MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA

Transcription

1 PARLIAMENTARY LIBRARY OF SOUTH AUSTRALIA RESEARCH SERVICES MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA A revised version of Information Paper No.15, incorporating data from the 1997 State election. JENNI NEWTON INFORMATION PAPER 16

2 1998, Parliamentary Library of South Australia Not to be reproduced in part or whole without the written permission of, or acknowledgement to, the Parliamentary Library of South Australia. This information paper has been prepared by the Research Service of the South Australian Parliamentary Library. While all care has been taken to ensure that the material is both accurate and clearly presented, the responsibility for any errors remains with the author. February 1998 ISSN

3 CONTENTS INTRODUCTION... 1 THE FAIRNESS CRITERIA IMPLEMENTATION OF THE FAIRNESS REQUIREMENT THE 1991 REPORT... 3 THE 1994 REPORT... 4 THE ACCURACY OF THE CHECKING EXERCISE THE PROBLEM BEFORE 1993: ESTIMATING THE TWO PARTY PREFERRED VOTE AT EACH BOOTH... 6 THE PROBLEM NOW: DECLARATION VOTES A. THE NUMBER OF DECLARATION VOTES B. THE EFFECT OF DECLARATION VOTES 1. Did declaration votes make a difference to the formality of the count? Did declaration votes make much difference to the outcome of the election? Was this effect because declaration voters were much more radical or conservative than voters who lodged their votes at booths? Was the effect of declaration votes greater in those seats which recorded more declaration votes (i.e. was the effect due to the number of declaration votes)? C. DO WE NEED TO INCLUDE DECLARATION VOTES IN A CHECKING EXERCISE D. HOW SHOULD WE ALLOCATE DECLARATION VOTES TO BOOTHS? Allocate declaration voters pro-rata to booths at the average two party preferred rate for all declaration voters in the seat Allocate declaration voters pro-rata to booths at the average two party preferred rate for all ordinary voters in the seat Allocate ALP and LIB declaration voters pro-rata to booths SUMMARY APPLYING THE FAIRNESS CRITERIA LOOKING TOWARDS THE NEXT ELECTION PATTERNS IN ACTUAL RESULTS: SWINGS IN RECENT SA STATE ELECTIONS Uniform swings? Proportionate swing? Characteristics of the individual seats? THEORETICAL APPROACHES TO SWINGS 1. Fischer and Charnock The Cube Rule THE CUBE RULE AND THE DEMOCRATS A SITTING MEMBER ADVANTAGE THE NUMBER OF MARGINAL SUMMARY AND CONCLUSION REFERENCES

4 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA "It's not the voting that's democracy, it's the counting." Tom Stoppard, Jumpers (1972) Act 1 After each State election in South Australia, an Electoral Districts Boundaries Commission is required to examine State electorates with a view to redistribution. In deciding whether boundaries need to be amended, the Commission is required to take into account many factors, including "as far as practicable" making sure that the boundaries are fair to all parties. This paper is about the process by which the boundaries are judged to be fair. INTRODUCTION Since 1975, South Australian State electoral boundaries have been redistributed on a regular basis by an Electoral Districts Boundaries Commission (EDBC), made up of a Supreme Court Judge, the State Electoral Commissioner and the State Surveyor General. Redistributions took place in 1976 (with effect at the 1977, 1979 and 1982 State elections), in 1983 (effective at the 1985 and 1989 elections), in 1991 (effective at the 1993 State election) and most recently in 1994 (effective at the 1997 State election). Redistributions do not require Parliamentary approval but become operative on the order of the EDBC; while a 1994 amendment now allows the EDBC to issue a draft report before making a final order and report, the final order of an EDBC can only be appealed on points of law. Under the Constitution Act (SA), the EDBC must take several factors into account when deciding where to draw electorate boundaries. Each resulting electorate must have the same number of electors, give or take 10 per cent (s.77). The electorate boundary must be an unbroken line, unless islands are involved. That is, no electorate can be made up of two (or more) separate pieces of land, unless the pieces are islands (s.82(5)). As far as it is practicable for the Commission to make it so, the redistribution must be "... fair to prospective candidates and groups of candidates so that, if candidates of a particular group attract more than 50 per cent of the popular vote (determined by aggregating votes cast throughout the State and allocating preferences to the necessary extent), they will be elected in sufficient numbers to enable a government to be formed." (s.83(1» Within these limits, the Commission must also, as far as practicable, have regard to, "(a) the desirability of making the electoral redistribution so as to reflect communities of interest of an economic, social, regional or other kind; (b) the population of each proposed electoral district; (c) the topography of areas within which new electoral boundaries will be drawn; (d) the feasibility of communication between electors affected by the redistribution and their parliamentary representative in the House of Assembly; (e) the nature of substantial demographic changes that the Commission considers likely to take place in proposed electoral districts between the conclusion of its present proceedings and the date of the expiry of the present term of the House of Assembly and may have regard to any other matters it thinks relevant." Page 1 of 41 : Dated February 20, 1998

5 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA (Constitution Act (SA) 1934, s.83(2). The underlining is mine.) While the other requirements have been in the Act since 1975, the fairness requirement of section 83(1), was inserted into the Act in THE FAIRNESS CRITERIA The insertion into the Constitution Act of the fairness requirement has quite a history. In a Constitution Amendment (Electoral Redistribution) Amendment Bill which he introduced in 1989, the Hon. KT Griffin (then Liberal Shadow Attorney General) proposed several changes to the Constitution Act. He argued that under the boundaries as they then stood, and as they would apply at the election expected within months, the Liberal Party required 52% of the total State-wide two party preferred vote for House of Assembly seats in order to win government. He also noted that because the most recent redistribution of seats had taken place six years previously, population movements had caused the number of electors in seats to vary by mid from almost in Fisher to fewer than in Elizabeth. The Bill proposed a redistribution after each second State election and also the incorporation into section 83 of the Act of a requirement that the EDBC, when making a redistribution, should take into account a new criterion, namely "the desirability that a political Party or group gaining 50 per cent plus one of the two party preferred vote at a general election of members of the House of Assembly at which the proposed redistribution would apply should have a reasonable prospect of forming a government." (Han. KT Griffin, SAPO, 9 August 1989: p114) The Bill lapsed when an election was called. On November 25th 1989, the ALP (with the support of two Independent Labor candidates) won 24 of the 47 House of Assembly seats with 47.9% of the two party preferred vote. The Liberal Party (and one National Party candidate) won 23 seats with 52.1 % of the two party preferred vote. Much as Mr Griffin had predicted, the ALP took government with less than half of the two party preferred vote. In 1990 the Labor Deputy Premier (Hon. Dr Hopgood) introduced a Bill to amend the Constitution to require a redistribution to take place after every two elections. In his second reading speech, he equated fairness with equality of elector numbers, saying that the real problem with the redistribution process was that when the Parliament had changed from 3-year to 4-year terms, no account had been taken of the resultant increase in the length of time between redistributions required under the Constitution Act. The Hopgood Bill proposed to solve that problem by requiring more frequent redistributions so that the number of electors in each electorate was more frequently pulled into line. In reply, the Liberal Party strongly advocated a fairness requirement, emphasising that electorates with equal numbers of electors still could not prevent the 1989 situation in which a redistribution could "lock up the potential Liberal vote in as few seats as possible and to spread the Labor vote as widely as possible" (0 Baker, SAPO 10 April 1990: p.1377). "At the last election, there were 13 non-labor seats which polled higher than 65% of the two party preferred vote for the winning candidate... However, only five seats were won by Labor with the successful member getting more than 65 per cent. This is the perfect psephological example of the 'locked in' interest.... At the last election, the non-labor vote in these seats totalled Labor needed about fewer votes to win its 13 safest seats than the Liberal Party." (0 Baker, SAPD 10 April 1990: p.1377.) Page 2 of 41 : Dated February 20, 1998

