THE ALTERNATIVE VOTE IN AUSTRALIA

THE ALTERNATIVE VOTE IN AUSTRALIA Exacerbating a Culture of Adversarialism? Sarah John 1, Leigh Hargreaves 2 1 School of International Studies, Flinders University GPO Box 2100 ADELAIDE, 5001, Australia sarah.john@flinders.edu.au 2 Department of Physics, California State University, Fullerton 800 North State College Boulevard Fullerton, CA 92834, United States of America lhargreaves@fullerton.edu This paper models the outcomes of South Australian state lower house elections from 1997 to 2010 under five different ordinal vote counting methods. The results of the model confirm the findings of Rae (1971) that AV behaves in much the same manner as plurality voting. However, our data show that AV is unique among ordinal voting methods in this regard. In all other ordinal systems modelled, a greater number of parties gain representation in the lower house and in most years no party would win a majority of. Additionally, non-av ordinal methods display a greater sensitivity to subtle changes in voters preference orders. The paper then considers the wider ramifications of the AV system on political culture in Australia, which at present is characterised by a highly partisan and adversarial tenor. We conclude that the use of a non-av ordinal voting method would either produce multiparty lower houses (if voters preference orders are assumed to be sincere) or provide an incentive for major party voters not to preference each other last (if voters preferences are assumed to be strategic). In either case, a non-av ordinal voting system would provide a basis from which a more constructive and less antagonistic political culture could develop. 1

THE ALTERNATIVE VOTE IN AUSTRALIA: Exacerbating a Culture of Adversarialism? INTRODUCTION The relationship between political institutions and political culture is becoming more and more relevant to established democracies, which are increasingly looking to electoral system reform in the face of declining voter turn-out and concerns about citizen disengagement with democracy. Ideas that institutional reform can lead to a change in voters behaviour and their relationship with democratic institutions abound (eg. Blais 2008, Donovan and Bowler 2004, Norris 2004, Brischetto 1995, Zimmerman 1994). Voting system reform has been an especially popular target of reformers seeking to reinvigorate ailing Western democracies. In the last decade, the United Kingdom seriously considered reforming its voting system (which climaxed with British voters rejection of the Alternative Vote (AV) in May 2011) and several Canadian provinces have looked to completely overhaul their electoral system, including their voting system (most notably in British Columbia and Ontario). Voting system reform occurred in New Zealand, which abandoned the plurality system in 1996 and is due to vote on the retention of the Mixed Member Plurality (MMP) system in November 2011. Several cities in the United States (including San Francisco, California, and Minneapolis and St Paul in Minnesota) have adopted AV since 2000, and many more have held initiatives or referendums on voting system reform. In light of this reform movement, understanding the full consequences of voting systems is crucial. While much of the present attention around voting systems focuses on the conversion of votes into, other aspects such as the impact on political culture are also subjects of interest. For example, in Britain s recent referendum campaign over AV, a major issue was the potential effect of AV on British political culture. Commentators wondered if AV could promote more civil politics or encourage legislators and parties to deal more constructively with the substantive issues (eg. Economist Online 2011). So central was political culture to the UK debate that when the BBC interviewed former Australian Prime Minister, John Howard, one of the main themes was whether AV would tone down yahoo politics. Mr Howard categorically rejected any connection between voting systems and the behaviour of legislators or parties: Well, I think most people who watch our [Australian Federal Parliament] Question Time would think it s fairly rowdy, but look, I don t think the voting system has anything to do with the behaviour [of legislators]. We [in Australia] have, we have not only preferential voting, or AV, or whatever you want to call it, but we also have compulsory voting. I think voting systems don t have anything to do with behaviour, I think that s a fairly fatuous argument frankly (BBC News, 2011). Despite Mr Howard s certainty that the link between them is absurd, voting systems clearly have some impact on the behaviour of political actors. The most obvious example of this impact is the relationship between compulsory voting and eligible citizens propensity to vote. Another example is the wellestablished association between legislators tendency toward coalition building and compromise in legislatures elected using proportional representation (PR). The relationships between more nuanced and subtle institutional choices and behaviour are still largely unexplored. Far from being settled, the 2

