Geography and election results: disproportionality and bias at recent elections to the Australian House of Representatives RON JOHNSTON1* and JAMES FORREST2 1 School of Geographical Sciences, University of Bristol, Bristol BS8 1SS, UK 2 Department of Human Geography, Macquarie University, Sydney, NSW 21089, Australia * Corresponding author: Email: [email protected] This paper has been submitted for publication NOT TO BE CITED WITHOUT THE AUTHORS’ PERMISSION

Geography and election results: disproportionality and bias at recent elections to the Australian House of Representatives Abstract The Alternative Vote system used for elections to the Australian House of Representatives is generally believed to disadvantage the Australian Labor Party in its contests with the Liberal and National parties. However, most of the analyses on which such conclusion are based over-simply the situation by not separating out the translation of votes into seats according to whether the election outcome in a district is determined using the first-preference or two-party preferred (2PP) votes. Analyses of bias at five recent elections which do recognise that separation find that there is little bias against either party in the districts where the determination is using the 2PP votes, but that there is considerable bias in those where the outcome is decided on the first-preferences. Furthermore, that bias is not in one direction, but rather favours the largest party in those contests. The reason for this is identified in the geography of support for the two parties, which produces the equivalent of a ‘cracked gerrymander’ in sufficient districts to have a significant impact on the outcome. KEY WORDS

electoral bias; Australia; House of Representatives,

Australia’s electoral systems are virtually unique in the democratic world: no other country uses the Alternative Vote (AV) system deployed for the Commonwealth House of Representatives, six of the State Lower Houses, and one State Upper House, and very few use the Single Transferable Vote (STV) system adopted for the Commonwealth Senate, two of the State Lower Houses, and four of the State Upper Houses. This uniqueness is frequently remarked upon, and is the subject of detailed examinations (as in Farrell and McAllister, 2006), but it means that Australia is difficult to place in comparative studies – especially those of elections to the lower houses of national legislatures. Two features of the operation of single- member constituency electoral systems, which Gudgin and Taylor (1980, 516) associated with ‘the territorial basis of this form of electoral law’, are disproportionality and bias: do the systems treat the parties disproportionally in the translation of votes into seats, and is this disproportional treatment biased in that it favours one party more than another? Several attempts to answer this question were made in Australia a few decades ago – starting with Rydon’s (1957) pioneering paper (see also Soper and Rydon, 1958) – leading to a general conclusion that the operation the AV system for electing the House of Representatives was biased in its outcomes against the Labor Party (Taylor and Gudgin, 1977; Gudgin and Taylor, 1979) – more so than a similar bias operating against Labour in both New Zealand and the UK. Indeed, Taylor and Gudgin (1977, 118) concluded that the extent of this bias ‘… means condemning the ALP to a

2

likelihood of many more years in opposition than their British and New Zealand comrades’. With one exception (Jackman, 1994: see below), little has been written on this subject in recent decades – and it gets no more than a very brief ment ion in Farrell and McAllister (2006). And yet, after losing the general election in 1980, the ALP won the next five contests, four of them by substantial majorities over the Liberal-National coalition (the exception was 1990). What has changed since the earlier findings of an electoral system biased against the Labor Party? In the UK, too, after a run of defeats the Labour party not only won three elections in a row (1997, 2001, 2005) but the biases in the British electoral system which largely worked against it until the 1960s, followed by a period when there was no substantial bias against either party, very much favoured Labour over its Conservative opponent from 1997 on (Johnston et al, 2001; Johnston, Rossiter and Pattie, 2006; Johnston, Pattie and Rossiter, 2008). The time seems right for a re-evaluation of the Australian case. This paper essays that task, presenting analyses of the results of five recent elections to the House of Representatives (1993-2004) using a different method to that deployed by Rydon, Soper, Gudgin and Taylor – although based on the same core arguments; our chosen method has a much more interpretable metric. It also explores the particular features of the Australian AV electoral system which make the measurement of disproportionality and bias less than straightforward. In the only recent study of disproportionality and bias in Australian election results, Jackman (1994, 329) also concluded – from analyses of all 19 elections between 1949 and 1993, for both federal and state elections – that there was still a moderate antiLabor bias. However, if the elections were split into periods, that bias only applied prior to 1984; between then and 1993 the bias favoured the ALP (i.e. it got a greater share of the seats than of the votes). In this work, Jackman (like earlier analysts such as Soper and Rydon) used share of the two-party preferred vote total (2PP) rather than share of the first preference votes cast as his criterion against which disproportionality and bias could be assessed. (The 2PP total is the outcome when all of the preferences have been allocated under the AV system, so that only two parties remain in contention in each district. Those preferences are allocated in each district irrespective of whether this is necessary to determine the result, which in many districts is based on the first preference votes because one party’s candidate obtains a plurality of the votes cast at that stage. See Rydon, 1986, for a discussion of why use of the 2PP data is considered a superior approach, although she does conclude that ‘voting figures do not lend themselves to precise or exact mathematical analysis and … there is no one simple method of handling election results’ – p. 73. 1 ) Additional analyses extended this finding and Jackman (1994, 347) concluded that because of biases in the operation of the system, Labor needed to win about 55 per cent of the 2PP votes in 1

Jackman (1994, 323) claims that in studying the operation of Australia’s electoral system ‘One complication is how to determine a party’s electoral strength under the alternative vote. … The number of first preference votes a party receives is often a misleading indicator in this respect, since seats are decided after distributing preferences’. He adds, in a footnote, that ‘Though, of course, the alternative vote reduces to simple majority rule when only two candidates contest a seat’. This ignores the fact that the result is not determined in all districts by the allocation of preferences: in a substantial number – a majority at some elections – the result is determined on the first preferences because one candidate has a plurality of the votes cast. The full set of 2PP votes has been published for every election since 1983, following a recommendation of the Joint Select Committee on Electoral Reform.

