Appendix to ‘Did WorkChoices impact on the NSW election results?’

Ben Spies-Butcher and Shaun Wilson, Macquarie University

May 2007

Details of the NSW model

Newspoll results (discussed in the main article) suggest we can assume that lower to middle income voters and voters from the working-age population (aged 18 to 64) are most likely to vote against WorkChoices because they express the most opposition to both the material and symbolic impact of the laws. For the 82 seats for which a two-party preferred swing against Labor in NSW in 2007 is readily available, we are able to test whether seats with higher numbers of working-age voters and lower income earners bucked the trend against Labor, recording smaller swings. We assume that seats behave like the profile of individuals that comprise them, but without individual-level data, this is the best possible.

Fortunately, the NSW Electoral Commission makes available seat-level data about the age and income of voters. We can easily calculate the proportion of adults aged 18 and over who belong to the working age population (18 to 64 years). But the Commission’s income data does not exactly correspond to the relevant Newspoll cut-off point of $70,000 a year household income. However, the Commission’s $999 per week (or $51,999 a year) cut off point is a fair substitute, given that we expect that most individual voters earning less than that belong to households earning below $70,000.

We do not claim that the NSW election result was a referendum on WorkChoices only. Other variables are necessary to capture geographic and political effects. We include a dummy variable for the geography of the seat (using the four way classification scheme available from the Commission: inner and outer metro as well as regional and rural seats). Inspecting the data, we decided to use an additional dummy variable to ‘soak up’ the large swings against Labor that were contained in the Hunter region where the Government faced acute, local problems.

Additionally, incumbency is a factor in limiting swings against governments, so we include a variable to model the possible impact of Labor suffering larger swings in seats where the incumbent has retired or has no sitting member. A final variable is included to capture the differential impact of Labor facing contests with the National Party: this seemed prudent since the National Party did relatively well in the elections, reclaiming two seats (Tweed and Murray Darling from Labor). The Liberals claimed none.

Appendix Table 1 presents the final model with its coefficients and measures of statistical significance. The model has a reasonable R-square of 0.30 and no multicollinearity.

Appendix Table 1: Regression model: What explains the swing against Labor in NSW?
	B coefficient	Significance
Social Demographics:
Proportion earning under $52,000 per year	0.193	**
Working age population	0.258	*
Location (control: Inner met)
Outer met	–3.019	*
Regional	0.212
Rural	1.356
Hunter seat	–4.902	**
Contest:
Labor versus National Party	–2.741
No Labor incumbent	–2.429	*
Constant	–40.455	**

Notes: R-square = 0.30, n=82 electorates. All demographics from NSW Electoral Commission’s ‘Electoral District Profiles’. Dependent variable is two-party preferred swing to Labor calculated from ABC Elections website estimates. 0.01<=p<0.05; *0.01<p.