relationship between AV and the behaviour of legislators is an open question. Whether AV might support a more civil political culture or increase yahoo politics is not known. This paper addresses a gap in our knowledge and examines the likely long term effect of Australia s use of AV on its political culture. The paper begins by examining AV s role in converting votes into as a platform for later consideration of AV s impacts on political culture. Results from the 1997, 2002, 2006 and 2010 South Australian House of Assembly elections are modelled, using actual ballot data, under the existing AV system and four other ordinal systems. South Australian elections are used as a proxy for most Australian lower houses electoral systems to assess AV against other ordinal voting methods. The model compares ordinal methods because they operate to a common majoritarian logic and are identical from the point of view of electors marking the ballot paper. These ordinal systems offer a unique opportunity to examine the impact of AV compared to other similar systems controlled for most variables. It is observed, consistent with existing literature, that electoral outcomes under plurality and AV are generally inconsequentially different. However, if AV is replaced by any of the other ordinal systems considered, electoral outcomes are markedly different and highly dependent on the ordinal vote counting method employed (if voters preference orders are assumed sincere). AV is found to be unique in its tendency to redirect preferences back into for the two established major parties and in inhibiting the success of new moderate parties as well as short term reactionary movements. Based on the observation that non-av ordinal systems produce significantly different electoral results to AV, the paper then explores the likely effects on political behaviour and culture of Australia s use of AV over the last seven or so decades in which it has been standard. The extent to which the characteristics of AV exacerbate, or at least fail to discourage, the fairly rowdy, antagonistic or unrestrainedly partisan political culture in Australia is postulated using two alternate assumptions about the nature of voters preference orders in the modelled elections. Firstly, the paper investigates the political cultural consequences of AV if the modelled voter preferences are assumed to be sincere. AV is found to be unique amongst ordinal methods in supporting a rigid, adversarial, two-party system. Under all non-av ordinal systems, Australian lower houses would be more multiparty and perhaps take on a characteristic more like the coalition based proportional systems of Western Europe, with their focus on negotiation and compromise. Secondly, the paper considers the implications of a non-av ordinal system if voters preferences are assumed to be strategic. The paper finds that AV provides incentives for a strategic major party voter to preference the other major party last, encouraging the tendency to view the other major party as the enemy and enabling a highly corrosive party dialogue. By contrast, by giving greater weight to preference orders, non-av ordinal voting systems necessarily de-incentivise preferencing strategies that exacerbate an aggressively partisan political culture. We argue, irrespective of whether preferences are sincere or strategic, all other ordinal systems would support a more cordial and constructive political dialogue, especially between the major parties, leading over time to a more constructive political culture. We conclude that AV is unique amongst ordinal voting systems in that it supports and perhaps encourages hostility between the largest parties thus contributing to Australia s harsh political culture. 3

POLITICAL CULTURE AND VOTING SYSTEMS The relevance of political culture to the design of political institutions has long been recognised, typically by scholars explaining the success of democracy in some societies but its failure in others. A vibrant literature and ongoing debate between political culturalism and rational choice institutionalism over the importance of culture relative to institutions in producing stable democracies is one of the central questions of modern political science. (Lipsett 1990; See also Norris 2004, Chapter 1, Pye and Verba 1965, Hartz 1964, Roeder and Rothchild (eds) 2005) This paper does not enter the debate between culturalists and institutionalists, but begins with the rational choice assumption that institutions provide incentives to actors behaviour, to which we add our own assumption: that these incentives, in the long term, may have an impact upon the political cultural of a democracy. With these assumptions in hand, we consider the impact of a subtle institutional choice the choice of AV over other ordinal voting methods and plurality voting made generations ago on Australia s political culture today. In Australia, a significant attempt at considering the impacts of AV on various aspects of Australian politics has been undertaken. In terms of election results (that is, the conversion of votes into ) the consensus is that AV produces very similar legislatures to plurality. American political scientist Douglas Rae famously noted in 1971 that the Australian system behaves in all its particulars as if it were a single member district plurality formula (Rae 1971, p. 108, also Bean 1997, p. 110, Farrell and McAllister 2006). Beyond this certainty about AV s similarity to plurality in terms of election results, less is known about AV. Farrell and McAllister maintain that the Australian analysis of AV would be well served if it were to break free from the fixation with votes into calculations and focus on the more fine grain impacts of voting systems (Farrell and McAllister 2006, p. 173). In so considering these micro-behavioural impacts, Farrell and McAllister hypothesised that AV would lead to greater voter influence over the system, but instead find that AV when in concert with other aspects of the electoral system (compulsory voting, party control over pre-selections and the use of How to Vote (HTV) cards) results in a system heavily focussed on the party not on the voter. Indeed, they conclude that the Australian electoral system gives parties a great deal of control over politics and that that control is on the rise (Farrell and McAllister 2006). Orr and Ewing (2011) come to a similar conclusion with regards to AV s propensity to increase the power of political parties relative to voters. While these studies have implications for political culture, few studies expressly consider the relationship between AV and political culture. Reilly honed in more closely on the cultural impacts of AV, arguing that the AV system provides strong incentives for the major parties to keep their focus on the middle ground at all times (Reilly 2001a, p. 79, Reilly 1997; See also Bean 1986). In the context of ethnically divided societies, Reilly concluded that AV provides centripetal incentives, which encourage moderate candidates and parties and a consensual politic (Reilly 2001a, Reilly 2001b). Closely related to Reilly s findings is an ongoing, and intense, debate between Fraenkel and Grofman and Horowitz about the impact of AV in militating against ethnic divisions in Fijian politics (Fraenkel and Grofman 2007, Horowitz 2007, Fraenkel and Grofman 2006, Fraenkel and Grofman 2004, Horowitz 2004). The tendency of AV to produce a moderate political culture is thus still an open question made all the more complicated by Jansen s findings that the existing political culture into which AV is imposed is more important to AV s impact (at least in its initial years of operation) than AV s mechanics (Jansen 2004). 4