3

1950 to get a legislative majority. However, this figure reduced over the subsequent decades so that by the early 1990s the re was no anti- ALP bias. This paper carries the story forward, starting with the last election in the sequence analysed by Jackman, using an alternative procedure, and taking into account the split between those districts where the result is determined by the first-preference votes and those where the 2PP totals are used. This split is important in order to take into account the number of seats won or lost on the first preferences as well as those where the determination is made on the 2PP votes. Because of this decision to investigate biases with first-preference as well as 2PP votes, the paper also looks separately at the two parties (Liberal and National) which have governed as a coalition during the period but both contest some of the districts (almost invariably where a sitting member of the coalition is not contesting the seat). Disproportionality in the election results To establish the degree of disproportionality at elections to the Commonwealth House of Representatives, Tables 1 and 2 present the 1993-2004 results in a number of ways, using seats:votes ratios (a party’s percentage of the seats allocated divided by its percentage of the votes cast) as the key indicator. A ratio of 1.0 indicates proportional representation – a party gets a share of the seats commensurate with its share of the votes; a ratio greater than 1.0 indicates that it gets a greater share of the seats than votes (i.e. over-representation), whereas a ratio of less than 1.0 indicates underrepresentation (a smaller share of the seats than votes). The top half of Table 1 gives these ratios – and the vote and seat shares on which they are based – at each of the five elections, using first preference votes as the indicator of ‘most preferred party’ ; in this, data are provided both separately and combined for the Liberal and National parties. These data show a fairly consistent pattern of support for the three main parties across those five elections, with the Liberal-National Coalition (LNC) parties getting 39-47 per cent of the vo tes and the Australian Labor Party (ALP) getting 38-45 (and between 37.6-40.1 only at the last four contests). The greatest variation is for the ‘Other’ parties, whose support ranged from a maximum of 20.4 per cent to a minimum of just over half that (10.5). Within the coalition, the Liberal Party won between 34 and 41 per cent of the votes, and National between 5 and 8. Of particular note is the 1998 result, when the ALP obtained more votes than the coalition parties combined but a substantially smaller share of the seats. Turning to the seats:votes ratios, the main feature is the difference between the 1993 election, won by the ALP, and the other four, won by the LNC. At the first, the ALP enjoyed substantial over-representation whereas the coalition as a whole had a ratio of exactly 1.0, largely as a consequence of a 50 per cent over-representation for the smaller of the two member parties (a ratio of 1.51). From then on the coalition enjoyed significant – though declining somewhat – over-representation for both of its members, whereas the ALP experienced a ratio below 1.0 in 1996 and smaller ratios than those for the LNC at the subsequent three elections. As is commonly the case in elections to legislatures using single- member constituency systems, therefore, the winning party (or coalition in this case) achieves a substantial ‘winner’s bonus’ whereas the smallest parties are very substantially under-represented (a seats:votes

4

ratio as low as 0.10 in 2001 when the ‘Other’ parties together obtained 19.2 per cent of the votes cast but only 3 of the 150 seats in the House of Representatives). Although first-preference votes cast under the AV system indicate their mostpreferred party for the great majority of voters, nevertheless for most they do not lead to its candidate being elected there. In at least forty per cent of the districts at the five contests studied here, and nearly two-thirds in 1998, second and lower preferences were counted to provide the 2PP totals (which indicate in almost all districts – the exceptions are those won by an independent candidate – which of the two main ‘blocs’, ALP and LNC, is preferred for government ). Because 2PP data are available for every district, irrespective of whether this allocation procedure was necessary as no candidate crossed the threshold (50 per cent + 1) for election on the first preferences, it is possible to evaluate the degree of disproportionality in the allocation of seats against those final figures (votes for the LNC parties are combined here because they are, in effect, no longer distinct entities). 2 The lower block of Table 1 gives these data. Here, the winner’s bonus pattern is even clearer: in 1993 the ALP had a seats:votes ratio exceeding 1.0 whereas the LNC had a ratio below 1.0. The situation was reversed at each of the next four elections – won by the LNC (although again, as with the first-preference votes, in 1998 the ALP defeated the coalition in terms of votes but lost by a substantial margin in the allocation of seats). This suggests that LNC benefited more tha n the ALP at the contests it won: the average LNC ratio of 1.14 across the four elections 1996-2004 indicates twice the level of over-representation to that achieved by the ALP in 1993 (1.07), whereas the average ratio of 0.85 for the ALP over the last four elections is considerably smaller than 0.91 for LNC in 1993. The LNC appears to get a more substantial ‘winner’s bonus’ than the ALP (which suggests a pro-LNC bias, to be analysed below). The Other parties, very few of whose candidates get through to the ‘final round’ of allocating 2PP votes, apparently do very well; this is because they get a small share only of the total number of 2PP votes – because they provide one of the two final candidates in just a few seats – relative to their share of the seats (even though the latter is also small). What if we split the results, into those districts won at the first stage of the AV procedures by a party whose candidate exceeds the (50 per cent + 1) threshold, and those for which the lower preferences are allocated to generate the 2PP totals on which the determination is made? Table 2 replicates Table 1: the top half gives the data for the districts won at the first stage and the bottom half for those won at the second stage. The data in the first block further stress the importance of the ‘winner’s bonus’ in the allocation of seats relative to votes. At the 1993 and 1998 elections, the ALP led the LNC parties in those districts where a winner was declared at the first stage (84 of the 147 districts in 1993 but only 50 of the 148 in 1998); it enjoyed a seats:votes ratio above 1.0 at those two contests, with the Liberal party having ratios below 1.0; the reverse was the case at the other three elections. The National party had a very large ratio above 1.0 at all five contests; the Other parties did not win any seats at the first preference stage at any of the elections. 2

These data were not available for one seat – won by one of the ‘Other’ parties – in 2001.

5

Turning to the districts where the result was determined at the final (2PP) stage, the lower block of Table 2 indicates that the two main ‘blocs’ (LNC and ALP) were never separated by more than three percentage points in the distribution of the 2PP votes. In general, this closeness is matched by the allocation of seats as the seats:votes ratios are much closer to 1.0 for LNC and ALP than they are in the districts where the election was determined at the first stage. For the Other parties, the large ratios in excess of 1.0 at four of the elections reflect the fact that although the 2PP distributions led to their victory in a few districts, in most they were eliminated before the final stage of allocations and so had no 2PP votes in the great majority of districts. 3 The data in Tables 1-2 provide clear evidence of disproportionality in the results of those five elections, whichever vo te total allocation is chosen. But how disproportional? Table 3 gives Gallagher’s (1991) Least-Squares Index of Disproportionality for each of the four blocks of data in Tables 1-2. This index can vary from 0 to 100 (with the latter representing complete disproportionality); proportional representation electoral systems tend to have indices below 5 and Australian elections tend to produce an index closer to those for first-past-the-post systems (Gallagher, 1991; Anckar, 1997). One very clear conclusion can be drawn from Table 3: the Australian AV system produces more disproportional results when applied to the first-preference than to the 2PP votes. This is because in the latter the ‘other’ parties are almost totally excluded; they are one of the last two remaining candidates in only a very small number of districts, so that their under-representation in the first-preferences as shown by the seats:votes ratios is, in effect, excluded from the later analyses. This exploration of disproportionality at five recent House of Representatives’ elections has shown that, in common with the situation in other single- member constituency systems which use the first-past-the-post method, Australia’s AV electoral system also generates a characteristic ‘winner’s bonus’ – irrespective of whether the winner is the Australian Labour Party or the Liberal-National Coalition. However, that bonus is more substantial in districts where the election result is determined at the first stage of counting than at the second stage, after all lower preferences have been allocated. Thus the extent of disproportionality overall – shown in particular in the first block of data in Table 1 – appears to be a function of the proportion of districts where the election is determined at the first stage. This sustains the argument developed here that analyses of bias resulting from operation of the AV system in Australian federal elections should be illuminated by taking the division of seats determined into those where the result is determined on the first preferences and those where the 2PP allocations are deployed into account. Bias might be more prevalent at one stage than the other, suggesting that the proportion of seats determined at the first stage may have a significant influence on the overall outcome. From disproportionality to bias Disproportionality – as illustrated by studies of British general elections since 1950 (Johnston et al, 2001) – can itself be unequally distributed across the various parties 3