6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA The Liberal argument was that if single-member electorates were to be retained, then more frequent redistributions, with account taken of the political value of each vote, needed to occur. At the same time, a willingness was also expressed to consider the West German top-up system or multi-member electorates with proportional representation. The Bill was sent to a Select Committee on April 10 th In November of 1990, the Select Committee reported to the Parliament, and recommended: an immediate redistribution, and further redistributions after each State election; retention of single-member electorates for the next State election, but after that (and in the light of the results of that election) a review of both the West German top-up system and single-member electorates (as far as I know this has not been formally done, although the parties may well have made their own judgements based on the 1993 State election); and adoption of the fairness requirement originally proposed in the Griffin Bill. In order to implement the Select Committee's recommendations, sections of the Constitution Act needed to be amended, and so a referendum was required. The referendum was held on 9 February 1991, and the question asked was "Do you approve the Constitution (Electoral Redistribution) Amendment Bill 1990?" The vote in favour was 76.7% across the State, with a clear majority in each electorate. As a result, the Constitution Act (SA) was amended to require that the Electoral Districts Boundaries Commission should be required to formulate a redistribution after each State election (not every third election). This is more frequent than any other State, Territory or Federal requirement except that for the ACT. The amendments also require the EDBC to draw the boundaries in such a way that, as far as is reasonably practicable, the party which won the majority of votes should be assured of holding the majority of seats. This fairness requirement is unique to South Australia, although the Court Liberal government in Western Australia did express an interest in electoral reform for their Legislative Assembly, including equal-sized electorates and a fairness criterion ("Electoral reform mooted", West Australian, 17 December 1996:p9) IMPLEMENTATION OF THE FAIRNESS REQUIREMENT THE 1991 REPORT The Commissioners have addressed the fairness requirement in a number of ways. Firstly, both the 1991 and the 1994 Commissions have applied the proposed boundaries to the results from the previous State election, in a checking exercise. This exercise is also done by each of the major parties, Professor Dean Jaensch (Politics and International Studies Department, Flinders University of South Australia), Antony Green (ABC TV Election Unit, Sydney) and myself. In a sense, what we ask is: "If the last election had been held on these boundaries, would the party with the majority of votes have won the majority of seats?" As with economic models, all other factors are generally held constant. This retrospective application of election results is a kind of bottom-line exercise. If a proposed set of boundaries showed that, at the last election, the party with the majority of votes would clearly not have won the majority of seats, then we might question how the Commission proposed that those same boundaries would better reflect voters' intentions at the next election. Page 3 of 41 : Dated February 20, 1998

7 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA The 1991 Commission drew the metropolitan boundaries, and then checked them for the political consequences of the proposed boundaries. "The check of the metropolitan area was mainly made against the 1989 voting patterns. It was found that the requirements of section 83 (1) would not have been met on an overall review of these seats. A new kind of imbalance had developed simply out of a faithful application of the criteria in Part V of the Act other than section 83 (1). The past voting patterns were closely checked in every metropolitan booth, and the necessary final adjustments were then made in order that the total metropolitan distribution would comply, as far as practicable, with the requirements of section 83 (1 )." (EDBC 1991 Report: para 54.2.) I understand this passage to mean that the Commission checked the 1989 voting patterns in the proposed metropolitan seats, found that the new boundaries would not have given enough metropolitan seats to the party with the majority of two party preferred votes, and adjusted the boundaries until the required relationship between seats and votes occurred. From comments made at paragraph 58.3, it seems that in some areas both 1985 and 1989 State election results were scrutinised. In their submissions to the 1991 EDBC, both the Labor and the Liberal parties based their proposed sets of boundaries on 1989 election results. The Commission saw this as evidence of agreement that past voting patterns could be appropriately used to indicate future voting patterns. The Commission noted the possibility that the parties may have based parts of their submission on "other more recent information" (EDBC 1991 Report: para ) - presumably party polling - but saw such information as "likely to be of minimal value" (op.cit.). "The Commission agrees with the thrust of the Liberal Party's submission, that voting patterns in the 1989 election are the most relevant because they are the most recent. They also reflect the wishes of electors at a time when the distribution of seats was as even as possible (24 : 23) and when the overall popular vote was quite even (48 : 52)." (EDBC 1991 Report: para 59.1) The Commission's Report acknowledged that the checking exercise would still not guarantee that a fair result would be achieved at the next election, for many reasons, including that: some voters will move into, and out of, an electorate; some voters can be expected to change their allegiance; and the checking exercise contains inherent errors. Nonetheless, the Commission "... has taken on board these conflicting views and bears in mind all the imperfections and uncertainties associated with using past voting patterns as an instrument for checking the political consequences of its redistribution. In the end, it has decided to use past voting patterns as they were the only guide available to it." (EDBC 1991 Report: para 58.4, my emphasis) IMPLEMENTATION OF THE FAIRNESS REQUIREMENT THE 1994 REPORT In 1994 the Commission staff checked the boundaries formulated by the Commission, against the results from the 1993 State election. While the 1993 data were the most recent voting figures then available to the Commission, they were never seen as indicative of the way that voters might align themselves at the next State election, for the simple reason that in 1993 there was a massive swing against the ALP government, largely as a result of what had come to be known as "the State Bank disaster". No-one imagined that all of those voters who changed their allegiance in 1993 would continue to vote Liberal (either as a first or subsequent Page 4 of 41 : Dated February 20, 1998

8 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA preference), but by the same token those voters were not all expected to return to Labor at the next election either. The 1994 Commission did not call attention to the extraordinary nature of the 1993 voting figures, and the extra limitations which these data imposed upon the task of making the distribution fair for parties contesting the next election: "The test of electoral fairness in s.83(1) is easily stated but, as will be obvious, quite difficult to apply. The data is incomplete and much must depend on judgement and conjecture and even guesswork. There is no rule of thumb by which electors' future voting intentions may be reckoned. There is more than one way in which the political effect of changing an electoral district may be assessed. " (EDBC 1994 Report: para 10.1) In discussing their checking exercise in relation to Unley and Kaurna (at paras 10.3 to 10.4, and 10.7 to 10.10), the 1994 Report shows that the Commissioners were sensitive to factors which might bring in new voters to an electorate or might move current voters out of an electorate. Projecting the future population of electorates occupied a great deal of the Commission's efforts. However, when it came to the problem of estimating how a swing back from the 1993 results would be likely to occur (and the effect of that swing on the fairness of the boundaries as they would stand at that next election), the Commission found major difficulties. Both the 1991 and 1994 Commissions have accepted, with some reservations, the concepts of a two party preferred vote (2PP), of ranking electorates by their swing-to-lose factor, and of showing these ran kings in the form of an electoral pendulum. The pendulum is based on an idea that if a party gains a swing of - say - 11 %, then it will win all of the opposing party's seats requiring a swing-to-iose of 11 % or less. The 1994 Commission rejected the probability of a uniform swing at the next election, remarking that "... experience shows that election swings are rarely, if ever, uniform." (EDBC 1994 Report: para 6.2). It noted that in fact at the 1993 State election the swings were decidedly non-uniform. With an average swing to the Liberal Party of 8.9%, the 14 rural seats swung towards the Liberal Party by 4.6% on average, whereas the 23 metropolitan seats swung towards the Liberal Party by, on average, 10.5% (EDBC 1994 Report: para 8.5 and Appendix 12). The Commission noted that the ALP had won only 39.1 % of the two party preferred vote at the 1993 election, and that in order to win government it therefore would require an average swing of 11 % across the State at the next election. It published, for the first time in South Australia, its own estimates of the swing-to-iose in each seat according to the new boundaries; the Pendulum at Figure 1 shows the Commission's swing-to-iose estimates for these new seats. It should be noted here that the Commission's estimates are based on the results from the 1993 general election; it is not common practice to take the results of by-election contests into account when preparing estimates of the effect of a redistribution of boundaries. This is partly because by-elections are sometimes not contested by a major party (for example there was no Liberal Party candidate for the Elizabeth by-election of ) so there may be no two party preferred result, but the main reason is that a redistribution affects every seat and so we need results from every seat on a comparable basis. The Pendulum at Figure 1, based on the Commission's estimates of the new seats after the 1994 redistribution, shows that 14 seats currently held by Liberal Members would each fall to labor with a swing to Labor in their areas of under 11 %. A fifteenth, Florey, would fall with a swing of 11 %. With a uniform swing of 11 % in each of the new seats, the ALP would win a majority of the two party preferred vote and would win 25 seats (one more than a simple majority). Page 5 of 41 : Dated February 20, 1998