It is important to note that all the above literature compares AV against only plurality voting. Little is presently known about AV s consequences vis-à-vis other ordinal voting systems. Indeed, there is a dearth of data about other non-av ordinal systems generally. Proponents and reviewers of different ordinal systems argue their merits often by reference to social choice theory criteria such as proportionality, fulfilment of the Condorcet criterion, representation of minorities and stability (Blais 1991, Punnett 1991, Hallett 1978, Geller 2002, Chen and Heckelman 2005, Horowitz 2003). These judgments are typically based on only assessments of the methods theoretical operation and mathematical properties. Because so few ordinal voting systems are presently in use, there is an absence of empirical evidence about how different ordinal systems operate relative to each other. The most useful observed evidence about different ordinal voting systems comes from a small literature that models the outcomes of elections within organisations such as labour unions using real ordinal ballot data (Felsenthal, Maoz and Rapoport 1993, Felsenthal and Machover 1995, Levin and Nalebuff 1995, Regenwetter et al 2007). This literature tends to indicate that electoral outcomes are largely insensitive to the chosen ordinal voting method. This runs contrary to rational choice and social choice theory, both of which predict electoral outcomes to depend greatly on the voting method chosen. Additionally, the literature models only the votes to outcomes of elections and has nothing to say about the cultural impacts of ordinal voting systems. More ought to be known about AV relative to other ordinal vote counting methods since, in single member district systems, they are united by their common core goal to elect a candidate broadly acceptable to a majority in his or her district. AV has much more in common with other ordinal methods than it does plurality (or PR). A greater understanding about AV relative to other ordinal methods would help Australia assess its political institutions institutions that ought to be increasingly questioned in light of the recent tendency toward minority governments (especially in the states), moribund political party memberships and increased levels of informal voting and non-registration. Additionally, a greater understanding about ordinal methods would have beneficial effects for the continuing electoral reform debate in other established democracies, which tends to be unduly limited to abandoning plurality in favour of three options: AV/Single Transferable Vote, MMP or PR. MODELLING FIVE ORDINAL VOTING SYSTEMS IN SOUTH AUSTRALIA To examine the link between voting systems and political culture, five different ordinal voting methods 1 (AV, Coombs Method, the Bucklin Method, Borda Count and Condorcet Voting) were modelled using election results from four consecutive South Australian House of Assembly elections. South Australia is used as a proxy for AV elections held across Australia. South Australia was chosen for the model because it uses an electoral system that is close to standard across Australia 2 and preference data are more easily modelled for lower house elections there than elsewhere. Due to a peculiarity of South Australian election law, which encourages almost all candidates to lodge a HTV card with the Electoral Commission of South Australia, preference flows can be readily inferred from ballot data. Whereas in all other jurisdictions data collection proves to be an almost insurmountable 1 The term ordinal voting is used here to avoid confusion stemming from Australians common reference to AV as preferential voting, despite AV being but one species of preferential voting. 2 Every 4 years, elections are held to fill the 47 vacant in the lower house of the legislature (the House of Assembly). Each seat corresponds to a geographical district, with eligible voters residing in each district electing a single representative to hold that seat for the next 4 years. Enrolment and voting are compulsory. 5

obstacle, in South Australia the impact of different ordinal vote counting rules on results can be demonstrated (Hargreaves and John 2011). The five ordinal vote counting methods were chosen because they vary only in their computation after the voter has cast his or her vote. Each of these methods is identical from the perspective of the voter (Figure 1). Voters are always required to rank each candidate in order of their most to least preferred candidate. A sincere voter would mark the ballot paper identically under each system. Additionally, the five methods were chosen for comparison because, as ordinal voting systems, they run according to the same majoritarian logic the methods are not concerned with proportionality (as is PR) or the intense support of a plurality (as is plurality voting). Instead the methods seek to elect the candidate that is most acceptable to a majority of voters. Figure 1: The format of the ballot paper under all five selected ordinal voting systems The first voting system modelled is the system actually used in South Australian House of Assembly Elections: the Alternative Vote (AV). In AV, vote counting is conducted in rounds. In the first round, all ballots are distributed amongst the candidates according to the first preference expressed on the ballot. If one candidate has more than 50% of the total vote, that candidate is elected. If no candidate has 50% of the total vote, the candidate with the fewest votes is excluded and votes for the excluded candidate are then re-allocated to other, non-excluded candidates, according to the second preferences on each ballot. If still no candidate has achieved 50% of the vote, a further candidate is excluded, with this process repeating until one candidate achieves 50%. The second voting method considered, Coombs Method, was proposed by Clyde Coombs in 1964 but is currently not employed anywhere for elections for public office (Coombs 1964, Grofman and Feld 2004). Coombs Method can be thought of as AV in reverse because the only difference between AV and Coombs Method is the order in which non-elected candidates are excluded. Whereas, if after the first round of counting no candidate has a majority of votes, AV excludes the candidate who has received the fewest votes, Coombs Method excludes candidates by distributing all votes according to the least preferred candidate on the ballot and eliminates the candidate with the most last preferences (even if they simultaneously have the most first preferences). The excluded candidate s preferences are then re-distributed per AV and a second round of counting occurs. This continues until a candidate with greater than 50% of the vote is found, that candidate being elected. Thirdly, the Bucklin Method of vote counting is modelled. Bucklin voting has its origins in the American West during the Progressive Era (1890-1920). Like AV, it is also explicitly majoritarian: James W. Bucklin devised the system as part of a plan to restore majority elections and true representative government (Bucklin 1911, p. 90). Bucklin voting spread across the US West for local city elections; several US states including Colorado, Washington, Minnesota and Oklahoma used it in the early 20 th century. As the Progressive Movement began to decline, the Bucklin Method was discarded in favour of plurality voting (Bucklin 1911, Richie 2004). As with AV and Coombs Method, the Bucklin Method proceeds in rounds and in the first round all ballots are distributed amongst the candidates according to first preferences. If no candidate has achieved 50% of the vote, all candidates receive an additional vote for each second preference they received. Unlike AV and Coombs Method, these second preference votes are added to the initial vote tally and no candidates are eliminated. If any candidate has now received greater than 50% of the vote, that candidate is elected. If two (or more) 6