Candidates from parties other than the ALP and LNC reached the final stage of the 2PP allocation in only 2 districts in 1993 and in 5, 4, 3 and 4 respectively at the succeeding four elections.

6

contesting an election: in particular, one party may enjoy a larger ‘winner’s bonus’ when it is electorally successful than may another. This suggests that the electoral system is biased, in its outcomes, because it favours one party more than others – or, to use an alternative terminology, the mechanisms that produce disproportionality do so asymmetrically (Grofman and King, 2007; King et al, 2005). Measuring bias Identifying the extent of such bias and the reasons for it has attracted considerable attention from researchers, who have developed a range of procedures for their measurement. All are based on the recognition of two key, geographical (or territorial) variables underpinning the production of disproportionality and bias – variations in district size (either population or number of registered voters) and the distribution of party support across those districts. The importance of these two is reflected in their significance as the two main components of any deliberate manipulation of electoral cartography for partisan gain – malapportionment and gerrymandering (Johnston, 2002). Variations in district size can result in bias – i.e. favour one party over another in a predominantly two-party contest – if one party in a contest tends to be stronger (i.e. more likely to win) in districts that are smaller than the average whereas the other party tends to be stronger in the larger districts – most of the analyses of electoral bias focus on single- member district systems where two parties dominate (in this case LNC and ALP). If, for example, there are two groups of seats, one averaging 50,000 voters and the other 70,000 then, assuming that there are no abstentions and/or informal votes and also that no other parties, only 25,001 votes are needed to win a district in the first group as against 35,001 in the second. Thus a party which is strongest in the first group may get more seats for a given vote total than would one whose main strength is in the second; the former gets a better return on its votewinning because of the geography of where tho se votes are relative to variations in district size. 4 The second main component of electoral bias operates because of variations in what is usually termed the efficiency or effectiveness of the geography of a party’s support across a set of districts. An ideal type example is a party that wins 51 per cent of all of the votes cast. These would be most efficiently distributed if it gained 51 per cent in every district – thus winning them all with a bare majority of the votes cast. The further it is from this ideal, the less efficiently are its votes distributed. Votes can be split into three types: wasted (those which do not win the party any seats, cast in districts that it loses), effective (those which do contribute towards it winning seats), and surplus (those which also do not win the party any seats, cast in districts that it wins by more than the number that are effective). Take a district in which party A wins 15,000 votes and party B wins 12,000. All of B’s 12,000 votes are wasted; 12,001 of A’s are effective, and the other 2,999 are surplus. A party’s goal is thus to minimise its wasted and surplus votes and maximise its effective total – while not placing itself in the situation where a relatively small shift in voter support (from 51 to 49 per cent, for example) could lead it to lose a lot of seats. Thus where it wins it wants to win well but not too well; there is no point in winning by a wide margin and 4

This is illustrated in Chapter 1 of Johnston et al. (2001).

7

piling up lots of surplus votes. Where it loses, on the other hand, it may as well lose badly and not accumulate lots of wasted votes – save in ‘marginal’ districts where there is a chance that it could win at a future election. A method of measuring the amount of apparent bias in an election result and decomposing its sources according to these two major geographies, adopted in a number of recent studies, was developed by a New Zealand political scientist, Ralph Brookes (1953, 1959, 1960). His procedure – modified to incorporate factors not present in his empirical studies of New Zealand (Johnston, Rossiter and Pattie, 1999) – is deployed here. It defines bias very straightforwardly as the difference in the number of seats that two parties would win if they obtained the same share of the votes cast:5 in an unbiased situation, each should get the same number of seats for the same share of the votes cast. To estimate what the seat allocation would be, he applies a uniform shift model to produce a hypothetical election result, generating an equal shares situation. 6 If, for example, party A won 53 per cent of the votes across all districts at the election and party B won 47 per cent, then A’s share would be reduced by three per cent of the total in each district, with those votes being allocated to B; each would then have 50 per cent of the national vote share, and the ‘winner’ of each district could then be calculated. The greater the difference between the two parties in the number of seats they gain, the greater the bias – or asymmetry – in the election result. Brookes’ procedure allows this bias figure to be decomposed into its main contributors: the size and efficiency effects discussed above. The former can be split into three sub-components. The first is variation in district size, as already illustrated. The second is the role of abstentions, which operates in the same way; if turnout varies across districts, one party may be advantaged if its support is concentrated in areas with relatively low turnout whereas its opponent’s is in areas where fewer voters abstain. The third is ‘minor party votes’, which – as with abstentions – may reduce the number of effective votes needed in a constituency for one of the main parties to defeat the other, if one is stronger in districts where the minor party(ies) is too. Of course, if a minor party wins seats, this may disadvantage one of the main parties more than the other. We thus have five components to electoral bias, all of which can be measured using the same metric in Brookes’ procedure – the advantage in number of seats that one party has over the other when they have an equal share of the votes: 1. Electorate size; 2. Abstentions; 3. Minor party votes; 4. Minor party victories; 5. Efficiency. Of these, an equivalent of the abstentions component is very unlikely to be important in the Australian context because of compulsory voting; few people fail to vote, although some deliver a spoilt ballot (termed informal voting). At the 2004 election, for example, the percentage of informal votes (i.e. the number of voters on the electoral roll less the number of valid votes cast as a percentage of the former) 5

This ‘equal shares’ concept was also deployed by Jackman (1994). There has been much discussion on the use of uniform shifts in Australian electoral studies, as discussed in Jackman (1994) 6

