Election 2007: Did the union campaign succeed?

Ben Spies-Butcher, Macquarie University
Shaun Wilson, Macquarie University

The 2007 election was only the sixth bringing a change of government since World War Two. It was also only the second time in Australian history that a prime minister lost his own seat. Interestingly, on both occasions the incumbent government was challenging the pillars of Australia’s industrial relations system. Before the election, the Coalition consistently denied that WorkChoices was the main reason voters were preparing to switch to Labor. After the election, though, this denial was surprisingly and quickly surrendered. The new Liberal leadership and party director conceded the role of industrial relations in the party’s loss and, in February 2008, the Party dropped its opposition to Labor’s alternative industrial relations system. The forthcoming Australian Election Study (AES) findings will allow researchers a deeper understanding of the Coalition’s defeat. In the meantime, the seat-by-seat analysis we present here provides some early insight into the impact of WorkChoices—and the campaign of the Australian Council of Trade Unions (ACTU) against the laws—on the result.


The union movement, led by the ACTU, led a concerted campaign against the Coalition government’s bid for re-election in 2007. The ACTU’s campaign—Your Rights at Work (YRAW)—began as soon as the 2004 election was decided. At this point it became clear that, from July 2005, the Coalition would have a majority in both houses of parliament, allowing for a much more radical policy agenda. The ACTU (and its state affiliates) organised mass rallies and a large scale advertising campaign, centred on an appeal to the rights and interests of ‘working families’—a theme later adopted by the Opposition Leader Kevin Rudd and used to tie together a range of stresses on lower and middle income Australians that went beyond industrial relations. About a year before the election in 2007, the ACTU also established a number of marginal seat campaigns, each with a full time organiser, to mobilise local union members around the ACTU ‘Your Rights at Work’ (YRAW) campaign, and targeting Coalition MPs. Unions also implemented a systematic program of communicating with their members in these seats. They sought to locate union members who were ‘undecided’ or ‘swinging’ voters, and to engage them in discussions about WorkChoices and the upcoming election.

The 2007 election was only the sixth bringing a change of government since World War Two.

The standard labour movement defences against hostile industrial reform—strike action and appeals to other laws and jurisdictions—had been made more difficult and politically risky by WorkChoices, and a new strategy was required for different times. By appealing directly to voters and portraying the 2007 election as a referendum on WorkChoices, traditional union repertoires were adapted to the remaining opportunities for ‘political unionism’, which took aim at the government’s political, not industrial, vulnerabilities.

Significant circumstantial evidence suggests that the ACTU’s campaign was effective from the beginning. The union media campaign coincided with a slump in the Coalition’s support in opinion polls over 2005–2006. In March 2007, exit polls from the New South Wales state election showed that federal industrial relations issues were a key factor in many people’s decisions (O’Malley & Davis 2007). The federal Labor Party also took on the branding of ‘working families’ and the major themes of the YRAW campaign. Finally, of the 25 seats targeted by the ACTU, 21 changed hands at the 2007 election (see Table 1).

Table 1: Seats changing hands, ranked by size of swing to the ALP
Seat Swing to ALP ACTU target?*
Forde 14.4% No
Leichhardt 14.3% Yes
Dawson 13.2% Yes
Longman 10.3% Yes
Blair 10.2% Yes
Lindsay 9.7% Yes
Petrie 9.5% No
Dobell 8.7% Yes
Makin 8.6% Yes
Flynn 7.9% Yes
Page 7.8% Yes
Macquarie 6.6% Yes
Moreton 7.6% Yes
Wakefield 7.3% Yes
Robertson 7.0% No
Eden-Monaro 6.7% Yes
Deakin 6.4% Yes
Corangamite 6.2% Yes
Bennelong 5.5% No
Bonner 5.0% Yes
Kingston 4.5% Yes
Bass 3.6% Yes
Hasluck 3.1% Yes
Solomon 3.0% Yes
Braddon 2.6% Yes
TOTAL 25 21

Source: ABC (2008); the list of ACTU target seats were supplied by
Ms Sally McManus, NSW Secretary of the Australian Services Union
*The four additional ACTU targeted seats that did not change hands
were Bowman, La Trobe, McEwen and Stirling.