candidates achieve greater than 50% of the original vote in the same round, the candidate with the largest vote wins (Dabagh, 1934). If no candidate has yet received 50% of the vote, all candidates receive a further vote for any third preferences they receive, and so on, until a winner is found. The fourth vote counting method considered is the Borda Count. The Borda Count is a points based method of counting named after its inventor, French mathematician Jean-Charles de Borda in 1770. Borda counting (or a close variant thereof) is used in parliamentary elections in Nauru and Kiribati (Reilly, 2002). For an election featuring N candidates, each candidate receives points from every ballot, N-1 points for each 1 st preference, N-2 points for each 2 nd preference, and so on down to 0 points for each N th preference. Once points from all ballots are allocated, the candidate with the most points is declared the winner (Black 1976). The fifth and final ordinal method of vote counting considered is a form of Condorcet Voting. Condorcet Voting was proposed by Marquis de Condorcet, another 18 th Century French mathematician (Williams 2004, pp. 206-212). In Condorcet Voting every possible combination of two candidates are compared, in each case all ballots are distributed between the two candidates according to preferences, and the candidate with the larger vote tally is declared the winner of that match up. In Australian parlance, a match-up is the two candidate preferred result between any two candidates, with every possible candidate pairing considered. If a candidate exists who wins every pairwise match up, that candidate is the Condorcet winner and is elected. 3 The only known instance of Condorcet Voting being employed in an election for public office was for city elections in Marquette, Michigan in the 1920s (Hoag and Hallett 1926). For the purposes of interpreting the outcomes under the five ordinal systems in the next section, we assume voters in South Australian elections from 1997 to 2010 expressed sincere preferences under the AV system, thus enabling their preferences to be taken at face value and the modelled results to be reliably presented and interpreted. Later in the paper we consider the consequences if those preferences are in fact not sincere, but instead strategy driven. RESULTS: SIMILAR VOTING SYSTEMS, DIFFERENT LEGISLATURES Despite the similarity of each ordinal method, when modelled using South Australian electoral data, the five ordinal methods deliver very different results. Indeed, the modelled outcomes indicate that outcomes in legislative elections are highly sensitive to the chosen ordinal voting method and that AV is unique in its proximity to plurality. The aggregate number of won by candidates of the various political parties contesting the modelled 1997, 2002, 2006 and 2010 South Australian House of Assembly elections under the different voting methods are presented in Tables 1-4, discussed below. 3 It is possible that votes can be distributed so that every candidate loses at least one match up and Condorcet Voting requires a tie breaker method for such scenarios. Dozens of different methods exist to resolve ties far too many to describe individually here. For this article the Schulze method, proposed by Schulze (2003), has been used. The particular choice of the Schulze method was largely arbitrary, chosen as the most recently formulated of the methods meeting Condorcet s criteria. 7

Table 1: Seats Won by Political Party, Modelled Results by Count Method, 1997 South Australian House of Assembly Election Alternative Coombs Bucklin Borda Count Condorcet Plurality Party Vote Method Method Voting ALP 21 13 10 9 12 16 Liberal 22 13 13 7 12 31 National 1 1 1 1 1 0 Independent 2 4 6 6 6 0 Democrats 1 15 16 23 16 0 United Australia Party 0 1 1 1 0 0 Two Major Parties* 43 26 23 16 24 47 Percentage of won Two Major Parties* 91 55 49 34 51 100 Table 2: Seats Won by Political Party, Modelled Results by Count Method, 2002 South Australian House of Assembly Election Alternative Coombs Bucklin Borda Condorcet Party Vote Method Method Count Voting Plurality ALP 21 18 10 8 17 21 Liberal 21 17 14 11 15 23 National 1 1 2 1 1 1 Independent 4 5 6 8 6 2 Democrats 0 2 9 7 6 0 Greens 0 3 3 1 0 0 South Australia First 0 1 1 3 1 0 Family First 0 0 2 8 1 0 Two Major Parties* 42 35 24 19 32 44 Percentage of won Two Major Parties* 89 74 51 40 68 94 Table 3: Seats Won by Political Party, Modelled Results by Count Method, 2006 South Australian House of Assembly Election Party Alternative Vote Coombs Method 8 Bucklin Method Borda Count Condorcet Voting Plurality ALP 28 27 22 19 24 29 Liberal 15 14 11 8 12 15 National 1 1 2 3 2 1 Independent 3 1 0 1 1 2 Democrats 0 1 2 5 3 0 Greens 0 2 5 1 2 0 Dignity for the 0 1 2 6 2 0 Disabled Family First 0 0 3 4 1 0 Two Major Parties* 43 41 33 27 36 44 Percentage of won Two Major Parties* 91 87 70 57 77 94