8

averaged only 4.6, with a standard deviation of 1.6; in seats won by the ALP it averaged slightly higher (5.4 per cent) compared to seats won by the Liberals (4.8 per cent) and National (4.1), but those differences are very unlikely to have a significant impact on the bias. Similarly, the small number of seats won by minor party candidates is unlikely to have any impact, although their vote shares could; in 2004, for example, their tally averaged 16.2 per cent of all votes cast (though with little difference between the seats won by: ALP – 15.5; Liberal – 14.7; and National – 16.2). It is also unlikely that differences in electorate size will have a major impact given that Australia now has a policy of equal-sized electorates (a different situation from the country quotas that stimulated Rydon’s – 1968 – exploration of ‘Malapportionment: Australian style’). A federal redistribution is undertaken every seven years with a requirement that no district’s population should vary by more than 5 percentage points from the average 3.5 years after the new boundaries have been introduced; this involves using census data plus Australian Bureau of Statistics estimates of change over the next seven years. Table 4 shows variations in the number of voters on the district electoral rolls at each of the five elections studied here. Although there are extremes, because of the minimum number of five seats guaranteed to Tasmania, for example, nevertheless the main feature is the small variation in the great majority of cases – only about 10,000 voters between the 10th and 90th percentile values, for example, and around 5,000 as the inter-quartile range. (For full details on Australia’s redistributions see AEC, 2004.) Finally, what of gerrymandering? Again, there is evidence of past gerrymanders, especially in some states (Queensland and South Australia, for example, as Jackman, 1994, demonstrates), but federal redistricting is a non-partisan procedure which precludes any explicit gerrymandering (Hughes, 1977). It does not, however, prevent the outcome of a redistricting exercise being that one party’s votes are more efficiently distributed than another’s. This could occur, for example, if one party draws most of its electoral support from members of socio-economic groups which are geographically more segregated from the remainder of the population than are the supporters of its main opponent: in this case, the former party would accumulate more surplus votes in its areas of strength and experience bias operating against it. Thus the equivalent of an ‘unintentional gerrymander’ can result from the interaction of the geographies of party support and the geography of the territorial containers which aggregate that support for the translation of votes into seats (as Johnston and Hughes, 1978, demonstrated for Brisbane). Given the small role that the various malapportionment components are likely to make in the Australian context, therefore, it is likely that any bias in the five election results studied here – suggested by Tables 1-3 – is the result of ‘unintentional gerrymanders’ producing different proportions of effective votes for the various political parties. Bias patterns Applying Brookes’ procedure to elections to Australia’s House of Representatives raises a number of important issues. The first relates to the nature of the party system. Brookes’ method is particularly relevant to the study of two-party systems. Australia’s is not quite that, because of the Liberal-National coalition. If the two parties never opposed each other in a district, they could be combined into a single party. But they

9

do compete in some districts: 39, 17, 19, 18 and 9 in 1993, 1996, 1998, 2001 and 2004 respectively. To take this into account, when analysing the first preferenc e votes we have adopted two separate strategies: • Use the largest of the two coalition parties as the ALP’s opponent; • Combine the Coalition parties’ votes to create a hybrid party opposing the ALP. The first of these identifies bias as it affects the largest of the two parties in each district in the usual way; the other assumes that the great majority of supporters of one of the coalition parties would give their second preferences to the other in a straight fight with the ALP. The second issue relates to the two stages of the AV system. As the earlier discussion of disproportionality showed, the (possibly asymmetric) ‘winner’s bonus’ was more pronounced at the first than the second stage. Thus as well as estimating the bias across all districts, we also analyse bias only in those districts at which the election was determined at each stage. This gives us the following separate analyses: 1. Analysis of the biases across all districts using the first-preference votes for the largest LNC party; 2. Analysis of the biases across all districts using the combined first-preference votes for the LNC parties; 3. Analysis of the biases across all districts using the 2PP votes (when the votes for the Liberal and National parties are combined, minus any leakage because those who supported one of them allocated their lower preferences to a candidate other than the person standing for the other coalition party); 4. Analysis of the biases across those districts where the result was determined on the first-preference votes, using the first-preference votes for the largest LNC party; 5. Analysis of the biases across those districts where the result was determined on the first-preference votes, using the combined first-preference votes for the LNC parties; 6. Analysis of the biases across those districts where the result was determined on the 2PP votes. In all of these analyses, a positive bias figure indicates that the asymmetry favours the Liberal-National coalition whereas a negative figure indicates a pro-ALP bias. Tables 5 and 6 report the results for the first two analyses, of all districts using the first-preference votes: Table 5 treats the competition as between the ALP and the largest party representing the coalition locally; Table 6 treats the competition as between the ALP and the two coalition parties combined. They both show a very similar pattern, with total bias almost invariably favouring LNC (the exception is for 1996 in Table 6). It was fairly small save at the 1998 and 2004 contests, however, when the coalition would have received a substantially larger number of seats than the ALP if they had equal votes shares redistributed in the way described here. That procoalition bias would have been worth 19-20 seats to the LNC in 1998 – when the two (LNC and ALP had almost equal vote shares in the actual contest) – and 10-11 seats in 2004. As expected, vote efficiency was by far the main source of the bias. If we look at the bias across all districts using the 2PP vote shares rather than first preferences – which means that the votes for the two LN coalition parties are necessarily combined – very much the same results appear in Table 7 as Tables 5-6.

10

The main exception is that only the 1998 contest shows very substantial bias favouring the LNC – again, almost entirely the result of differences between the two blocs in vote efficiency. These first three analyses produce a very similar pattern of results, therefore. Bias is very small at most of the elections. Where it is substantial (1998 and 2004 in the firstpreference analyses, 1998 only in the 2PP analyses) it favours the LNC; and the predominant pro-LNC bias source is vote efficiency. Turning to the separate analyses of the two stages of the AV system, Table 8 reports analyses for districts where the election was determined won on the first preference votes (with the coalition party included in the analyses being that winning the largest number of votes). Here we see a very different pattern from that in the previous analyses – except that, as is the case throughout, the predominant bias source is vote efficiency. The result appears to have been substantially biased at every contest, but not in the same direction. In 1993 and 1998, when the ALP won most votes in those districts where the contest was decided on first preference votes alone, it enjoyed biases of 12 and 14 seats respectively. When the coalition parties were in the lead in 1996, 2001 and 2004, on the other hand, they were the beneficiary (shown here as negative bias). Their advantage was even greater – by 28, 19 and 40 seats respectively, which are very substantial numbers when the number of districts where the result was determined at this stage was only 82, 62 and 87. In 1996, therefore, when 82 seats were determined, instead of getting 41 seats each with equal vote shares, the coalition would have been allocated 55 to only 27 for the ALP. Table 9 reports a similar analysis to Table 8, for all districts where the result would have been determined in the first preferences if the Liberal and National vote totals were combined in every district (hence the larger number of seats in Table 9 than 8 at each election except 2004). The pattern is exactly the same as that shown in Table 8: the result is biased towards the largest party (ALP in 1993 and 1998, LNC at the other three elections) – and the bias is greater when the LNC is in the ascendancy than when the ALP is. Finally, Table 10 reports on analyses in the districts where the election was determined at the second stage after the distribution of preferences producing the 2PP totals. Here we see very little bias – relatively little disproportionality was displayed for these districts in Table 2 – except in 1993, when the ALP was a very substantial beneficiary from a more efficient vote distribution. The final set of three analyses provide a very different set of findings from those obtained in the first three, therefore. If we split the districts according to which stage the election was determined there (i.e. on first preferences or 2PP votes), we find separate bias patterns. In those where the determination occurred at the first stage, there was considerable bias at every election (and not just at two of them, as Tables 45 suggested). Furthermore, that bias was not in the same direction at each contest: the party with most votes at this stage benefited from the bias, because its votes were more efficiently distributed than its opponents’. In those districts where the result was only determined using the 2PP votes, however, bias was in general slight – certainly so at the four elections won by the LNC.