If the ACTU campaign was central to the election outcome, this fact is of considerable significance. While Australian politics has been profoundly shaped by the union movement, political scientists have argued that unions have a declining role to play in Australian politics (see, for example, Leigh (2006)). Certainly, union membership has declined, and unions’ industrial power has considerably lessened. But if unions can still influence election results, then there are important implications for both the future of the union movement and for Australian politics.

More generally, there have been few recent attempts by non-party organisations to influence an election through the mass mobilisation of members or supporters. Compulsory voting has made such ‘get out the vote’ campaigns less significant in Australia than in countries like the United States. If the YRAW marginal seat campaigns were effective, this points to the value of grassroots mobilisation. It may also suggest a re-engagement of social movements and citizens in the electoral process. For this reason, it is important to identify not only the effect of the general campaign, focused as much of it was on media advertising, but also on those aspects that correspond more closely to the actions of a traditional social movement.

Political scientists have argued that unions have a declining role to play in Australian politics.

However, elections are complex events. People decide how they will vote for a range of reasons. In addition to the focus on industrial relations in the YRAW campaign, other important issues were debated in the lead-up to the election, including climate change, interest rates, public hospitals, and education. Voters are also influenced by non-policy factors, such as leadership style, party affiliation, and their local candidates. Given this complexity, how do we assess the impact of the union campaign?

Our strategy is a modest one, given the task. We develop a linear regression model based on the seat-by-seat results, and the swing against the Howard Government (two party preferred) recorded in each. This approach uses statistical correlation to test whether there is a relationship between particular features of a seat (for example, its age composition) and the anti-government swing recorded in that seat. A particular advantage of this type of model is that it can separate out the effects of different factors.

There are two important limitations to our approach. The first is the so-called ‘ecological fallacy’ (see Schwartz 1994). We are looking at seat data, yet seats do not vote—individuals do. Attributing seat-level changes to ‘representative’ individuals assumes a close correspondence between individual voter behaviour and seat outcomes. This correspondence is not always reliable and can be a source of faulty reasoning. We are conscious of this limitation. However, in the absence of publicly available and reliable exit polling, this strategy remains the best so far available. This approach also has the advantage of identifying the marginal seat campaign effects, which are easily captured at a seat level. Second, as with all analysis based on statistical correlation, by itself, our model cannot reveal deep causation. However, it can help test some of the explanations that have emerged since the election and help build a picture of the underlying electoral dynamics at play on 24 November.


Our model is given in Table 2. It models the impact of a range of factors on the swings recorded in individual seats, factors identified in the broader literature as potentially significant, either in Australian elections or in this campaign. Political scientists have identified certain political characteristics of a seat with potential voting behaviour in elections. Using 2006 census data we can identify characteristics of seats, such as the average income or the proportion of people within a certain group (such as an age or ethnic group). We can also identify aspects of seat location, and of course electoral variables, like the swing, whether a seat is safe, and which party holds the seat. Here we use census data compiled by electorate for George Megalogenis (2007a) and published on The Australian website as well as census data complied by the Parliamentary Library of Australia (2007). Broadly, the data reveals an electorate that is increasingly educated, multicultural and moving north. According to Bowe (2008a), Queensland is likely to pick up a further seat in a future redistribution, and with a possible 30 seats in the House of Representatives is beginning to rival Victoria (37) in electoral significance. Brisbane and the greater Gold Coast are emerging as the third major east-coast metropolis, their growing regional power underlined by the fact that both the Prime Minister and Treasurer come from Brisbane.

On election night, it also became clear that recorded swings were varying by state.

The ‘incumbency advantage’ has proven important to sitting members in other Australian elections. Seats with an incumbent member seeking re-election have tended to experience smaller negative swings than seats without an incumbent (see Spies-Butcher & Wilson (2007) on the NSW election). This is usually ascribed to the personal vote and profile of the incumbent. Because Australian lower house elections depend on winning a majority of single member electorates, political and financial resources are concentrated on marginal seats. Indeed, the Coalition’s election strategy, reportedly aimed at reducing swings in its marginals, raised the hope of holding office with a majority of seats but a minority of the two-party preferred vote (Shanahan 2007a). To account for effects of incumbency, we include a dummy variable for those seats with no incumbent member. We also include dummy variables for the type of seat in electoral terms; that is, whether the seat was safe or marginal (the cut off being a margin of 5 per cent), and which party held the seat.