Table 4: Seats Won by Political Party, Modelled Results by Count Method, 2010 South Australian House of Assembly Election Party Alternative Vote Coombs Method Bucklin Method Borda Count Condorcet Voting Plurality ALP 24 20 13 11 21 25 Liberal 20 18 18 17 19 20 National 0 0 0 1 0 0 Independent 3 3 3 3 3 2 Democrats 0 0 0 1 0 0 Greens 0 2 11 3 2 0 Dignity for the 0 0 1 1 0 0 Disabled Family First 0 0 1 7 0 0 Save the RAH 0 3 0 3 2 0 Two Major Parties* 44 38 31 28 40 45 Percentage of won Two Major Parties* 94 81 66 60 85 96 Looking at the results across the four elections, three closely related observations ought to be made. Firstly, in all modelled elections AV and plurality produce very similar Houses (Table 5). With the exception of 1997, in which plurality and AV are noticeably different, the AV result differs only slightly from plurality, seemingly confirming Rae s assertions in 1971 that AV does behave like plurality for all (or at least most) intents and purposes. AV s tendency to produce results similar to the plurality results is largely due to the characteristics that it shares with plurality, namely of rewarding intense support (in AV s case, first preferences) whilst ignoring intense opposition (last preferences). Table 5: Difference between AV and Plurality (P) Results, Seats Won by Political Party, 1997, 2002, 2006 and 2010 South Australian House of Assembly Elections 1997 2002 2006 2010 Party AV P Difference AV P Difference AV P Difference AV P Difference ALP 21 16 5 21 21 0 28 29 1 24 25 1 Liberal Party 22 31 9 21 23 2 15 15 0 20 20 0 National Party 1 0 1 1 1 0 1 1 0 0 0 0 Independent 2 0 2 4 2 2 3 2 1 3 2 1 Democrats 1 0 1 0 0 0 0 0 0 0 0 0 Other 0 0 0 0 0 0 0 0 0 0 0 0 Secondly, the four non-av ordinal methods of vote counting tend to elect multiparty legislatures with both a greater number and a more diverse array of parties gaining representation. Candidates from new, reactionary and issues based parties are elected under the other methods, thus non-av ordinal voting methods would present a challenge to the existing dominance of the two established major parties (the ALP and the Liberal Party). As shown in Figure 2, plurality and AV produce Houses with candidates from the fewest parties (2.50 and 3.00 parties respectively). The Bucklin Method and Borda Count produce legislatures with the most parties, with an average of 6.00 and 6.75 parties respectively. 9

parties elected Figure 2: Political Parties elected by Modelled Count Method, 1997, 2002, 2006 and 2010 South Australian House of Assembly Elections 9 8 7 6 5 4 3 2 1997 2002 2006 2010 1 0 Alternative Vote Coombs Method Bucklin Method Vote Counting Method Borda Count Condorcet Plurality It is conspicuous that, by merely changing the criterion for eliminating unpopular candidates from least first preferences to most last preferences, AV and Coombs Method elect very different legislatures. Coombs Method produces an average of 5.25 parties in the legislature and elects candidates from the Greens (2010, 2006, 2002), the Democrats (2006, 2002, 1997), Save the RAH (a single issue party, contesting its first election, formed in response to plans to rebuild the main Adelaide hospital at a different site) (2010), Dignity for Disabled (a party contesting its first election on platform of disabled people s rights) (2006), South Australia First (a party formed in 1999 by ejected ALP MLC Terry Cameron) (2002) and the United Australia Party (a splinter group from the Liberal Party, contesting their first election) (1997) None of these candidates would have been elected under AV (with the exception of one of the Democrats candidates in the modelled AV results for 1997 4 ). The third important observation is that all the non-av ordinal methods produce more varied results over the 4 years modelled. This phenomenon is due to the non-av ordinal methods greater sensitivity to voters ordering of their preferences than AV. As represented in Table 6, AV produces the least variable results, with the mean difference between the maximum and minimum number of candidates from a political party elected in the studied elections of just 4%. By contrast, the mean difference for plurality is 10% and for the non-av ordinal systems between 13% and 20%. This demonstrates that the other voting methods are more subtle than AV. They can give voice to changes in voter sentiment that AV misses by making greater use of a voter s full preference list. In different ways, each non-av 4 We note that the election of the Democrats candidate did not actually occur in the real AV election and is an artifact of the approximations made in our model. Refer to Hargreaves and John (2011) for a discussion of these approximations and their limitations. 10