11

‘Unintentional gerrymanders’ – packed and cracked The findings reported here regarding the direction of electoral bias at Australian House of Representatives elections differ from those for the only other country for which a series of elections has been examined. There has been no evidence of several switches in the direction of bias between elections in the UK (Johnston et al, 2002; Johnston, Pattie and Rossiter, 2008), so why this switching in Australia, with proALP bias in the districts determined at the first stage of the AV contest in 1993 and 1998 – when Labor was the largest party in those seats – but a pro-LNC bias in 1996, 2001 and 2004 when they were the overall victor in those districts? One possible answer to this can be found in a US categorisation of gerrymanders into two main types. Packed gerrymanders occur when one party’s votes are concentrated in a small number of districts, which it wins by large majorities, whereas its opponent’s (the beneficiary – and perhaps the instigator – of the gerrymander) are more evenly distributed across a larger number of districts, which it wins by smaller (though sufficient) majorities: there are relatively few marginal seats. Cracked gerrymanders, on the other hand, occur whe n the beneficiary (and probably instigating) party wins in a majority of the districts, but by fairly small majorities in a lot of them, producing a considerable number of marginal seats that could change hands with a relatively small shift in partisan support. Cracked gerrymanders can deliver more districts to the gerrymanderer than packed gerrymanders, but they are also more risky because a majority of districts could be lost if the voters shift away from the gerrymandering party: its majority is much safer with a packed gerrymander. There is no gerrymandering in Australia, of course, but nevertheless the equivalent of one could arise as an unintended consequence of the redistricting procedure, if the underlying geography of party support produces such an outcome – with a party’s vote efficiency being a function of an ‘unintentional gerrymander’ (as Johnston and Hughes, 1978, suggested for Brisbane). Our results suggest this has been the case in Australia – in particular in those districts determined at the first stage of the AV process. Furthermore, this is more likely to be a cracked than a packed ‘unintentional gerrymander’, involving a substantial number of districts won by relatively small margins and so potentially lost at a succeeding election if there is a small shift in popular support away from the winner. No non-partisan cartography involving a neutral Boundary Commission is likely to produce an ideal ‘intentional gerrymander’ – whether packed or cracked – as was initially demonstrated by Taylor and Gudgin (1976; see also Chisholm et al, 1997; Gudgin and Taylor, 1982). However, it might approximate one type rather than the other and in this way be the context for biased election results. We anticipated that the outcome of Australian Commonwealth redistricting would be closer to a cracked than a packed gerrymander and Table 11 indicates this was so. For the districts determined at the first stage of the AV process, the first block characterises them according to the margin of victory (in five catego ries from the most marginal to the safest) and the party estimated to have won there with equal voting shares. At each of the five contests, both blocs (ALP and LNC) had a substantial number of very safe seats, which would have been won by a majority of mo re than 20 percentage points over its nearest rival. Not surprisingly, given the class-based geography of support for each party and the geographical separation of the classes – especially in the towns – there

12

is a substantial number of districts where one bloc’s supporters dominate. 7 In addition, however, there are clear elements of the equivalent of a cracked gerrymander, a small number of districts won by one of the parties with a small majority that would be vulnerable to a shift in support towards its opponent. Thus in 1993 and 1998, when the ALP was the leading party in the districts determined at the first stage, it won all of the marginal seats (those won by a majority of 10 percentage points or less). At the other three contests, however, when the Liberal-National parties won a majority of the votes in those districts, they won virtually all of the marginal seats. These patterns are further illuminated by the data in Table 12, which show the numbers (with equal vote shares) of wasted votes per seat lost and of surplus votes per seat won, plus the percentages of votes that were effective. Not surprisingly, the ALP had the larger percentage of effective votes in 1993 and 1998, when the bias favoured it, with the LNC parties having the (much) larger percentage at the other three contests. Of particular interest is the number of surplus votes per seat won, which was much smaller for each party at the elections when they won in these districts (1993 and 1998 for ALP, 1996, 2001 and 2004 for LNC) than when they lost. The clear implication is that when whichever of the major parties (ALP or LNC) won, it did so with relatively small majorities in a number of seats. These marginal seats apparently change parties according to which is the largest overall in the districts as determined on first preference votes, and this is the source of the biases observed there – the equivalent of an ‘unintentional cracked gerrymander’. A similar pattern does not characterise the districts where the election was determined after preferences had been distributed, however. Instead, there was little difference between the two parties in the distribution of districts across the various marginality levels, except in 1993 when there was a substantial pro-LNC bias (Table 10). Table 13 indicates that most districts were marginal at this stage of the procedure, and that each party had approximately the same distribution of seats that it would have won with equal vote shares across the five categories. When preferences are distributed to create the 2PP totals, therefore, the two main voting blocs – one based on the LN coalition and the other on the ALP – are very close competitors in most districts, and no bloc appears to have the edge over the other. Conclusions The study of disproportio nality and bias in election results is complicated by two main factors in the case of Australian federal contests, relative to other countries which employ single- member constituency systems. The first is the use of the Alternative Vote, which precipitates an immediate run-off contest in those districts where no candidate wins a plurality of the first-preference votes. The second is the role of the Liberal-National coalition, involving two parties which share many policy stances and are committed to governing in a coalition if together they win the election but nevertheless contest some seats. The National (formerly Country) party is subject to a challenge from a Liberal candidate only when a sitting National member does not stand for re-election (and the same is true if a Liberal member representing a rural district does not seek re-election), but the number of such contests has declined 7

Many of the National party’s victories over the ALP in districts where the result was determined on the first-preference votes were by very large majorities, placing them in the fifth category in table 11.