On election night, it also became clear that recorded swings were varying by state. This was made more evident by the delay of results from Queensland in the ABC television coverage. Before the Queensland count became available, the result appeared closer. When the Queensland count came in, the result consolidated for Labor, with the stronger swings in that state than were recorded elsewhere. Likewise, both in the lead-up to polling day, and on the night, Western Australia was more favourable to the Coalition. Here, we test if there was a significant variance in the swing by state.

The swing also appeared to vary by urban geography. In one post-election analysis, William Bowe (2008b) from the political website Poll Bludger reports that swings in inner-metropolitan areas were significantly lower than in outer-metropolitan areas. There were also particularly large swings in some regional and rural areas, especially in northern Queensland. Again, we test for this by including variables for the location of a seat based on Australian Electoral Commission classifications (rural, provincial, inner-metropolitan and outer-metropolitan).

The last group of variables in the model tests a number of election-specific hypotheses about the result. We begin by including a variable for mortgage stress—seats with a high proportion of households paying more than 30 per cent of their income in mortgage repayments. Rising interest rates were a key issue in the election, with the Reserve Bank raising rates for the first time during an election campaign, but with many commentators still ascribing the Coalition’s success in the 2004 campaign to the public’s greater confidence in the Coalition’s ability to manage the economy. (We also tested for the number of mortgage holders in an electorate, but this proved less significant than the mortgage stress measure included here.)

We also looked for evidence of backlash from single parents, following political commentator George Megalogenis’s claim that swings appeared higher in seats with more single parent families, a possible reaction against welfare-to-work reform (see Megalogenis 2007b; Australian Broadcasting Corporation 2007). However, this was not included in the final model as it did not prove significant.

Finally, we include two variables related to the ACTU’s campaign against the Coalition’s WorkChoices policy. We divide this effect into (i) the broad campaign to win the votes from working families, later endorsed by the Labor Party, and (ii) the specific impact of the YRAW campaign on seats targeted by the ACTU.

Our model shows that older voters were less likely to swing to Labor than younger age groups.

First, we attempt to capture the potential swing among working families. Here we use demographic variables, often identified as important in shaping voting behaviour, to help us identify seats with a large proportion of ‘working families’. ACTU research reportedly showed about 75 per cent of Australians viewed themselves as ‘working people’, up to annual incomes of about $80,000 (Steketee 2007). Unions later described their target as ‘footy dads’ and ‘soccer mums’. Here we focus on families in ‘middle Australia’, which we argue broadly matches this description. We construct a ‘working families’ variable by combining middle Australian households (defined by income) with data on families with dependent children. We began by identifying broadly middle income families. Drawing on data from the Parliamentary Library (2007), we identify seats by the proportion of families with incomes above $650 ($33,800 per year) per week and below $2,000 per week ($104,000 per year). (These cut-offs are dictated by data availability, but we believe they are a reasonable approximation of the income points we require.) This measure excludes families on very high incomes, and those on very low incomes, where income is likely to primarily come from government benefits or small businesses. Next, we add to this the proportion of households with dependent children (having standardised the two initial variables). The final measure, called ‘working families’ in the model, ranks seats by the number of families with children and middle income earners present. To illustrate, among the top five seats for working families were Leichhardt and Forde, which recorded the two largest swings of the election. At the other end, we have seats like Higgins and Wentworth—wealthy Melbourne and Sydney electorates—which recorded little or no swing away from the Liberals.

Second, we decided to separately test for the effect of specific marginal seat campaigning undertaken by the ACTU. We do this by including a variable for the 25 seats in which the ACTU employed a full-time marginal seat co-ordinator one year out from the election. We are interested to see if a specific campaign effect went beyond the general effect generated in seats with working families. We recognise that unions campaigned in other seats, and quite intensively, especially as the election drew near. Measuring these ad-hoc efforts is difficult without knowing and measuring the precise level of resources involved in each campaign. The advantage of the ACTU role in the marginal seat campaign is that its stated impact can be reliably identified and measured.