9 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA While Frome would only require a swing of 8.2%, and Stuart 9.0%, the Commission argued that the smaller swings to Liberal in rural seats at the 1993 election showed that these two seats would be unlikely to fall to Labor even on a Statewide average two party preferred swing of 11 %; instead one or more metropolitan seats individually requiring a swing of more than 11 % would be likely to fall. "Our aim must be to ensure that the Labor Party would win 14 more seats under the 1994 redistribution with a swing in its favour of 11 %. As we have said, it is unlikely to win Stuart or Frome on an 11% State-wide swing. However, if on that hypothesis the Labor Party is getting less than 11 % of the two party preferred vote in Stuart and Frome, it must be getting more than 11 % in one or more other seats. The difficulty comes in reckoning where the above-average swings in its favour are likely to occur." (EDBC 1994 Report: para 8.11, my emphasis) The Commission noted that at the 1993 State election, marginal metropolitan seats recorded larger swings than safer metropolitan seats (at para 8.16), and concluded that the new boundaries would be likely to be fair to both parties. My own feeling is that the Commission did not make out a case that the ALP would win 14 more seats - and only 14 - with a non-uniform swing of 11 %. The problem is that the Commission specifically rejected the likelihood of a uniform swing occurring at the next election (EDBC 1994 Report: para 6.2) and also specifically dismissed the probability of a swing back to the ALP being non-uniform but proportionate to the swing towards the Liberal Party in "..it would obviously be quite wrong to suppose that a general swing to the Labor Party at the next election would simply reverse, and to a proportional degree, the swings against it in the individual electoral districts in There are any number of variables that would combine to make that a most unlikely eventuality." (EDBC 1994 Report: para 8.16) Having disposed of the two most obvious ways of guessing what pattern a future swing would take, the Commission does not seem to have a strong basis for its judgement that the new boundaries would be fair. My feeling is that the pattern of swings is an unexplored area in the Commissions' work. However, it is not the only area which seems to call for further work; the other is the accuracy of the checking exercise. Given that the checking exercise is logically prior to application of a predicted swing, the checking exercise should be examined first. THE ACCURACY OF THE CHECKING EXERCISE THE PROBLEM BEFORE 1993: ESTIMATING THE TWO PARTY PREFERRED VOTE AT EACH BOOTH The detailed outcome of the 1991 Commission's estimates was not published, but the Commission did report that the result was that if the new boundaries been in place at the 1989 State election, the Liberal Party would have won 24 of the 47 House of Assembly seats, and hence government. My own estimate at the time, and the estimates made by the ALP and the Liberal Party, was that although several of the seats would be very marginal, if the new boundaries had been in place in 1989 the Labor government would have been returned with 24 seats. Dean Jaensch gave both parties 23 seats and found Newland too hard to call (Newton 1992: summary table). Page 6 of 41 : Dated February 20, 1998

10 6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA The Commission had superior access to data and a great deal of well-respected expertise, and its estimate was, rightly, accepted. My point in raising these different estimates of the outcome of the redistribution, is not to criticise any particular estimates, but to show that it is possible for several justifiable estimates to be made. One of the difficulties which existed in 1991 and which has since been removed, was that the State Electoral Office produced figures showing the distribution of preferences for an electorate as a whole, but not for each booth. This meant that when a portion of an electorate needed to be moved, an assumption needed to be made about how the preferences would have been allocated in the relevant booths, to reach a two party preferred figure for the area being moved. In the absence of any better guide, it was common practice to assume that the pattern of distribution of preferences in a group of booths would be the same as that for the electorate as a whole. Since the 1993 State election the State Electoral Office has been able to provide two party preferred counts for each booth. Not only does this remove a possible error in the earlier process, but it gives us an idea of the size of the previous error. It is quite possible now to compare the pattern of distribution of preferences in the seat as a whole, with the pattern of distribution of preferences at each booth. In general, in 1993, the pattern of distribution of preferences in metropolitan booths was not very different from the average when the seat was counted as a whole, but there were some fairly major differences in country booths. Table 1 illustrates this point. It looks at two party preferred figures for Lee, the seat which became the most marginal following the 1993 State election. Preferences were distributed from five candidates: Behn (Call To Australia), Brown (Independent for Natural Law), Cotton (Independent), Wasylenko (Independent Labor) and Clements (Australian Democrats). On the left of the table are the actual first preference votes received at each booth, and for the seat as a whole, and the actual two party preferred count at each booth. In the left-centre panel is the actual distribution of preferences which occurred in 1993 at each of the Lee booths, and in the final count for the seat as a whole. This can be inferred from the two columns to the left. It is simply the proportion of first preference votes for candidates other than ALP or Liberal, which must have been allocated to the ALP or Liberal candidates (when preferences were distributed) in order to arrive at the two party preferred figures published for each booth. In the right-centre panel are the estimated two party preferred figures for each booth, which I have calculated by applying the pattern of distribution of preferences in the seat as a whole, to each booth (on a pro-rata basis). In this case, in each booth I have allocated 64.3% of all preferences to the Liberal candidate, and 35.7% of all preferences to the ALP candidate. This is how I used to make booth estimates, before the State Electoral Office was able to provide the booth figures on a two party preferred basis. On the right of the table are measures of the difference between the two methods of showing the two party preferred vote for each booth - the actual and the estimated method. Table 1 shows that in most cases the estimates of how preferences would have been distributed (using the average figures for the seat as a whole) would have been reasonably close to the actual result. The largest differences would have been at Woodville West, where the actual distribution of preferences gave the ALP 68.8% of preferences, compared to the ALP's share of only 64.3% of preferences in the seat as a whole, and in West Lakes Shore, where the actual distribution of preferences gave 40.7% of preferences to the Liberal Party, compared to only 35.7% of preferences in the seat as a whole. Page 7 of 41: Dated February 20,1998

11 6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA However, when we apply these differences to the number of first preference votes lodged at these booths, and look (at the right side of Table 1) at the difference between these two methods in terms of their overall effect on the two party preferred count, we find that the estimates were generally within 10 votes of the actual result and even in the case of the very big West Lakes Shore booth the estimate was only 22 votes out. As a proportion of all formal votes in Lee, this is 0.1 percentage points. In effect what we see in Lee is that the old method would have introduced an error which sometimes favoured the ALP and sometimes the LIB candidate, and which was of the order of up to 0.1 percentage points per booth. In a redistribution checking exercise, the error in a whole seat rather than one booth would depend on how a redistribution split the seat. If a hypothetical redistribution kept the borders of Lee intact, then there would be no difference between the two methods of estimating the two party preferred vote in the seat. If, however, the seat were split so that - for example - the northern booths of Semaphore Park, Semaphore Park South and West Lakes Shore were moved off into another seat, then having the actual two party preferred counts for each booth makes the process of estimating the two party preferred vote for the new seat more accurate. Table 2 shows that using the old method of estimating two party preferred booth figures in Lee we would have estimated the two party preferred vote for the northern group of booths to be ALP:LlB 42.8:57.2 but we now know (because we have actual counts) that it was actually 42.5:57.5. That is, the old method would have generated an error of 0.3 percentage points in this case. Is Lee representative of all seats? Looking at the estimates made of the effect of the 1991 redistribution, what margin of error should be added to the estimate for each seat? Did the 1991 estimates show a fair result or was the margin of error so high that no such conclusion should have been drawn? I have calculated the figures for each booth in each seat on the same basis as Table 1 showed for Lee. These calculations take up 15 pages, which I can supply on request; Table 3 shows the first of those 15 pages. The far right hand column of Table 3 shows, for each booth, the difference between the actual two party preferred vote in the booth and the estimate of the two party preferred vote which needed to be made before This difference - or error - is expressed in Table 3 in terms of 1. the percentage of preferences which the ALP (or LIB) candidate actually received at each booth compared to the average percentage of preferences that same candidate received for the seat as a whole; 2. the number of votes (on a two party preferred basis) this difference represents; and 3. the percentage of all formal votes in an average seat which this difference represents - i.e. the size of the error attached to each booth's two party preferred estimate. The reason why I have expressed the error value as a percentage of the average number of formal votes in a seat (rather than the actual number of formal votes in the relevant seat) is that in a checking exercise it is necessary to look at the error which would arise from combining several booths from several seats, so a common denominator is needed. In 1993 the average number of formal votes in a seat was I should state here that the full calculations for Table 3 show a problem with the distribution of declaration vote preferences in Custance. I believe that this problem lies in the data supplied, and because it generates an incongruous result for Custance I will exclude that one seat wherever that seems necessary. As is shown in Table 3, the error induced by having to assume that the distribution of preferences in each booth would follow the pattern of distribution of preferences in the seat as a whole was, in general, fairly small; when we look at each booth of each seat, we find that each booth carried an error in the order of +/- 0.1 percent of all formal votes in the seat. Page 8 of 41 : Dated February 20, 1998