method of ordinal voting gives greater meaning to second and subsequent preferences and changes in the ordering of these such that whether voters place the Democrats candidate second or third on the ballot paper makes a difference to the results generated. In AV, second or later preferences are utilised only after the voter s most preferred candidate is eliminated (which rarely happens to the major party candidates) in essence, second or later preferences typically only have meaning when they are redirected to one of the major party candidates. Indeed, the common Australian parlance of directing preferences to Labor or Liberal itself reveals the fate of minor party first preferences under AV. However, for most voters, their second (and subsequent) preferences are never given any meaning by AV. Table 6: Maximum and Minimum Candidates Elected from each Political Party by Count Method, 1997, 2002, 2006 and 2010 South Australian House of Assembly election Party Alternative Vote Coombs Method 11 Bucklin Method Borda Count Condorcet Voting ALP Minimum 21 13 10 8 12 16 Maximum 28 27 22 19 24 29 Difference 15.0% 30.0% 26.0% 23.0% 26.0% 28.0% Liberal Party Minimum 20 13 11 7 12 15 Maximum 22 18 18 17 19 31 Difference 4.3% 10.6% 14.9% 21.3% 14.9% 34.0% National Party Minimum 0 0 0 1 0 0 Maximum 1 1 2 3 2 1 Difference 2.1% 2.1% 4.3% 4.3% 4.3% 2.1% Independent Minimum 2 1 3 1 1 0 Maximum 4 5 6 8 6 2 Difference 4.0% 9.0% 6.0% 15.0% 11.0% 4.0% Democrats Minimum 0 0 0 1 0 0 Maximum 1 15 16 23 16 0 Difference 2.0% 32.0% 34.0% 47.0% 34.0% 0.0% Greens Minimum 0 0 0 0 0 0 Maximum 0 3 11 3 2 0 Difference 0.0% 6.0% 23.0% 6.0% 4.0% 0.0% Other Minimum 0 1 1 1 0 0 Maximum 0 3 5 11 3 0 Difference 0.0% 4.0% 9.0% 21.0% 6.0% 0.0% Mean Difference 3.9% 13.4% 16.7% 19.6% 14.3% 9.7% Plurality As a consequence of their sensitivity to preference order, the non-av ordinal methods tend to reward moderate (as opposed to polarising) parties and candidates. It should be noted here that the Democrats a party whose raison d'être was moderation elect candidates in every non-av ordinal method in 1997, 2002 and even in 2006 (by which time the party had been declared dead and buried by commentators). In 1997, the non-av ordinal methods elect between 15 and 23 one short of a majority Democrats candidates (Table 1). The 1997 result reflects both a high (for a minor party) first preference count (16.45% state-wide) and the fact that the Democrats often ranked significantly higher in major party voters preference orderings than did the other major party. By contrast, One Nation,

dubbed one of Australia s pariah parties because both major parties tended to encourage their voters to put it last in their preference listings (Green 2010), do not elect any candidates under any of the methods. Additionally, the Greens in 2010 in 2010 Australia s most significant minor party, but a less moderate and more polarising party than the Democrats do not do as well in any year as the Democrats in 1997 relative to the proportion of first preferences they received, except under the Bucklin Method. The Bucklin Method rewards second preferences handsomely and this property is reflected in its tendency to elect Greens candidates. In 2006, The Bucklin Method would elect five Greens whereas the other non-av ordinal methods elect only one or two (Table 3). We speculate that this is because the Greens were frequently the second preference of ALP voters, and the ALP had an unusually high first preference count in 2006 (45.2% state-wide) relative to the other years modelled (2010: 37.5%; 2002: 36.3%; 1997: 35.2%) (ECSA 2011, p. 26). Similarly, in 2010 the Greens would have (almost) joined the ranks of a major party, electing 11 candidates compared to 13 for the ALP (Table 4). Taken together, these results indicate that, contrary to indications in earlier organisational modelling studies (Felsenthal, Maoz and Rapoport 1993, Felsenthal and Machover 1995, Levin and Nalebuff 1995, Regenwetter et al 2007), when applied to legislative elections ordinal methods vary significantly in how they transfer votes into. They highlight that all ordinal voting methods are neither identical nor equal. Thus, we conclude that the ordinal voting system chosen really matters to the constitution of the legislature. DISCUSSION: INSTITUTIONS AND POLITICAL CULTURE Assessments could be made using the above results about the relative merits of the different ordinal systems using social choice criteria such as disproportionality, representation of minorities, stability, responsiveness to changes in voter sentiment and so on. For example, there is a clear argument to be made, which deserves far more attention than space here permits, about the superiority of non-av ordinal methods in terms of responsiveness and the facilitation of emergent and new political parties and movements. Our primary focus is on the consequences that AV might have for Australian political culture. Australian political culture is identifiable by the absolute domination of political discourse by the two major sectional political parties (and the smaller contribution of the National Party). As noted by Collins (1985), Australian political culture is, and has been for many decades, unusual for the narrow scope of politics and the nature of its contest being unashamedly utilitarian and interests based. Within this culture, political parties make little resort to ideals and ideas to clothe their naked intent (Collins 1985, p. 155; Hartz 1964, Hughes 1973). There is a neat compatibility of AV with compulsory preferencing and the pragmatism of a utilitarian political culture seeking to manufacture a majority and ensure that voter choices do not interfere with the two sectional parties control over the system. The rowdiness of Australian politics and the antagonistic relations between the major parties, at least in public, in which often complex issues and ideas are bifurcated into two simple and diametrically opposed positions, is an observable Australian phenomenon. Efforts at approaching problems in an evolutionary, cross-partisan fashion are extremely rare. While it may be tempting to romanticise about the rowdiness of Australian politics, we must be mindful to juxtapose the rowdy, antagonistic nature of politics and the aptness of AV in this environment against the overwhelmingly negative views held by 12