We also examined a number of demographic variables associated with voter preference. For age, we include the proportion of the population over 65, focusing on the strong support enjoyed by the Coalition among older voters (Shanahan 2007b). Publicly available data on education was limited, and so here we included the proportion of the population with less than a Year 10 education. In some versions of the model, we also included variables for the proportion of the population who were first or second generation migrants, and a general measure for income. Both these variables proved troublesome in that they interacted with other variables in ways that threatened the reliability of the model. (This is the problem of ‘multicollinearity’.) Migrant levels correlate closely with location, while income was closely correlated with the low education variable. As we are using seat level data, there are only very small differences in gender composition between seats.

Note that the model includes 147 seats, three less than the number contested. The three omitted seats are New England, Kennedy, and Melbourne. In these three seats, a non-Labor, non-Coalition candidate was included in the two party preferred result, making the reported swing incompatible with the measure of swing we use in the model.

Table 2: Seat-level model:
What explained the swing against the Howard Government?
  B coefficient Significance
Over 65 -.19 *
Low Education .07 *
No incumbent member -.40  
Party affiliation (control: Marginal ALP)    
     Safe ALP seat -.66  
     Safe Coalition seat -.18  
     Marginal Coalition seat -1.61 *
State (control: Victoria)    
     NSW .18  
     Qld 1.80 **
     SA 2.56 **
     WA -3.08 **
     Other States and Territories -3.15 **
Location (control: Inner metropolitan)    
     Outer metropolitan -.62  
     Provincial .60  
     Rural .23  
Households experiencing mortgage stress 10.48  
ACTU campaign    
     Working families .03 **
     25 ACTU targeted marginal seats 1.59 **
Constant -1.28  

Notes: Adjusted R-square = 0.526, n=147 electorates.Demographic
data from Parliamentary Library (2007) and Megalogenis (2007a).
Dependent variable is swing to Labor calculated from Australian
Broadcasting Corporation (2008). *0.01<=p<0.05; **0.01<p.


The results, presented in Table 2, suggest several factors influenced the election outcome. There is some evidence that each type of variable had some impact. Consistent with some pre-election polling, the model shows older voters were less likely to swing to Labor than younger age groups. However, the correlation is sensitive to the exact construction of the model, and so we are less confident in this result. We return to education below.

Surprisingly, there is little evidence that the political affiliation of the seat, or the departure of a sitting incumbent, made any significant difference to the result once other factors were controlled for. The exception was marginal Coalition seats (relative to marginal Labor seats) where the swing was significantly lower—providing some evidence that the government strategy to hold its own marginals had an effect. This effect remained relatively stable across a number of models.

There is little evidence that mortgage stress, in itself, was decisive to the outcome of the election.

State level variations appear much more significant. We took Victoria as the base category, both because it is relatively large and because it seemed to follow the overall pattern of the election. As expected, Queensland produces a much larger swing than Victoria, as does South Australia. On the other side, Western Australia produces a much smaller swing as do other States and Territories category (comprising Tasmania, the Australian Capital Territory and the Northern Territory), while New South Wales is not significantly different to Victoria. Again, this is consistent with other analysis of the election result, and these relationships proved stable over different models.

The story is more complicated when it comes to where seats are located relative to large urban centres. This model does not reveal any significant variations for either rural, provincial or outer-metropolitan areas relative to inner-metropolitan areas. Importantly, these variables changed substantially depending on the composition of the model. The exclusion of the education variable resulted in significant swings in rural and provincial areas, while the exclusion of the ‘working families’ and mortgage stress variables resulted in outer-metropolitan areas recording significant swings. We suggest that these results indicate that the larger swings recorded outside the inner city were primarily the result of the social profile of these electorates, rather than their location per se.

Returning to the effects of education, our model shows that those seats with a ‘low education’ profile were more likely to swing to Labor. But the effect of education varied in significance, depending on what other indicators of socio-economic disadvantage were included in a model. What we can conclude, more generally, is that seats with the combined profiles of low income and poor education were more likely to swing to Labor.