12 6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA Some booths, by virtue of a large number of preferences which had to be distributed, a very distinct pattern of preference distribution or a large number of voters who used the booth, generated a larger error. Table 4 shows that there were 14 seats which had at least one booth with an error of 0.2 percentage points. That is, the old method (which was then unavoidable but is now no longer necessary) would have produced an estimate for a redistributed seat which was inaccurate to the extent of at least 0.2 percentage points. While safe seats would not be greatly affected, in marginal seats this error would be sufficiently large to be a problem in a redistribution checking exercise. The single booths generating the highest error using the old method with voting figures from the 1993 State election, would have been Port Lincoln (0.4% of all formal votes lodged in Flinders) and Port Lincoln South (0.5% of all formal votes lodged in Flinders). In a redistribution a segment of a seat represented by several booths which carry a larger error might be separated from the seat and added to another seat. In this case the error attached to each booth would be cumulative. However, looking at the booths listed in Table 4 and the relevant maps it seems to me that the booths with larger errors are rarely neighbours, and when they are the seats are not marginal ones. Without extending this exercise to every likely possible combination of booths, I would guess that the old method of estimating a two party preferred result may have introduced an error of - at the most percentage points. That is, I believe that the two party preferred estimates made with State data prior to 1993 (or indeed Federal data prior to 1993) need to be treated as if they could have a margin of error of up to 0.5 percentage points. After the 1991 redistribution, my own estimate and that of the ALP and the Liberal Party, was that although several of the seats would be very marginal, if the new boundaries had been in place at the time of the State election in 1989 the Labor government would have been returned with 24 seats. Dean Jaensch gave both parties 23 seats and found Newland too hard to call. The three most marginal seats looked to be Newland, Unley and Colton. The various estimates of the two party preferred status of the "new" seats (as ordered by the 1991 EDBC) to which I have access are shown in Table 5. If we recognise a margin of error of up to 0.5 percentage points due to the old methodology, then both Newland and Unley could well have been either Labor or Liberal seats on the basis of the 1989 State results. What does that say about the fairness of the 1991 redistribution? pointless to argue with either the Commission's statement that Simply that it would be "It is likely that, on the present redistribution, the Liberal Party would have governed with 52 per cent of the popular vote in " (EDBC 1991 Report: para 55.1) or with the other estimates which seemed to indicate that the ALP would still have won 24 seats (and hence government). The figures were just too close. There is no need now to dwell on a methodological problem that has been overcome. For the 1993 State election, two party preferred results are available for each booth, so the problem of how to distribute preferences in individual booths does not arise and the methodological error is avoided. But now another error, albeit of smaller proportions, has emerged because the figures from individual booths do not include declaration votes. Page 9 of 41 : Dated February 20, 1998

13 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA THE ACCURACY OF THE CHECKING EXERCISE THE PROBLEM NOW: DECLARATION VOTES A declaration vote is one "given to an elector who is unable to vote at a polling booth in his/her District on polling day" (State Electoral Department, 1993, Everyday Electoral Terms). Declaration votes include absent and postal votes. By their nature these votes cannot be allocated to particular polling booths, so they are presented separately in the statistics. Before 1993 when booth figures became available on a two party preferred basis, it was common practice to allocate a number of declaration votes to each booth on a pro-rata basis, before allocating preferences (pro-rata) to finally estimate the two party preferred vote for each booth. Now that we no longer need to go through the exercise of allocating preferences to estimate a two party preferred figure for each booth (because we have actual counts now), I suspect that declaration votes will be left out of the process altogether. The reasoning would probably be that there are so few declaration votes, and that we really don't know how to allocate them to individual booths, so that it may be better to concentrate on the ordinary votes lodged at booths. Is it important to include declaration votes in the checking exercise, or could they be ignored? A. THE NUMBER OF DECLARATION VOTES At the 1993 State election the number of declaration vote certificates issued in State electorates ranged from 1389 in Chaffey to 3036 in Adelaide; in 1997 the number of declaration vote certificates issued ranged from 2371 in Finniss to 3525 in Waite. Table 6 shows the proportion of all votes which were declaration votes, for each seat contested at the State elections of 1993 and (There were boundary changes between the elections but the table stands because there is no need to compare a given electorate's results in 1993 and 1997). Table 6 shows that in 1993 declaration votes made up 11.5% of all votes State-wide, ranging from 6.8% of all votes in Chaffey, to 15.5% of all votes in Norwood and 15.6% of all votes in Adelaide. In 1997, declaration votes were more common in every electorate, probably because the election was held during the last weekend of school holidays. Of all ballot papers issued in 1997, 14.4% were issued to declaration voters. In Waite 18% of all votes were declaration votes, and the lowest proportion was recorded in Finniss with 12%. While 18% represents an appreciable proportion of all votes lodged, declaration votes still might be unimportant if they follow the same pattern of formality and the same two party preferred pattern as ordinary votes lodged in the same seat. The following tables therefore look at the difference between declaration votes and ordinary votes lodged at booths within each electorate. In the following tables the results from several seats are marked "rethrow". There is always a possibility in a seat that the final two candidates preferred by the electors of that seat will be from parties other than the ALP and the Liberal Party, so the final two party preferred result in the actual count for the seat may be NAT: LI B, IND LAB:ALP, IND:L1B etc. In these circumstances, the State Electoral Office has, since the 1985 State election, asked for the first preferences of the non-alp non-lib candidate to be reallocated ("rethrown") for Page 10 of 41 : Dated February 20,1998

14 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA statistical purposes only, and has provided us with data showing the ALP:LlB two party preferred vote in those seats. For example, the actual result in Chaffey in 1993 was NAT:LlB 43.7% : 56.3% but a rethrow produced an ALP:LlB two party preferred result of ALP:LlB 21.0%: 79.0%. Rethrows were conducted in 1993 for Chaffey and Flinders (both NAT: LIB) and for Davenport and Waite (both DEM: LI B), and in 1997 for Napier (ALP:DEM), for Heysen, Waite, Davenport, Schubert, Kavel, Finniss (all DEM:LlB), for Chaffey and Flinders (both NAT:LlB) for MacKiliop (IND LIB :LlB), and for Gordon (IND:LlB). B. THE EFFECT OF DECLARATION VOTES 1. Did declaration votes make a difference to the formality of the count? Tables 7A and 7B show the proportion of all ordinary votes, all declaration votes and all votes (ordinary + declaration) which were formal or were deemed to be formal as a result of the operation of the ticket vote provisions. (Table 7 A is for 1993, 7B for 1997.) I have arranged the seats in order of their swing-to-iose, in case we can see a contrast between ALP and LIB seats, safe and marginal seats etc. The far right hand column of both versions of Table 7 shows the difference between the proportion of formal votes in the ordinary vote count and in the count of all votes for each seat: it therefore shows the effect on the proportion of formal votes of including declaration votes in the count for each seat. What is fairly clear is that including declaration votes did not change the proportion of formal votes in the Statewide count by very much at all. In 1993 including declaration votes in the count raised the proportion of formal votes by 0.2 percentage points in Spence and Adelaide (because a higher proportion of declaration votes were formal) and lowered it by 0.3 percentage points in Peake (because a lower proportion of declaration votes were formal). Apart from those three seats the effect was so small that it could easily be ignored. In 1997 including declaration votes raised the proportion of formal votes by 0.3 percentage points in MacKillop, by 0.2 percentage points in 18 seats and lowered it in only one seat - Fisher, by 0.1 percentage points. For the State as a whole, the effect of including declaration votes in the count was to increase the proportion of votes which were formal by 0.1 percentage points. The size of the effect, and the direction, did not follow an obvious pattern in either 1993 or 1997, and my feeling is that the difference between declaration votes and ordinary votes lodged at booths in terms of their formality is so small that we can ignore it and concentrate on formal declaration votes and formal ordinary votes. The following sections therefore relate to formal votes. 2. Did declaration votes make much difference to the outcome of the election? Tables 8A and 8B show the two party preferred vote for all ordinary votes in each seat, and the two party preferred vote when al/ votes are taken into account (i.e. when declaration votes are added into the count). The EFFECT OF DEC VOTES column at the right of Tables 8A and 8B is calculated as the difference between these two columns and shows the effect on the two party preferred vote in each seat, of including declaration votes in the count. The two versions of Table 8 show that, for the State as a whole, declaration votes did not have a great effect in 1993 nor in 1997, but did have a big enough effect on some seats that we should look at them in more detail. In 1993 declaration votes raised the ALP two party preferred vote by 0.1 percentage points and lowered the Liberal two party preferred vote by 0.1 percentage points, Page : Dated February 20,1998

