Lilley, Lingiari, Oxley, Page, Parramatta, Perth, Petrie, Rankin, Reid, and Roberson: Both the AEC and the model consider these seats marginal for Labor. If the country underwent a Coalition swing these seats may switch hands.
Capricornia, Lindsey, and Moreton: The model predicted narrow Coalition wins in these seats however strong performances from the Labor incumbents were sufficient to hold them. If the incumbents were to retire these seats would be favoured to flip even in an environment of no overall swing.
Griffith: Held by popular former Prime Minister Kevin Rudd who was expelled from leadership by his own party three months before the 2010 election. Neither the AEC nor anyone else considers Griffith to be marginal. On the other hand the model considers Griffith to be quite endangered in the hands of anyone other than Rudd as he over performed more than any other Labor candidate.
Kingsford Smith: Both the model and the AEC consider Kingsford Smith to be right on the edge of marginal. The model says that unpopular minister (and former Midnight Oil frontman) Peter Garrett didn’t perform well. This may have to do with how much the Coalition made Garrett the focus of their attacks regarding government incompetence and waste.
La Trobe: Labor won La Trobe from the Coalition at the last election by a narrow margin. However the model considered Coalition incumbent Jason Wood, notable for his speech on genetically modified orgasms, to have performed quite well to have nearly held the seat. Laura Smyth should be able to keep this seat for the Labor party against all but the strongest swing.
McEwen: McEwen was won by Labor at last election when the Coalition left the seat open. As Rob Mitchell, who lost at the 2007 election by just 27 votes after leading on election night, is now the incumbent he should be safer than the Mackerras pendulum would suggest.
Marginal Marginals
Adelaide, Barton, Chisholm, McMahon, Melbourne Ports, Richmond: The AEC doesn’t consider these Labor seats marginal however the model considers incumbency keeps these seats safer than they otherwise would be.
Calare, Cowan, Cowper, Flinders, Forrest, Goldstein, Higgins, Hume, Indi, Gippsland, Grey, Mayo, and Pearce: These seats are all regarded as non-marginal by the AEC however the model predicts that strong performances by their Coalition incumbents that kept them out of the marginal category and that a retiring incumbent and a largish swing against the Coalition could threaten them.
Fisher: The model says that Fisher really should be safe for the Coalition but that “Slippery” Pete Slipper underperformed far more than any incumbent from either party. There are ethical clouds hanging over Slippers head and much of the rest of the Coalition really don’t like him much. There remains a good chance he won’t be selected as the Coalition nominee at the next election.
Leichhardt and Flynn; Based on the 2010 election results the AEC lists both Leichhardt and Flynn as marginal. However the model indicates that the low Coalition results can be put down to resistance from the Labor incumbents, the new Coalition members should survive all but an unprecedented swing.
Ultimately this model would need to be tested in the next Australian federal election, currently scheduled for 2013, to determine whether it has any predictive value or was only useful for describing the 2010 election itself.