Interestingly, the model shows little evidence that mortgage stress, in itself, was decisive to the outcome of the election. This variable was not significant in any of our models. However, this model measures the impact of mortgage stress on the swing between the 2004 and 2007 elections. If, as some have claimed, there was a strong swing to the Coalition at the 2004 election from the ‘mortgage stressed’, then this group’s continued support for the Coalition in the election of 2007 might not be captured in the model. In any case, our model gives little support to the claim that dissatisfaction with rising interest rates, and their effect on mortgage stressed households, benefited Labor. However, the full effects of mortgage stress may be masked. If stressed households registered the impact of rising rates but still saw the Howard Government as best able to manage interest rates, as polls before the election indicated (albeit by falling margins; see Newspoll 2007), then the benefit of rate rises for either party would have been ambiguous. Concluding that interest rates did not explain the swing is not the same as saying that interest rate politics was unimportant.

The ACTU campaign had a significant impact on the outcome of the election.

We find limited evidence for George Megalogenis’s interesting claim about the voting response of single parents. Seats with more single parents did appear to swing significantly more in some models—consistent with Megalogenis’s argument. However, in most of the models the variable was not significant, particularly when age was included in the model, suggesting that much of the effect was likely due to younger people being both more likely to be single parents and more likely to have moved to Labor.

But the most noteworthy finding, for our purposes, comes from the final set of variables relating to WorkChoices. We find evidence that the ACTU campaign had a significant impact on the outcome of the election. First, our ‘working families’ variable shows a significant, consistent, and strong relationship to the size of the anti-Howard swing. Seats home to middle income families with kids (the ‘working families’ demographic) were significantly more likely to swing to Labor. This result only reinforces our earlier finding that seats with larger older populations were significantly less likely to swing to Labor, as older families are far less likely to have members in the workforce and have dependent children.

Perhaps more surprisingly, our model also suggests a significant and sizeable effect from the YRAW marginal seat campaigns. Again, this was a very stable result in several different models, indicating an additional swing of between 1.3–2 per cent in seats with ACTU-led YRAW organisers and campaigns, once other factors are controlled for. Importantly, this effect is independent of the strong and significant swing recorded in seats with higher numbers of working families. One way of illustrating the impact of the specific ACTU campaign effects is to count the number of targeted seats with very small margins (under 1.3 per cent). We find five seats—Bass, Corangamite, Hasluck, Flynn and Solomon—were all won by margins less than or equal to 1.3 per cent, the lower estimate in our models attributed to the campaign’s influence. A further two seats—Braddon and Deakin—were won with margins of less than 2 per cent, the upper limit of the estimate.

We are not claiming Labor would not have won the election had it not been for the ACTU campaign. Had the ACTU relied on a Labor-led campaign against WorkChoices, and not independently campaigned in marginal seats, it is more than likely Labor’s appeal to working families and the ACTU’s media campaign would have produced a broadly similar result. What we do claim, on the basis of this modelling, is that seats targeted by the ACTU produced significantly larger swings, and their campaign appears to have added to Labor’s margin of victory.

Perhaps the era of activist electoral politics is not yet dead, but waiting to be remobilised.

We are not yet in a position to find out how well the campaign effects identified here stand up against attitudinal responses to WorkChoices that will be available to researchers when the Australian Election Study is released later this year. It is possible that attitudes to WorkChoices will do the ‘heavy lifting’ and dominate the ‘campaign effects’ (and ‘working family’ demographic effects) we identify using aggregate seat and census data. At the same time, though, the AES will also provide additional measures of labour movement activism in the election campaign (that is, union member voting and campaigning behaviour) that will help complete the overall picture of union influence.


In America, politics in recent years have been shaped by greater mobilisation of the union vote for the Democrats under a reformist AFL-CIO leadership that won office in 1996 (and their new rivals in the ‘Change to Win’ coalition). Union mobilisation of the vote is an offshoot of political unionism that has tried to respond to the declining capacity of industrial unionism (because of low union density and limited union impact on wages) and to the recognition that genuine political allies and legal change are increasingly necessary for organised labour’s revival. As Margaret Levi (2003) has made clear, the union movement depends not only on a strong shopfloor presence but on a favourable legal and political environment as well. Better laws are critical to the labour movement’s long-term hopes, both in the United States and in post-WorkChoices Australia.

The impact of vote mobilisation in America has been substantial. In the 2000 presidential elections, despite ongoing declines in union density in the workforce, unionised workers made up over a quarter of all voters—up from 19 per cent in 1992 (Chang 2001, p. 375). Higher numbers of union voters, and a stronger preference for the Democrats among these voters, were also crucial to limiting Bush’s second term majority and to the substantial win by the Democrats in the 2006 congressional elections (Greenhouse 2006).