15 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA and in 1997 they raised the Liberal two party preferred vote by 0.3 percentage points and lowered the ALP two party preferred vote by 0.3 percentage points. In seats won by the ALP candidate, declaration votes reduced the two party preferred ALP vote by only a tiny amount percentage points (on average) - in 1993 but in 1997 declaration votes reduced the two party preferred ALP vote in seats won by the ALP by an appreciable amount percentage points (on average). In 1997 a pattern emerged in ALP seats, where in all cases except Elder the effect of including declaration votes was to reduce the ALP's two party preferred percentage for the seat. In seats won by the Liberal Party candidate, declaration votes reduced the two party preferred LIB vote by 0.2 percentage points (on average) in 1993 and increased it by 0.2 percentage points in 1997 (again, on average). Unlike the ALP seats in 1997, there was no pattern to the direction of declaration votes in LIB seats in 1997 (nor in ALP or LIB seats in 1993). In some cases declaration votes reduced the LIB winning margin (the greatest effect was 0.5 percentage points in Schubert) and in some cases they increased it (1.2 percentage points in Davenport, 1.0 in Stuart and Frome). It is clear from Table 8A that in 1993 declaration votes were able to lift a winning candidate's two party preferred vote by up to 1.1 percentage points (in Eyre) or to reduce a winning candidate's two party preferred vote by up to 1.0 percentage points (in Peake). Were there any seats in 1993 where the ordinary vote count put the seat within 1.0 percentage points of being won or lost? One, Hanson. In 1993, the two party preferred vote in Hanson was ALP: LI B 49.1 % : 50.9% when votes cast at all of the booths were counted. It was possible, based on the figures shown in Table 8A, that the inclusion of declaration votes could have changed the two party preferred vote by 1.0 percentage point either way. Hanson's result could therefore have been anywhere between ALP:LlB 48.1% : 51.9% and 50.1% : 49.9% (in fact it emerged at 48.8% : 51.2%). It seems quite possible that Hanson may have been won by the ALP rather than the Liberal Party candidate if the declaration votes had not been so supportive of the Liberal candidate. Table 8B shows that in 1997 declaration votes were able to lift a winning candidate's two party preferred vote by up to 1.2 percentage points (in Davenport) or to reduce a winning candidate's two party preferred vote by up to 1.6 percentage points (in Giles). Were there any seats in 1997 where the ordinary vote count put the seat within 1.6 percentage points of being won or lost? Four, and indeed one seat - Hartley - was won by the ALP on the ordinary vote count but then swung to the Liberal Party once declaration votes were included into the count. The ALP won Norwood (51.6% 2PP), Mitchell (51.1 %) and Hartley (50.2%) on the basis of the ordinary votes lodged at booths. In addition the ALP was within 1.6 percentage points of Stuart (49.4%) on the ordinary vote count. If the declaration votes in Mitchell and Norwood had favoured the Liberal Party more (or perhaps if there had been more declaration votes) and if the declaration votes in Hartley and Stuart had favoured the ALP more (or if there had perhaps been more of them) any of these four seats could have had a different outcome. In Giles the ALP two party preferred vote was reduced by 1.6 percentage points when declaration votes were added in to the count. If this had happened in Stuart - a neighbouring seat and in many ways quite similar to Giles - then the seats would have changed from 50.6% LIB (two party preferred ordinary vote) to 51.0% ALP (two party preferred full count). The actual results when declaration votes were included into the count were as follows: the ALP margin in Norwood was reduced by 0.9 percentage points but the ALP candidate still won the seat, in Mitchell the ALP margin was reduced by 0.2 percentage points and the ALP candidate still won the seat, in Hartley the ALP margin was reduced by 0.7 and the seat passed to the LIB candidate, and in Stuart the LIB margin was increased by 1.0 percentage point and the LIB candidate won the seat. Page 12 of 41 : Dated February 20, 1998

16 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA 3. Was this effect because declaration voters were much more radical or conservative than voters who lodged their votes at booths? I have often heard it said that absent and postal votes (declaration votes) are more likely to favour the Liberal Party than are votes lodged at the booths of the same seat. I have also heard it said that declaration votes will favour the party which held the seat before the election. Tables 9A and 9B show, for each seat, the two party preferred percentages for ordinary votes lodged at booths, and for declaration votes. Declaration votes differed from ordinary votes by up to 11 percentage points (in Giles in 1997 the two party preferred vote for ordinary votes was 63.0% ALP but for declaration votes it was only 52.1 % ALP). Should this be judged a big difference or a small difference? In electorates where the two party preferred results can be quite radically different from one booth to another, perhaps the most surprising thing is the relative conformity of declaration votes with ordinary votes for each electorate. Table 9A does not show any evidence of a general tendency of declaration votes to favour the Liberal Party in 1993 nor in 1997 (Table 9B). In general, declaration votes favoured the ALP in seats where ordinary votes favoured the ALP, and declaration votes favoured the Liberal Party in seats where ordinary votes lodged at booths favoured the Liberal Party. (The exceptions were Giles, Napier and Peake in 1993, and the four most marginal seats in Florey, Mitchell, Norwood and Hartley.) Tables 9A and 9B also show quite clearly that declaration votes can favour the same party as the ordinary votes but can favour that party more or less strongly. That is, in a LIB seat the declaration votes can be either more strongly LIB than ordinary votes, or less so. It is tempting to say that in general declaration votes favoured the party which won the ordinary votes count, but a little less strongly. While this does seem to be true it is not terribly helpful because the Difference columns of Tables 9A and 9B show that the exceptions to this generalisation are often the seats which are most delicately balanced: in 1993 the generalisation did not apply in six of the 10 remaining ALP seats, five of the 14 seats the ALP lost to LIB, and in seven of the 23 seats the Liberal Party retained. in 1997 the generalisation does apply in all of the seats the ALP retained, and in all except one of the seats which the ALP won from the Liberal Party (the ALP seat of Elder did not comply). The four most marginal ALP seats on the ordinary vote count (Florey, Mitchell, Norwood and Hartley) probably did comply with the generalisation if we accept that declaration votes favoured the party which won the ordinary vote count (ALP) but less strongly, to the extent that support for the ALP in the declaration vote count fell below 50%. The four most marginal LIB seats after Hartley (Stuart, Frome, Colton and Unley) did not complyand of the remaining LIB seats half did and half did not comply. If declaration voters were consistently more likely to vote for the sitting Member, then we should be able to see it most clearly in the seats which changed hands. In 1993 the ALP lost 14 seats to the Liberal Party, and it is true that in nine of these 14 seats declaration voters were a little more likely to have voted for the ALP than ordinary voters did. On average in these 14 seats the ALP won 43.4% of the ordinary vote count but more % - of the declaration vote count. In 1997, 11 of those same 14 swinging seats swung back to the ALP, and in 10 of the 11 the sitting Members and their party (this time the Liberal Party) again won a bigger share of the declaration vote count than they won of the ordinary vote count. On average in Page 13 of 41 : Dated February 20, 1998

