Friday, 28 November 2014

Method to the Madness

[Backfiled: originally published 11/1/2015]

The methodology for the predictions of the Victorian State election were based on applying the polled swing to the current margin, and although attention was paid to the historical incumbency this was not particularly helpful as 14 were not contested in the previous election.

Here, because the Infographinomicon is at least nominally a source of infographics, is a chart of Victorian state seat incumbency since 1900:

However, coincidentally, all of the predictions were not contradicted by the seat histories. The predicted successful party of each seat has held the seat in question at some point since the last two elections for all seats with a recent electoral history. Of the 14 seats without a recent electoral history, 11 are entirely new seats. The remaining three are Ringwood, St Albans and Werribee. Ringwood was predicted for Liberal, and was L&C or Liberal for 24 years of its 34 year run between 1958 and 1991. St Albans was predicted for Labor, and in its previous incarnation from 1985 to 2001 had been a consistent ALP stronghold. Werribee was predicted for Labor, and with the exception of its first election in 1976 was always won by Labor until it too was abolished in 2001.

The predictions, as previously noted, are also entirely consistent with applying the uniform 5.1% swing to Labor* to the each seat’s margin in all seats where the 2PP vote is split between Labor and a coalition party, after adjusting the margin to factor in boundary changes (based on the VEC’s district profiles).

However, this method does not provide predictions in the seats where the 2PP contest is not divided between the ALP and a coalition party.

For a more nuanced approach, inter-party swings were calculated as follows:

Where A2010 is the percentage of the primary vote won by the party in question at the last election, B2010 is the percentage of the primary vote won by the second 2PP party at the last election, and APoll and BPoll are the primary vote polling data for those respective parties according to Newspoll’s October 27-30 data, the most up-to-date poll that distinguished between Liberal and National support at the time the predictions were made. In other words, the swing is calculated as the polled shift from one party’s primary vote to or from its 2PP rival, ignoring all other parties and preference flows. The calculations were:

Labor to Liberal Swing:
2010: 38/74 (51.4%)
Poll: 35/76 (46.1%)
Swing: -5.3%

Liberal to Labor Swing:
2010: 36/74 (48.6%)
Poll: 41/76 (53.9%)
Swing: +5.3%
Labor to National Swing:
2010: 36/43 (83.7%)
Poll: 41/45 (91.1%)
Swing: +7.4%
National to Labor Swing:
2010: 7/43 (16.3%)
Poll: 4/45 (8.9%)
Swing: -7.4%
Labor to Green Swing:
2010: 36/47 (76.6%)
Poll: 41/54 (75.9%)
Swing: -0.7%

Applying these swings to the adjusted polling results yields the following predictions:

Giving our 2014 predictions. As previously noted, both Gembrook and Mount Waverly were a little sketchy considering their electoral past, predicting Liberal in two twice-Labor once-Liberal seats during a pro-Labor swing, but given their short electoral histories ant the fact that this was not outside the realms of possibility, no tossups were used.

Examination of the accuracy of these predictions will follow soon.

* Polling 53.5 2PP compared to 48.4 after the 2010 election.

No comments:

Post a Comment