With the Victorian election almost upon us there are many guides to individual seats available. To add to these I have developed a simple regression model to predict the 2010 Labor primary and two-party preferred votes and the 2010 Green vote in each electorate from the social composition of the electorate as revealed by the 2011 census. As this is a ‘back of the envelope’ exercise I have kept it simple. The dependent variables are the portion of the electorates’ population that do not speak English at home and the portion not employed in professional-managerial occupations. I have also included a dummy variable for ‘agricultural electorates’; defined as those in which more than 5% of the workforce is employed in agriculture, forestry or fishing. This model sheds light on the distinctive features of individual regions and possible personal votes. The final results and residuals are available here.
The results are:
Labor 2010 Primary
|Non-Professional & Managerial||0.57||0|
R-Sq. (adjusted) 0.69
Labor 2010 Two-party preferred
|Non-Professional & Managerial||0.34||0|
R-Sq. (adjusted) 0.49
Greens 2010 Primary
|Non-Professional & Managerial||-0.33||0|
R-Sq. (adjusted) 0.45
There is nothing very surprising here but the gap (‘residuals’) between the predicted and actual party votes in electorates are revealing. Labor polls poorly in the eastern suburbs of Melbourne. The model predicts a Labor primary vote of 39% in Doncaster compared to the 26% Labor actually received. Labor failed to win 8 electorates predicted by the model: 7 in the outer east & Mordialloc but won 16 predicted as Liberal victories. These electorates were all in the inner city & provincial cities plus Eltham, Macedon & Monbulk. For all the talk about the ‘sand belt’ Labor’s 2010 performance here was only slightly below that predicted by the model. Below I discuss some key electorates and regions. Electorates where Labor or the Greens ‘over performed’ are those where they polled more than the model predicts, ‘underperformed’ are where model predicted a higher vote than actually occurred.
Labor over performs by 19% on two-party preferred on primaries and 18% on primaries. Much of this reflects local political culture – Ararat being part of the goldfields but it does suggest that Joe Helper had a substantial personal vote.
Labor over performed by 7% on two-party and primary vote. This is a different type of middle-class Melbourne than the outer east but it does provide some evidence of a possible personal vote for retiring Labor candidate Steve Herbert.
Labor underperformed notably on two party (14%) and primaries (10%) but the model does not predict Labor should have won it in 2010. It is no longer a natural Labor heartland electorate but it would be a Labor seat if it manifested the pro-Labor tendency of the other provincial cities.
Labor over performs in all four Bendigo & Ballarat electorates. The model predicts Coalition would have won all four in 2010. This may explain the Coalition pursuit of these electorates. Labor’s strong performance here (and to lesser extent in Geelong) confirms Antony Green’s emphasis on the importance of Labor’s regional electorates.
Some Labor people think they should do better here as the electorate is increasingly ethnically diverse. Labor does notably underperform but it should still be a safe conservative electorate (although the predicted Labor vote is actually slightly above that predicted for Ripon).
Labor over performed by around 5% on two-party and primary votes.
Rather like the goldfields region Labor over performed here (although not to the same extent), Bellarine & Geelong both predicted as losses by the model.
Their vote unsurprisingly is very high in the inner city, it is under predicted between 16 and 19% for Brunswick, Richmond, Melbourne and Northcote. Rather like Labor they underperform in the eastern suburbs.