17 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA these 11 seats the Liberal Party won only 45.3% of ordinary votes but slightly more % - of declaration votes. In 1993 the ALP sitting Members who kept their seats attracted a bigger proportion of declaration than ordinary votes in only six out of the ten cases, and LIB sitting Members won a greater share of declaration votes than ordinary votes in only 7 of the 22 cases. In 1997 none of the 10 sitting ALP Members won a greater share of declaration votes than ordinary votes, but 15 of the 26 "Liberal" (including NAT, IND, IND LIB) cases did. In summary, it does not seem that declaration voters have been consistently much more radical or more conservative than ordinary voters in the same electorate, although there is some evidence that in a seat which swings the sitting Member may lose a slightly smaller share of the declaration voters than of ordinary voters. In both 1993 and 1997, both the direction and the size of the effect varied markedly from seat to seat. The figures in the column marked EFFECT OF DECLARATION VOTES show that there were enough declaration votes in each seat for the different voting preferences of declaration voters to have an effect on the overall result in almost every seat. Comparing these figures with the results in the DIFFERENCE column it is clear that in both 1993 and 1997 the seats where declaration votes were appreciably different from ordinary votes (in terms of their two party preferred percentages) were the ones where declaration votes had the greatest overall effect on the count for the seat, suggesting that it was this difference, rather than the varying number of declaration votes in each seat, which was responsible for declaration votes changing the two party preferred result in most seats. In summary, in general the effect of declaration votes on the final count has been small, although in a few seats it has been so big as to demand attention; in terms of individual seats the effect of declaration votes on the overall count has not been predictable in either size or direction although the maximum effect of declaration votes in the two elections for which we have data has been +/- 1.6 percentage points. 4. Was the effect of declaration votes greater in those seats which recorded more declaration votes (i.e. was the effect due to the number of declaration votes)? To look at the effect of the number of declaration votes, I have standardised the number of declaration votes in each seat, by recalculating the final two party preferred vote for each seat using the average number of declaration votes (2237 in 1993 and 2749 in 1997), but retaining the actual two party preferred percentages for declaration votes lodged in that seat. For this exercise the number of ordinary votes has been kept the same as in the actual election count, but the number of total votes has been allowed to rise or fall. (Custance is not included in Table 10A or 11A.) The calculations are shown in Tables 10A and 10B and the results are summarised in Tables 11 A and 11 B. Tables 11 A and 11 B show the actual effect of declaration votes in each seat, and then the hypothetical effect of declaration votes in the same seat had that seat recorded the average number of declaration votes for that election. The difference between the two columns is very small - only 0.1 percentage points in every case (or about 20 votes) except Chaffey in 1993 (0.3 percentage points) and Spence and Stuart in 1997 (0.2 percentage points). My interpretation of the figures in Tables 10 and 11 is that they show that the number of declaration votes lodged in each electorate has not made an important difference to the final two party preferred vote in each seat. Page 14 of 41 : Dated February 20,1998

18 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA C. DO WE NEED TO INCLUDE DECLARATION VOTES IN A CHECKING EXERCISE? It is worth pointing out that no clear pattern has emerged by ordering the seats according to their marginality and party, although ordering seats according to which party held the seat before the election may be useful. If declaration votes can make a difference of up to 1.6 percentage points, as they did in the 1997 final count, then I believe it would be wrong, especially in the marginal seats, to exclude declaration votes from a redistribution estimation exercise. It is quite possible that the very large swing against the government in 1993, and the election date in 1997, generated larger-than-normal discrepancies between declaration and ordinary votes. It is possible that data from the next few State elections will show that, in general, declaration votes fall into line with booth votes, in which case it may be possible in future to ignore declaration votes when calculating the effect of a boundary change on an electorate. In the meantime, I believe that declaration votes should be included in any checking exercise. D. HOW SHOULD DECLARATION VOTES BE ALLOCATED TO BOOTHS? The problem with including declaration votes in an estimation exercise is that there is no way of knowing where declaration voters live, but they need to be allocated to booths so that when a seat is split in a redistribution declaration votes can be transferred with booth votes. Several options are available. 1. Allocate declaration voters pro-rata to booths, at the average value for all declaration votes in the seat. It has been customary to assume that declaration voters are spread across the electorate in the same way as ordinary voters. Declaration votes are usually allocated to individual booths on a pro-rata basis or to segments of the electorate. In the end both of these methods will give the same results, as long as declaration votes are allocated at the two party preferred value for all declaration votes in the electorate. Table 12A shows how this method has been applied, using Peake's 1993 results as an example. The aim of the exercise in Table 12A is to break Peake up into three segments which can then be attached to other electorates in a hypothetical redistribution exercise. The Brooklyn Park booth received 10.4% of all formal ordinary votes lodged within Peake, so 10.4% of all formal declaration votes lodged within the electorate need to be allocated to that booth: Brooklyn Park booth therefore receives 211 declaration votes. Brooklyn Park South received 6.7% of ordinaries so it receives 6.7% of declaration votes (136 declaration votes), and so on. The average value for all declaration votes in Peake was ALP:LlB 52.9:47.1 %, so those two party preferred percentages are used to split the declaration votes in each booth into Labor and Liberal votes. Adding these to the actual Labor and Liberal ordinary votes lodged at each booth gives an estimated final figure for each booth, in which each booth is accorded not only the ordinary votes lodged there but also a proportion of the declaration votes lodged in the electorate as a whole. The advantage of this method is that at the end of the exercise the two party preferred percentages for individual booths may have changed - and that seems reasonable because on average declaration votes are different from ordinary votes - but the two party preferred vote for the seat as a whole has not changed. The problem with this method - and it has not been acknowledged because in the end it Page 15 of 41 : Dated February 20, 1998

19 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA has seemed to be too difficult to remedy - is that all declaration votes in an electorate are allocated at the same value. This means that in Peake, for example, declaration votes will be allocated at the same rate of ALP: LIB 52.9%: 47.1% to Thebarton, which is a strongly Labor booth, and to Lockleys which is a strongly Liberal booth. Are declaration voters from Thebarton and from Lockleys really more like each other, than like ordinary voters from Thebarton and ordinary voters from Lockleys? 2. Allocate declaration voters pro-rata to booths at the average value for all ordinary votes in the seat. The Electoral Districts Boundaries Commission seems to have allocated declaration votes in a quite different way. The Commission has allocated declaration votes at the two party preferred value expressed by the ordinary votes in a given booth or segment of the electorate. The 1994 Report records the Commission's method in some detail, as follows. "Take the case of a simple movement of 1000 electors from District A to District B. The Electoral Office's records, which have been made available to us, show the two party preferred result for each polling booth within District A and the two party preferred declaration vote figures for that district as a whole. The Electoral Office's records also show the percentage of electors in each district who voted as formal ordinary voters and formal declaration voters. If, for example, 77.0% of electors in District A voted as formal ordinary voters and 10.0% of them voted as formal declaration voters, 870 (770 plus 100) votes are transferred from District A to District B. In assessing the political effect of the transfer, one has to identify - and this will not always be clear - the polling booth (or booths) where the 770 formal ordinary voters are likely to have voted last December. Usually the two party preferred result from that booth can then be applied to the 770 transferred votes, with the result being deducted from the two party preferred result in District A and added to the two party preferred result in District B. However, the catchment area of the polling booth in question may not be politically homogeneous. If that appears to be the case a judgement will have to be made as to whether the polling booth result, as applied to the transferred votes, should be weighted in the direction of one party or the other to allow for the estimated political disposition of the voters who are being transferred to District B. As for the 100 formal declaration voters, we can only assume that they voted the same way as the ordinary formal voters in the transferred area. Again, the votes, appropriately distributed between the parties, are added to the District B result and deducted from the District A result." (1994 EDBC Report: para 10.5, my italics.) Table 12B applies this method to the electorate of Peake, using the same hypothetical segments as were used in Table 12A. As the various segments of the seat are parcelled up ready to be sent to other electorates, a pro-rata parcel of declaration votes is attached (the quote above shows that the Commission assumes that the distribution of formal declaration votes across the seat will be the same as the distribution of formal ordinary votes.) As Segment 1 contains 18.7% of all formal ordinary votes lodged in Peake, it receives 18.7% of all formal declaration votes lodged in Peake. The Commission applies to these declaration votes, the two party preferred rate which is the average for the ordinary votes lodged at those booths included in the segment. At the end of the exercise there are three segments whose two party preferred percentages have not changed, but (because this method does not recognise that there is a difference between the declaration vote two party preferred percentage and the ordinary vote two party preferred percentage in each seat) the two party preferred vote for the seat as a whole is not correct. The advantage of the Commission's methodology is that it assumes that declaration voters have more in common with other ordinary voters who live in their local area, than with other declaration voters who happen to live in their electorate. Page 16 of 41 : Dated February 20,1998