Australians about their political institutions (Goot 2002, Bean 1993). If AV exacerbates tendencies in Australian political culture that contribute to disenchantment with the political system, alternatives ought, at the very least, to be considered. To assess whether AV is likely to have had an impact on the development of the caustic tendencies in Australia s political culture, we consider the observed unique characteristics of AV using two, mutually exclusive, assumptions: (1) that preferences expressed on voters ballots are sincere; or (2) that preferences expressed are strategic and would change according to the incentives offered by individual voting systems. For clarity s sake, in sincere votes the preferences expressed on the ballot are the genuine order in which the voter would like to see the candidates elected to office and the preference order would remain the same across all ordinal voting systems. In strategic votes, the preferences on the ballots do not give an honest indication of the voters preference for each candidate, but instead reflect rational choice type strategic calculations by the individual voter, or the party to which he or she is loyal, about maximising the candidate or party s chances for election and/or government. Under the first assumption, of sincere preferencing, the present research is indicative of electoral outcomes under the various ordinal voting systems. Under all the non-av ordinal voting systems, a greater number of parties gain representation (Figure 2) and one party does not command a majority of. A majority is returned only in 2006 (Table 3) and no fewer than four parties were elected in any of the non-av ordinal methods (Figure 2). It is notable that such multiparty legislatures are most commonly found under PR systems, and we therefore surmise the Australian legislatures could potentially assume some properties commonly associated with PR legislatures under a non-av ordinal voting method. (However, it should be noted that none of the non-av ordinal methods produce results proportional to first preferences that is neither their aim nor their effect.) It has been long established that PR changes the style of political parties and the nature of competition between them, in large part due to the multiparty nature of PR legislatures (Norris 2004, Katz 1980). Multiparty legislatures encourage compromise and bargaining since it is difficult for individual parties to assert outright control. Coalition-building skills become important if a party wants to succeed in its legislative goals and these skills are diametrically opposed to the adversarial and aggressive debating skills presently prized under AV. The absence of coalition-building skills presently in Australian politics is evident in the recent tumult caused by election of minority federal government in 2010 and also in the unified nature of the Liberal-National Party coalition, in which very little dissention is permitted. Thus, we argue that, if preferences are sincere, Australian parliaments and politics would necessarily become less intensely adversarial under a non-av ordinal voting system. Such a system would reward those who can master the arts of compromise and bargaining, unlike AV, which rewards extreme and rigid partisanship, combativeness and antagonism. Additionally, the greater sensitivity of the non-av ordinal methods to subtle changes in preference order would ensure parties could be less complacent about alienating other parties voters because whether a party s candidate is preferenced last or second to last really matters to the outcome under non-av ordinal systems. Under the second assumption, that the preferences expressed by voters and especially those voters following their political party s HTV cards are strategic, preferences are dependent on the characteristics and technicalities of the voting method employed. To the rational and strategic voter, intent on realigning their preferences to increase the chances of their most preferred candidate or party winning, the unique characteristics of AV are central. AV is the only ordinal voting method studied that does not penalise candidates who receive many last preferences; it is also unusual for being largely 13

insensitive to the exact order of preferences. Coombs Method, the Bucklin Method, the Borda Count and Condorcet Voting all make greater use of second and subsequent preferences. Shortly summarised, AV is the least subtle of the ordinal voting methods and thus requires the least subtlety in voter strategy. As a consequence of this lack of subtlety, AV encourages a strategic major party voter to preference the other major party last, which it gives voice to and normalises hostile partisan feelings and a tendency to view the other major party as the enemy, thus enabling a highly corrosive party dialogue. In all non-av ordinal methods, a strategic major voter and especially a strategic party trying to guide their party s voters would reconsider their common current strategy of preferencing the other major party last or near last. All other ordinal methods incentivise minimisation of last and near last preferences. In order to ensure the major parties each remain major, they would need to convince their voters to preference the other major party high up on the preference list, in the hope the other major party s voters would reciprocate. An individual voter, operating without guidance of party would not necessarily feel an imperative to change his or her preference order, but the major parties certainly would. If both parties agree and were successful in convincing their party s voters to follow this strategy, then the electoral results returned under the ordinal systems could be similar to AV and to plurality and both parties could expect to take it in turns controlling the legislature. Such an arrangement implies some Katz and Mair style cartel behaviour, but this is not beyond the realms of possibility (Katz and Mair 1995). While it may be possible to strategically defeat the multiparty tendencies of non-av ordinal voting systems, political culture and behaviour may not be so unaffected. Because the other voting systems all create powerful incentives on the rational major political party to encourage their voters to preference the other major party highly, a caustically adversarial politic in which each major party behaves as though the other is the vilest political force in existence might seem somewhat incongruous. Additionally, voters would have more power to determine the fortunes of both major political parties with their preference ordering. The major parties would need to be careful not to isolate voters of the other major party (especially under the Coombs Methods) for fear of receiving last or near last preferences. Publicly, a major party in discourse and in its HTV card could no longer declare the other major party to be the enemy. An admission that there is more common ground than divisive would need to be made. This in turn would place pressure on parties to tone done the harsh adversarial tone of politics lest the major parties voters believe the bumf and preference each other last. Under either assumption whether the ordinal votes in South Australian House of Assembly Elections are sincere or strategic a non-av ordinal voting system would provide pressure for the evolution of less adversarial and corrosive political culture over time. While the constitution of the legislature changes under each assumption, the net pressure for a gentler, more constructive political culture does not. These findings accord with McAllister and Farrell s (2006) conclusions about AV s tendency to support a high level of party control of the system: under non-av ordinal systems either there would be less domination by one or two political parties over the legislature or voters would be more greatly empowered to determine the fortunes of the two major parties. The paper s conclusion about AV s abrasive tendencies contradicts with Reilly s findings about the moderating and consensual impact of AV (Reilly 2001a, Reilly 2001b), but only because AV is compared against its natural cousins other ordinal methods rather than plurality, which aims to achieve something very different to that which ordinal methods aim to achieve. 14