Australia’s compulsory voting system means there is relatively little research on mobilisation campaigns. Moreover, if voters are obliged to vote, there is little need to develop vast grass roots networks to mobilise them. Yet the YRAW campaign appears to be an example of the success of such a strategy. The decline in union density means, almost automatically, a weakening in political influence—both in a diminished voter bloc and perceptions of weakness that embolden opposition. Like the American labour movement, the ACTU has offset its declining natural constituency by more strongly mobilising its remaining membership, renewing it in the process. And so the tactics the ACTU employed during the 2007 election were much closer to those of a grass roots mobilisation than to the simple increase in resources, or targeted promises, that accompany other marginal seat campaigns. This is important both in highlighting the continuing power and importance of the union movement in Australia, and in opening up the possibility of the broader significance of electoral mobilisation by social movements. Perhaps the era of activist electoral politics is not yet dead, but waiting to be remobilised.


Australian Broadcasting Corporation 2007, Insiders, 25 November.

Australian Broadcasting Corporation 2008, ABC Elections: Australia votes 2007 [Online], Available: http://www.abc.net.au/elections/federal/2007/ [2008, Feb 18].

Bowe, W. 2008a, ‘The luck of the draw: Federal redistributions’, The Poll Bludger, 15 January [Online], Available: http://www.pollbludger.com/790 [2008, Feb 19].

Bowe, W. 2008b, ‘Post-match report: Melbourne’, The Poll Bludger, 2 January [Online], Available: http://www.pollbludger.com/785 [2008, Feb 19].

Chang, T.F. 2001, ‘The labor vote in US national elections’, The Political Quarterly, vol. 72, no. 3, pp. 375–385.

Greenhouse, S. 2006, ‘Labor movement dusts off agenda as power shifts in congress’, The New York Times, 11 November.

Leigh, A. 2006, ‘How do unionists vote? Estimating the causal impact of union membership on voting behaviour from 1966 to 2004’, Australian Journal of Political Science, vol. 41, no. 4, pp. 537–552.

Levi, M. 2003, ‘Organizing power: The prospects for an American labor movement’, Perspectives on Politics, vol. 1, no. 1, pp. 45–68.

Megalogenis, G. 2007a, ‘Battlers who will shape the poll’, The Australian, 20 August [Online], Available: http://blogs.theaustralian.news.com.au/meganomics/index.php/theaustralian/comments/ [2008, Feb 19].

Megalogenis, G. 2007b, ‘Howard’s restless battlers’, The Australian, 30 June.

Newspoll, 2007, ‘Federal voting intention and leaders’ ratings + economy, interest rates and treasurer’, 20 November [Online], Available: http://www.newspoll.com.au/ [2008, Feb 18].

O’Malley, N. & Davis, M. 2007, ‘Unions crow at blow to Howard’, The Sydney Morning Herald, 26 March.

Parliamentary Library of Australia 2007, ‘Electoral division rankings: Census 2006 first release (2006 electoral boundaries)’, Research Paper No. 12 2007–08, Parliament of Australia, Canberra [Online], Available: http://www.aph.gov.au/LIBRARY/pubs/RP/2007-08/08RP12.pdf [2008, Feb 18].

Schwartz, S. 1994, ‘The fallacy of the ecological fallacy: The potential misuse of a concept and the consequences’, American Journal of Public Health, vol. 84, no. 5, pp. 819–824.

Shanahan, D. 2007a, ‘Paths to victory’, The Australian, 3 November.

Shanahan, D. 2007b, ‘Grey voters PM’s last stronghold’, The Australian, 1 November.

Spies-Butcher, B. & Wilson, S. 2007, ‘Did WorkChoices impact on the NSW election result?’ Australian Review of Public Affairs, Digest, May [Online], Available: http://www.australianreview.net/digest/2007/05/spies-butcher_wilson.html [2008, Feb 18].

Steketee, M. 2007, ‘Family formula gets workout’, The Australian, 17 November.

Ben Spies-Butcher is Lecturer in Sociology at Macquarie University.

Shaun Wilson is Senior Lecturer in Sociology in the Division of Society, Culture, Media and Philosophy at Macquarie University.