13

recently (from 39 in 1993 to 9 in 2004). Where both have fielded candidates and split the ‘right-wing’ vote this is in the knowledge that if the contest extends beyond counting of the first-preferences their votes will be combined in the final determination (which was the reason for introducing the AV system: Farrell and McAllister, 2006). Studies of disproportionality and bias in Australian election results have noted these two complications but adopted strategies which largely avoid them, thereby producing results that provide – at best – only a partial picture. Most, for example, have used the 2PP vote totals only in their calculations, ignoring the fact that at most elections a substantial proportion of the districts have their results determined on the firstpreference votes (57, 55 and 58 per cent of all districts in 1993, 1996 and 2004). In those districts, the 2PP vote totals are irrelevant to the translation of votes into seats. Similarly, most analyses have treated the LNC as a single party (reasonably so if they are analysing 2PP vote totals only), thereby ignoring the potential impact of them splitting the right-wing vote in some districts might have on the outcome. (At the 1993 election, when both LNC parties contested 39 districts, there were 10 cases in which the Liberal and National votes combined would have defeated the successful ALP candidate on 2PP votes – but did not because the determination was at the first stage using the first preferences.) By taking these two complications into account, this paper has provided greater insights into disproportionality and bias at recent Australian federal elections than have other studies. It has shown that bias is mainly a feature of the districts in which the contest is determined on the counting of first-preference votes, and it was substantial at each of the five elections studied here. That observed bias, not surprisingly given compulsory voting and the frequent redistributions in Australia, resulted not even in part (as in the UK) from differences in district population and turnout but almost entirely from differences between the two main party blocs in the efficienc y of their vote distributions. Furthermore, this equivalent of an ‘unintentional cracked gerrymander’ did not favour the same bloc at each election; rather it favoured the largest of the two at the particular (in the first-preference vote count only). This comes about because there is a relatively small number of marginal districts at each election which are determined at the first stage of the counting process and are almost all won at the elections studied by one of the largest of the two blocs. Geography is thus clearly implicated in the translation of votes into seats at recent Australian federal elections, and thus in the disproportionality that analysts of elections over a range of countries with different electoral systems have identified. In considerable part the implication of geography is probably serendipitous: it comes about because different groups of voters, who tend to support different parties (particularly ALP, Liberal and National) tend to be concentrated in different parts of the country – the geography of party support reflects the geographies of the labour and housing markets and of rural vs urban. Only a small number of the electoral districts, defined in an entirely non-partisan way and with little variation in their population size, combine a mix of areas of different types that make them marginal and thus potentially won by more than one party. At recent federal elections, most of these marginal constituencies have favoured the same party, producing the biases that we have reported here – but only if the district election results were determined on the first-preference votes alone. Where the allocation of preferences to produce the 2PP

14

totals was needed, there was very little bias and the two party blocs had similar distributions across the various marginality categories. An indirect influence of geography on electoral bias through the labour and housing markets and associated socio-economic (class?) segregation, along with the ruralurban cleavage, may be the sole cause of the bias production – with trends in the overall support for the two parties across all districts resulting in the ‘cracked gerrymander’ effect identified above. However, it is very likely that the political parties themselves see to accentuate its influence by focusing their campaigning on the districts where the elections can be won or lost, rather than where the outcome is fairly certain (i.e. in the ‘safe’ seats – see Forrest et al, 1999); this is certainly the case in the UK, where geographically-targeted campaigns have led to exaggerated biases at recent elections (Johnston et al, 2001; Johnston, Rossiter and Pattie, 2006). If this is the case, then geography is not just the context within which biased election results are generated but also a tool to be deployed by those wishing to create such biases (the political parties and their supporters). The ‘territorial forms’ of Australia’s electoral laws, like those of other countries which use single- member constituency systems, are thus deeply implicated in the disproportional and biased processes of translating votes into seats. Acknowledgements We are grateful to Charles Pattie for valuable comments on a draft of this paper References AEC 2004: Australia’s federal redistributions 1901-2003. Canberra: Australian Electoral Commission, Research Report 4. Anckar, C. 1997: Determinants of disproportionality and wasted votes. Electoral Studies 16: 501-515. Brookes, R. H., 1953: Seats and votes in New Zealand. Political Science 5: 37-44. Brookes, R. H. 1959: Electoral distortion in New Zealand. Australian Journal of Politics and History, 5, 218-223. Brookes, R. H. 1960: The analysis of distorted representation in two-party, single member elections. Political Science, 12, 158-167. Chisholm, M., Devereux,. B. and Versey, R. 1981: The myth of non-partisan cartography: the tale continued. Urban Studies 18: 213-218. Farrell, D. and McAllister, I. 2006: The Australian electoral system. Sydney: University of New South Wales Press. Forrest, J., Johnston, R.J. and Pattie, C.J. 1999: The effectiveness of constituency campaign spending in Australian state elections during times of electoral volatility: the New South Wales case, 1988-95. Environment and Planning A 31:1119-1128.

15

Gallagher, M. 1991: Proportionality, disproportio nality and electoral systems. Electoral Studies 10: 33-51. Grofman, B. and King, G. 2007: The future of partisan symmetry as a judicial test for partisan gerrymandering after LULAC v. Perry. Election Law Journal 6: 2-35. Copy at http://gking.harvard.edu/files/jp.pdf Gudgin, G. and Taylor, P. J. 1979: Seats, votes and the spatial organization of elections. London: Pion. Gudgin, G. and Taylor, P. J. 1982: The myth of non-partisan cartography: clarifications. Urban Studies 19: 405-407. Gudgin, G. and Taylor, P. J. 1980: The decomposition of electoral bias in a plurality election. British Journal of Political Science 4: 515-521. Hughes, C. A. 1977: Malapportionment and gerrymandering in Australia. In R. J. Johnston, editor, People, places and votes: essays on the electoral geography of Australia and New Zealand. Armidale: Department of Geography, University of New England, 93-110. Jackman, S. 1994: Measuring electoral bias: Australia, 1949-1993. British Journal of Political Science 24: 319-357. Johnston, R. J. 2002: Manipulating maps and winning elections : measuring the impact of malapportionment and gerrymandering. Political Geography 21: 55-66. Johnston, R. J. and Hughes, C. A. 1978: Constituency delimitation and the unintentional gerrymander in Brisbane. Australian Geographical Studies 16: 99-110. Johnston, R. J., Pattie, C. J., Dorling, D. and Rossiter, D. J. 2001: From votes to seats: the operation of the UK electoral system since 1945. Manchester: Manchester University Press. Johnston, R. J., Pattie, C. J. and Rossiter, D. J. 2008: Electoral distortion despite redistricting by independent commissions: the British case, 1950-2005. In L. R. Handley and B. Grofman, editors, The politics of redistricting. New York: Oxford University Press. Johnston, R. J., Rossiter, D. J. and Pattie, C. J. 1999: Integrating and decomposing the sources of partisan bias: Brookes’ method and the impact of redistricting in Great Britain. Electoral Studies 18: 367-378. (‘Addendum’, Electoral Studies 19: 649-650.) Johnston, R. J., Rossiter, D. J. and Pattie, C. J. 2006: Disproportionality and bias in the results of the 2005 general election in Great Britain: evaluating the electoral system’s impact. Journal of Elections, Public Opinion and Parties 2: 37-54.