CONCLUSION The results presented for South Australian House of Assembly elections 1997 2006 using five different ordinal methods conflict with previous empirical studies that observe that electoral outcomes are not closely related to the ordinal voting method employed. The results indicate that electoral outcomes vary significantly depending on the voting method employed. The modelled results of South Australian elections show that, if voters are assumed to be sincere in how they mark ballot papers, election results are very dependent on the subtle and almost arbitrary differences between ordinal methods. The findings confirm that AV behaves in most respects as if it was plurality. Additionally, AV is observed to be unique amongst ordinal voting systems in the extent to which it behaves not like an ordinal method of vote counting but a plurality method. AV is also unique in its tendency to direct ordinal votes into for the two established major parties and prevent the success of new movements and parties, especially moderate parties. Additionally, AV demonstrates a distinctive lack of sensitivity to changes in voters preference orderings. Tersely stated: AV is the least subtle of the ordinal voting methods examined. Far from producing the more civil politics Britons mused about during their debate about AV, AV with its tendency to squash uprising new and moderate political movements likely emboldens and exacerbates an adversarial political culture by allowing opposing major parties in the legislature to paint each other extremely negatively. AV facilitates the practice of mutual last or near last preferencing by the irrelevance of last and near last preferences to the AV counting method, which helps support and buttress an adversarial political culture. The tendency of major party voters to preference the other major party last or near last encourages a view of the other major party as the enemy. After a time, enough practice at preferencing the other major party last, combined with a harsh partisan ethic that paints the other major party as the arch nemesis, surely takes its toll on the mindset of a major party voter. Combined with the two-party dominance in the legislature (itself aided by AV), which makes coalition building and negotiation almost irrelevant, AV reinforces the adversarial and unrestrainedly partisan nature of Australian political culture. No other ordinal method possesses the caustic qualities of AV even though they all have same underlying goal and require identical information from voters. Whether voters are assumed to be honest and sincere or more sophisticated and strategic, every ordinal voting method other than AV would lead to a less adversarial and antagonistic politic. Sincere voters would elect more multi-party and coalition based legislatures, which require greater negotiation and relationship building skills in legislators. Strategic major party voters would no longer be able to preference the other major party last, thus creating a normative impetus toward viewing the other major party not as the arch nemesis, but as a potential partner or ally. AV is a poor choice amongst ordinal voting systems for increased civility in politics. Relative to other ordinal methods, AV incentivises aggressive partisanship, or at the very least it does not de-incentivise it. Indeed reformers looking for a majoritarian system that is likely to develop a more considered politics ought to stay away from AV and consider other ordinal voting methods (or perhaps even cardinal methods). However, political scientists looking to explain the continued corrosive character of Australian politics might look further into the role that AV has played. 15

BIBLIOGRAPHY BBC News (2011), John Howard explains Australian voting system (5 April 2011), UK Politics Section, http://www.bbc.co.uk/news/uk-politics-12974935 (accessed 29 April 2011) Bean, C. (1997), Australia's Experience with the Alternative Vote, Representation, 34(2), 103-110. (1993), Conservative Cynicism: Political Culture in Australia, International Journal of Public Opinion Research, 5(1), 58-77. (1986), Electoral Law, Electoral Behaviour and Electoral Outcomes: Australia and New Zealand Compared, Journal of Commonwealth and Comparative Politics, 24(1), 57-73. Black, D. (1976), Partial Justification of the Borda Count, Public Choice, 28(1), 1-15. Blais, A (ed). 2008, To Keep or To Change First Past the Post? The Politics of Electoral Reform, Oxford University Press, Oxford. (1991), The Debate over Electoral Systems, International Political Science Review, 12, 239-260. Brischetto, R. (1995), Cumulative Voting as an Alternative to Districting: An Exit Survey of Sixteen Texas Communities, National Civic Review, 84, 347-54. Bucklin, J.W. (1911), The Grand Junction Plan of City Government and Its Results, ANNALS of the American Academy of Political and Social Science, 38(3), 87-102. Cain, B.E., Donovan, T. and Tolbert, C.J. (2008), Democracy in the States: Experiments in Election Reform, Brookings, Washington, D.C. Chen, F.H. and Heckelman, J.C. (2005), Winning Probabilities in a Pairwise Lottery System with Three Alternatives, Economic Theory, 26(3), 607-617. Coombs, C.H. (1964), A Theory of Data, John Wiley and Sons, New York. Dabagh T.S. (1934), The Vote-and-a-Half System of Majority Choice by Plurality Vote, and of P.R., American Political Science Review, 28(1), 93-101 Donovan, T. and Bowler, S. (2004), Reforming the Republic: Democratic Institutions for the New America, Prentice Hall, Upper Saddle River, N.J. ECSA (Electoral Commission of South Australia) (2011), Election Statistics: South Australian Elections 20 March 2010, ECSA, Adelaide, <http://www.ecsa.sa.gov.au/apps/news/?sectionid=12&pageid=625> (accessed 4 July 2011). Economist Online (2011), Economist Debate: Alternative Vote, Economist, London, <http://www.economist.com/debate/days/view/688> (accessed 4 July 2011). 16