20 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA The problem with the Commission's methodology is that it does not produce the right answer at the end of the process. 3. Allocate ALP and LIB declaration voters pro-rata to booths. Table 12C shows that it is possible to assume that declaration voters are spread across the electorate in roughly the same way as ordinary voters, and also to assume that declaration voters have more in common with the ordinary voters at their local booth than they do with voters elsewhere in their electorate. This is possible if ALP declaration votes are allocated according to the distribution across the electorate of all ALP voters who lodged their votes at booths, and LIB declaration votes are allocated across the electorate according to the distribution of LIB voters who lodged their votes at booths. In Table 12C, looking at the two party preferred results for each of the booths in Peake, Thebarton had 7.2% of all formal ALP ordinary votes (and only 3.3% of all formal LIB ordinary votes) so it receives 7.2% of ALP declaration votes (and 3.3% of LIB declaration votes). Similarly, Lockleys received 14.6% of formal LIB ordinary votes but only 6.9% of ALP ordinary votes, so we allocate to it 14.6% of the LIB declaration votes and only 6.9% of the ALP declaration votes. This breaks the relationship between the ALP and the LIB votes in the declaration vote count. There is no need to allocate the declaration votes at the same two party preferred rate to each booth: it is enough to know how many ALP declaration votes and how many LIB declaration votes to allocate to each booth. When these are added to the actual counts of the ordinary votes lodged at each booth, the outcome is a two party preferred result for each booth which includes the ordinary votes and the declaration votes for the area. The method gives correct two party preferred percentages for the seat as a whole, and it also allows declaration votes to affect the ordinary vote count for each booth. Table 120 summarises the results from the previous three tables. It shows that this last method will produce better estimates than the others. It produces the correct number of ALP and LIB formal votes for the seat as a whole, It reflects the fact that the party preferences shown in a seat's declaration votes can be different from the same seat's ordinary votes, by showing a change to the two party preferred percentages in a segment once declaration votes are included, and It reflects the fact that some booths in a seat are much stronger for the ALP, and takes this into account by allocating more ALP declaration votes (and similarly for the Liberal Party). THE ACCURACY OF THE CHECKING EXERCISE: SUMMARY Both the 1991 and the 1994 Electoral Districts Boundaries Commissions have checked their proposed seats against the results at the previous election, in order to check that the party which won the majority of votes at that previous election would have won the majority of the proposed seats (and therefore would have won government). Because it was not possible to properly allocate preferences before 1993, the checking exercise carried out on the 1991 redistribution contained a margin of then-unavoidable error, which I estimate was up to 0.5 percentage points in any electorate. While the 1991 Commission estimated that the redistributed seats would have given government to the Liberal Party in 1989, both Newland and Unley could have been estimated as being either Labor or Liberal. Page 17 of 41 : Dated February 20,1998

21 6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA While the Commission supplemented the checking exercise with well-informed judgement, the checking exercise alone should probably not have been accorded the status it has had in recent years. The 1993 and 1997 data, and future data, will make it possible to estimate the effect of a redistribution with more accuracy simply because the State Electoral Office now produces data showing the two party preferred percentages for the ordinary vote at each booth. Declaration votes still need to be allocated in a purely statistical exercise, and the result will still relate to the previous election rather than the next one, but the checking exercise itself will be more accurate now. To summarise the previous section on declaration votes: the declaration vote count includes a similar proportion of informal votes to the ordinary vote count, so it does seem to be quite legitimate to ignore informal votes in any checking exercise; State-wide, declaration votes reduced the LIB two party preferred vote by 0.1 percentage points in 1993 and increased it by 0.3 percentage points in 1997, but in some seats the effect of including declaration votes was much larger, changing the final two party preferred percentages by up to 1.6 percentage points. The change does not seem to be consistently in the same direction (and therefore does not seem to be predictable); the effect generated by including declaration votes into the count seems to be due to the difference between the two party preferred vote for declaration votes and for ordinary votes, not due to the number of declaration votes; because of the unpredictability of declaration votes compared to the ordinary vote count, and the size of the effect, we do need to recognise declaration votes in any checking exercise; the most accurate way to include declaration votes is to allocate the ALP declaration votes to booths on the basis of those booths' share of the electorate's ordinary ALP votes, and to allocate LIB declaration votes to booths on the basis of those booths' share of the electorate's LIB ordinary votes. The question of how we relate the checking exercise to a future election remains unanswered. APPLYING THE FAIRNESS CRITERIA LOOKING TOWARDS THE NEXT ELECTION As stated earlier, the 1994 Commission rejected the probability of a uniform swing, and even of a non-uniform but proportionate swing - a situation where seats which swung the most from 1989 to 1993 would swing back the most at the next election. If there is no way to predict how far individual seats will swing with a given State-wide swing, then how is it possible to judge that a redistribution will result in the party winning the majority of the vote winning enough seats to govern? In recent years this question has been addressed in several ways: studying the patterns of actual swings at South Australian State elections; applying probability theory to swings; and studying the characteristics of individual seats. Page 18 of 41 : Dated February 20,1998

22 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA PATTERNS IN ACTUAL RESULTS: SWINGS IN RECENT SOUTH AUSTRALIAN STATE ELECTIONS. At the 1977 State election, the ALP won 53.4% of the two party preferred vote throughout the State (see Table 13), and the Liberal Party won the remaining 46.6%. At the 1979 State election, the ALP won 44.9% of the two party preferred vote throughout the State, and the Liberal Party won 55.1%. The average Statewide swing from 1977 to 1979 was therefore 8.5% to LIB, on a two party preferred basis (see Table 14). At the 1982 election the swing was 6 percentage points back towards the ALP. More recently the Statewide swings were 2.3% to the ALP in 1985, 5.3% back to LIB in 1989 and then another 8.9% to LIB in In order to show these swings for individual seats, the redistributions need to be taken into account. Tables 13 and 14 use: actual results for the State elections of 1977, 1979 and 1982 as these elections were all fought on the 1976 EDBC boundaries; Jaensch's estimates of the 1982 election results using the seats ordered by the EDBC in 1983 and then the actual results from the 1985 and 1989 State elections on those 1983 boundaries; my estimates of the 1989 election results using the seats ordered by the 1991 EDBC and then the actual results from the 1993 State election on those 1991 boundaries; and finally my estimates of the 1989 election results using the seats ordered by the 1994 EDBC, EDBC estimates of the 1993 election results using the seats ordered by the 1994 EDBC, and then the actual results from the 1997 State election on those 1994 boundaries. Where the final result for a particular seat did not come down to an ALP:LlB two party preferred contest, Tables 13 and 14 use rethrow data published by the State Electoral Office from the 1985 State election onwards. Prior to 1985, there are no rethrow data, so for 1977 to 1982, I have used Jaensch's estimated two party preferred estimates. There is one problematic result, namely Flinders in 1977, when the final two candidates preferred in Flinders were ALP and NAT. Because the NAT candidate was acting in coalition with the LIB government, Jaensch did not calculate an estimated ALP:LlB set of figures, using instead the ALP:NAT figures as if they were equivalent to ALP:LlB data. 1. Uniform swings? Table 14 shows quite clearly that the swings in individual seats were often fairly different from the Statewide swing; that is, there was not, strictly speaking, a uniform swing at any of the elections from 1977 to While at most elections the general direction of swings was uniform, the size of the swings in individual seats was very variable. At the 1979 election for example, the overall swing was 8.5% towards the Liberal Party but the swing in each seat ranged from 2.5% in Goyder to 21.3% in Mitcham. Swings were not even consistent in their direction in 1985, ranging from 9.8 % towards the ALP in the safe Liberal seat of Alexandra, to 8.0% towards the Liberal Party in the safe ALP seat of Whyalla. Still, if the swing in even just the more marginal seats was uniform, the concept of a uniform swing might remain useful. The pendula at Figure 2 show the state of the seats after the 1977, 1979 and 1982 State elections. This set shows the movements in seats which were drawn up as a result of the 1976 EDBC. The pendula incorporate the definitions of marginal, safe etc used by the Australian Electoral Commission and the State Electoral Office, as follows: marginal seats are those which require a swing of under 6% to change hands; fairly safe seats are those which require a swing of at least 6% but under 10% to change hands; and safe seats are those which require a swing of 10% or more to change hands. Page 19 of41 : Dated February 20,1998