16

King, G., Grofman, B., Gelman, A. and Katz, J. 2005: Amicus brief in the case of Jackson v. Perry in the U.S. Supreme Court (No. 05-276) – available at http://gking.harvard.edu/files/amicus-sym.pdf Rydon, J. 1957: The relation of seats to votes in elections for the Australian House of Representatives 1949-1954. Political Science 9(2): 49-61. Rydon, J. 1968: Malapportionment: Australian style. Politics 3: 133-147. Rydon, J. 1986: Two-party preferred: the analysis of voting figures under preferential voting. Politics 21: 68-74. Soper, C. S. and Rydon, J. 1958: Under-representation and electoral prediction. Australian Journal of Politics and History 4: 94-106. Taylor, P. J. and Gudgin, G. 1976: The myth of non-partisan cartography: a study of electoral biases in the English Boundary Commission’s redistribution for 1955-1970. Urban Studies 13: 13-25. Taylor, P. J. and Gudgin, G. 1977: Antipodean demises of Labour. In R. J. Johnston, editor, People, places and votes: essays on the electoral geography of Australia and New Zealand. Armidale: Department of Geography, University of New England, 111-120.

17

Table 1. The distribution of votes and seats and seats:votes ratios in elections to the Australian House of Representatives 1993-2004 Election 1993 First-Preference Votes V S Liberal-National Coalition Liberal 37.1 49 National 7.2 16

V

S

V

S

V

S

V

S

39.0 8.2

75 19

34.2 5.3

64 16

37.4 5.6

69 13

40.8 5.9

75 12

Australian Labor Others

38.8 14.0

49 5

40.1 20.4

67 1

37.8 19.2

65 3

37.6 10.5

60 3

44.9 10.8

80 2

1996

1998

2001

2004

Seats:Votes Ratios Liberal-National Coalition Liberal 0.89 National 1.51 TOTAL 1.00

1.30 1.56 1.35

1.26 2.04 1.37

1.23 1.55 1.27

1.23 1.36 1.24

Australian Labor Others

0.85 0.24

1.13 0.03

1.15 0.10

1.06 0.29

1.21 0.13

2PP Votes Liberal National Australian Labor Others

V 48.5 50.7 0.7

S 65 80 2

V 52.4 45.1 2.0

S 94 49 5

V 48.4 50.2 1.3

S 80 67 1

V 50.4 48.4 1.2

S 82 65 3

V 52.3 46.4 1.7

S 87 60 3

Seats:Votes Ratios Liberal National Australian Labor Others

0.91 1.07 2.00

1.22 0.73 1.70

18

1.12 0.90 0.52

1.09 0.89 1.67

1.11 0.86 1.18

Table 2. The distribution of votes and seats and seats:votes ratios in elections to the Australian House of Representatives 1993-2004 – districts are split into those won on first preferences and those won on 2PP votes

Election 1993 1996 Districts won on first preferences V S V S Liberal-National Coalition Liberal 37.8 30 38.5 43 National 5.1 6 11.2 14 Australian Labor 47.2 48 37.3 25 Others 9.9 0 13.0 0 Seats:Votes Ratios Liberal-National Coalition Liberal National TOTAL Australian Labor Others Number of seats

1998

2001

2004

V

S

V

S

V

S

32.7 3.4 46.0 17.9

15 3 32 0

38.9 7.1 36.8 17.2

34 7 21 0

44.1 5.6 35.6 14.7

56 7 24 0

0.95 1.40 1.00 1.22 -

1.36 1.52 1.40 0.82 -

0.92 1.77 1.00 1.39 -

1.24 1.59 1.44 0.92 -

1.46 1.44 1.46 0.77 -

84

82

50

62

87

Districts won on 2PP votes V S Liberal-National 50.6 29 Australian Labor 47.7 32 Others 1.7 2

V 48.7 46.4 4.4

S 38 24 4

V 50.9 47.1 2.0

S 63 34 1

V 48.8 49.7 1.5

S 41 44 2

V 47.0 49.4 3.5

S 24 36 3

Seats:Votes Ratios Liberal-National Australian Labor Others

0.91 1.06 1.87

1.18 0.78 1.37

1.26 0.74 0.51

0.97 1.02 1.53

0.81 1.16 1.36

Number of seats

63

66

98

87

63

19

Table 3. The Gallagher Least-Squares Index of disproportionality for elections to the Australian House of Representatives 1993-2004 Election 1993 1996 All seats First preferences 7.7 13.6 2PP 4.1 11.9 Districts determined on first preferences 10.5 17.4 Districts determined on 2PP votes 3.9 9.5

20

1998

2001

2004

14.4 6.0

14.6 4.5

10.6 6.1

18.0

18.9

17.0

13.0

1.5

8.5

Table 4. Frequency distributions for the number of registered voters

Minimum 10th percentile Lower Quartile Median Upper Quartile 90th percentile Maximum

1993 63,149 72,780 74,285 77,106 79,944 83,061 97,210

1996 63,499 73,844 76,445 79,715 82,333 84,963 98,800

1998 62,419 76,035 79,278 81,877 84,364 86,477 105,359

2001 53,705 78,534 82,366 84,987 86,876 89,329 111,547

2004 54,725 82,002 84,590 87,527 90,817 93,903 111,835

Mean Standard Deviation

77,446 4,852

79,329 5,058

81,841 5,763

84,242 6,722

87,280 7,018

21

Table 5. Results of the bias analysis for all districts using first preference votes and the largest LNC party (a positive value indicates bias to the LNC; a negative value indicates bias to the ALP) Election Number of Seats % Share of Votes LNC ALP Seats with equal Vote Shares LNC ALP Total Bias Electorate Size Abstentions Efficiency Minor Party Votes Minor Party Seats

1993 147

1996 148

1998 148

2001 149

2004 150

41 45

46 39

39 40

42 38

46 38

74 73

74 73

82 62

74 72

78 67

1 0.7 -1.0

1 0.5 -0.8

20 0.3 -0.5

2 -1.4 -0.8

11 -0.9 0.8

1.8

2.2

16.5

3.2

10.0

-0.3 0.0

-1.6 1.0

2.5 1.0

1.1 0

-0.1 1.0

22

Table 6. Results of the bias analysis for all districts using first preference votes and the combined LNC first preference votes (a positive value indicates bias to the LNC; a negative value indicates bias to the ALP)

Election Number of Seats

1993 147

1996 148

1998 148

2001 149

2004 150

44 45

47 39

40 40

43 38

46 38

76 71

72 75

82 63

74 71

78 67

5 0.4 -0.7

-3 -0.2 -0.7

19 0.7 -0.5

3 -1.2 -1.0

11 -0.9 0.7

Efficiency

5.2

-1.2

15.9

4.4

11.1

Minor Party Votes Minor Party Seats

0.2 0.0

-1.6 1.0

2.6 0.0

0.9 0.0

0.8 -1.0

% Share of Votes LNC ALP Seats with equal Vote Shares LNC ALP Total Bias Electorate Size Abstentions

23

Table 7. Results of the bias analysis for all districts using 2PP votes (a positive value indicates bias to the LNC; a negative value indicates bias to the ALP) Election Number of Seats

1993 147

1996 148

1998 148

2001 149

2004 150

44 45

53 45

48 50

50 48

52 46

76 71

72 70

84 60

74 71

76 70

5 0.4 -0.7

2 0.0 -0.8

24 0.0 -0.5

3 -1.3 -0.8

6 -0.2 -3.1

Efficiency

5.2

2.8

23.9

5.2

9.2

Minor Party Votes Minor Party Seats

0.2 0.0

-0.8 0.0

0.0 0.0

0.0 0.0

0.0 0.0

% Share of Votes LNC ALP Seats with equal Vote Shares LNC ALP Total Bias Electorate Size Abstentions

24

Table 8. Results of the bias analysis for those districts determined on the first preference votes and the largest LNC party (a positive value indicates bias to the LNC; a negative value indicates bias to the ALP) Election Number of Seats % Share of Votes LNC ALP Seats with equal Vote Shares LNC ALP Total Bias Electorate Size Abstentions Efficiency Minor Party Votes Minor Party Seats

1993 84

1996 82

1998 50

2001 62

2004 87

42 47

49 37

36 46

46 36

49 36

36 48

55 27

18 32

40 21

63 23

-12 0.2 -0.7

28 -1.1 -0.5

-14 -0.1 0.7

19 -0.4 -0.9

40 -1.8 1.5

-11.6

29.5

-14.2

19.9

40.0

0.1 0.0

0.1 0.0

0.7 0.0

0.6 0.0

0.0 0.0

25

Table 9. Results of the bias analysis for all districts determined using first preference votes with the combined LNC votes (a positive value indicates bias to the LNC; a negative value indicates bias to the ALP) Election Number of Seats % Share of Votes LNC ALP Seats with equal Vote Shares LNC ALP Total Bias Electorate Size Abstentions Efficiency Minor Party Votes Minor Party Seats

1993 94

1996 92

1998 53

2001 67

2004 87

44 46

50 37

37 45

47 35

50 36

46 48

60 32

21 32

45 21

63 24

-2 0.4 -0.7

28 -1.0 -0.5

-11 0.0 -0.4

24 -0.4 -1.0

39 -1.9 1.5

-1.7

29.5

-11.4

24.9

39.2

0.2 0.0

0.1 0.0

0.7 0.0

0.6 0.0

-0.1 0.0

26

Table 10. Results of the bias analysis for those districts determined using 2PP votes (a positive value indicates bias to the LNC; a negative value indicates bias to the ALP) Election Number of Seats % Share of Votes LNC ALP Seats with equal Vote Shares LNC ALP Total Bias Electorate Size Abstentions Efficiency Minor Party Votes Minor Party Seats

1993 63

1996 66

1998 98

2001 87

2004 63

51 48

49 47

51 47

49 50

47 49

22 38

32 29

47 47

41 44

31 29

16 -0.3 0.0

3 -0.2 -0.3

0 -0.9 0.9

-3 0.5 -0.2

2 0.6 0.2

-15.7

0.5

-0.9

-4.3

1.2

0.0 0.0

0.0 0.0

0.0 0.0

0.0 0.0

0.0 0.0

27

Table 11. The distribution of districts that would have been won by the LNC and ALP, if they had equal vote shares, by the margin of victory, in those districts where the result was determined by the first preference votes

Margin/Winner 0– 5% 5 – 10% 10 – 15% 15 – 20% 20% +

1993 1996 1998 2001 2004 LNC ALP LNC ALP LNC ALP LNC ALP LNC ALP 0 3 8 2 0 2 0 0 10 1 0 5 5 0 0 3 4 0 7 0 3 4 10 0 0 6 6 0 16 0 3 9 4 0 18 21 10 0 13 0 30 27 28 25 18 32 21 21 24 16

28

Table 12. The distribution of wasted votes per seat lost, surplus votes epr seat won, and percentage of votes effective in districts that would have been won by the LNC and ALP, if they had equal vote shares, in those districts where the result was determined by the first preference votes

1993 Wasted votes per seat lost Liberal-National 24,173 Australian Labor 21,856 Surplus votes per seat won Liberal-National 20,979 Australian Labor 15,733 Percentage of votes effective Liberal-National 29.1 Australian Labor 42.9

1996

1998

2001

2004

18,287 25,929

22,750 16,195

16,778 24,510

19,074 28,249

12,454 25,371

28,939 16,279

14,659 28,619

10,791 26,762

54.7 19.0

18.9 47.3

51.3 18.0

60.5 16.5

29

Table 13. The distribution of districts that would have been won by the LNC and ALP, if they had equal vote shares, by the margin of victory, in those districts where the result was determined by the 2PP votes

Margin/Winner 0– 5% 5 – 10% 10 – 15% 15 – 20% 20% +

1993 1996 1998 2001 2004 LNC ALP LNC ALP LNC ALP LNC ALP LNC ALP 17 11 14 13 15 19 13 11 13 9 11 6 10 5 12 5 7 6 5 7 7 2 3 6 10 10 8 12 10 6 0 0 2 1 2 7 8 7 1 4 7 0 3 4 8 6 5 8 2 3

30