23 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA The first pendulum in Figure 2 shows the state of the seats after the 1977 State election (i.e. going in to the 1979 election) and the second pendulum of Figure 2 shows the outcome of the 1979 election. There was an average swing Statewide of 8.5 percentage points away from Labor at the 1979 election (see Table 14), so if there had been a uniform swing, the Liberal Party would have won 4 extra seats at the 1979 election (Morphett which would change hands with a swing of 0.5% to LIB, Mawson and Todd which would both change hands with a swing of 6.6% to LIB, and Brighton which would change hands with a swing of 8.2% to LIB). In fact the Liberal Party won 7 extra seats. Morphett, Mawson, Todd and Brighton did all change to the Liberal Party. The swings to LIB were: Morphett 5.7 percentage points, Mawson 9.5, Todd 11.1, Brighton These seats did not each record the State average swing (8.5% to LIB) so there was no uniform swing even within this group, but the right seats did change hands. However, the Liberal Party also won another three seats which would not have been predicted to have changed hands under a uniform swing of 8.5% to LIB. These seats were Henley Beach (ALP 59.3% at the 1977 election, but swung 10.3 percentage points to LIB in 1979), Newland (59.8% ALP in 1977 but swung 15.9 percentage points to LIB in 1979) and Norwood (60.2% ALP in 1977 but swung 10.3 percentage points in 1979). Interestingly, Unley did not change hands although it sat between Newland and Norwood on the 1977 pendulum, so it cannot be said that the extra seats were the next three seats on the pendulum. The outcome of the 1979 election, shown in the second pendulum of Figure 2, is also the state of the seats going into the next State election, in The outcome of the 1982 State election is shown in the third pendulum of Figure 2. There was an average swing across the State of 6.0% back towards the ALP at the 1982 election, so a uniform swing would have delivered 7 seats to the ALP: Norwood, Henley Beach, Mawson, Todd, Brighton, Morphett and Mt Gambier. In fact, the ALP won only 4 of these, plus one unexpected seat. As expected, Norwood, Henley Beach, Mawson and Brighton changed hands and became ALP seats, but Mt Gambier, Morphett and Todd swung by only very small amounts (Mt Gambier 3.4 percentage points; Morphett 3.8; Todd 3.2) and stayed Liberal. Then the ALP also picked up Newland even though that seat would not have been predicted to have changed hands under a uniform swing of 6.0% to the ALP. The second set of pendula, at Figure 3, show the state of the seats after the 1982, 1985 and 1989 State elections. This set show the movements in seats which were drawn up as a result of the 1983 EDBC, and the figures shown in this 1982 pendulum are Jaensch's estimates of the way that voters in these (newly redistributed) seats voted at the (previous) 1982 State election. Table 14 shows that the average swing across the State at the 1985 State election was 2.3% towards the ALP. If a uniform swing had occurred we would expect that Hayward, Newland, Bright and possibly Fisher would have changed hands and become ALP seats at the 1985 election. Indeed, as we see from the second pendulum, those seats did move to the ALP in 1985, but in addition, the ALP won Adelaide. At the next State election, in 1989, there was a Statewide average swing of 5.3% back to the Liberal Party. A uniform 5.3% swing to LIB would have sent 6 ALP seats to the Liberal Party (namely Adelaide, Fisher, Newland, Bright, Hayward and Unley). The 1989 election was interesting in that the largest swings against the ALP government occurred in their own heartland - and also in safe Liberal seats such as Mt Gambier (swing Page 20 of 41 : Dated February 20, 1998

24 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA of 12.6% to LIB), not in the marginal seats. So while 5 of the 6 seats did change hands they all recorded swings to LIB well below the State average of 5.3% (3.9% in Adelaide, 4.2% in Fisher, 1.6% in Newland, 2.6% in Bright and 3.7% in Hayward). Unley's swing of only 2.7% was not enough to make it change hands. The third set of pendula, at Figure 4, show the state of the seats after the 1989 and 1993 State elections. This set show the movements in seats which were drawn up as a result of the 1991 EDBC, and the figures shown in this 1989 pendulum are my estimates of the way that voters in these (newly redistributed) seats voted at the 1989 State election. The figures for 1993 are actual results. At the 1993 State election, the Statewide average swing was 8.9% to LIB. The pendula in Figure 4 show that if every seat had swung by 8.9% then Unley, Newland, Norwood, Florey, Mawson, Peake, Mitchell, Kaurna, Torrens, Elder, Hanson, Wright, Giles, Lee and perhaps also Reynell would all have become Liberal seats. Reynell did change hands, as did all of the other seats, with the exception of Giles, which swung by only 3.4% and remained with Labor. Finally, the pendula at Figure 5 show the state of the seats after the 1989, 1993 and 1997 State elections, based on the boundaries ordered by the 1994 EDBC. The figures for 1989 are my estimates, those for 1993 are the EDBC's estimates and those for 1997 are actual results. The first pendulum in Figure 5 shows the state of the seats after the 1989 State election - i.e. going into the 1993 election. Table 14 shows that the Statewide swing at the election of 1993 was 8.9% towards the Liberal Party, and from the first pendulum it is clear that with a uniform swing 16 ALP seats would have become LIB seats (Mawson, Hartley and Unley, Newland, Florey, Mitchell, Kaurna and Norwood, Peake, Torrens, Hanson, Elder, Wright, Reynell, Giles and Lee). Such a sweep would have left the ALP with only nine seats. The second pendulum in Figure 5 shows that the ALP was indeed left with few seats. Of the 16 seats expected to change hands with a uniform swing of 8.9%, 15 seats actually did move and only Giles stayed relatively firm, experiencing only a 3.5% swing to LIB while around it Reynell swung by 10.1 %, Lee by 9.1 %, and even the traditionally solid Labor seat of Playford swung by 9.1 % (although it remained a Labor seat). At the 1997 State election no-one was surprised when the direction of the average Statewide swing was back towards the ALP, but at 9.4% the size of the swing was a surprise to many. The second pendulum in Figure 5 shows that a uniform swing of this size would have at least 12 seats and possibly thirteen moving back to Labor (Lee, Hanson, Kaurna, Reynell, Wright, Elder, Peake, Torrens, Norwood, Frome, Stuart, Mawson and perhaps Mitchell). The first nine did swing (Lee, Hanson, Kaurna, Reynell, Wright, Elder Peake, Torrens and Norwood all became ALP seats). The next three seats - Frome, Stuart and Mawson - did not change hands and then Mitchell did change hands, Colton did not, Florey did change, Unley did not and Hartley was terribly close but finally did not. If a uniform swing is one in which each seat swings by the same number of percentage points, then it is quite clear from these pendula that there has not been a uniform swing in South Australian State elections in at least the past twenty years. If uniform Statewide swings had occurred: in 1979 the Liberal Party would have won an extra 4 seats, but they won 7; in 1982 the ALP would have won 7 extra seats but they won only 5; in 1985 the ALP would have won 4 extra seats but in fact they won 5; in 1989 the Liberal Party would have won 6 extra seats, but they only won 5; in 1993 the Liberal Party would have won 15 seats and in fact they won 14; and in 1997 the ALP would have won 12 seats and in fact they won 11. Page 21 of 41 : Dated February 20, 1998

25 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA It is also true that the seats which changed hands were not always the ones which would have been predicted to change on the basis of a uniform swing: in 1979 Unley should have changed hands and Norwood should not have changed; in 1982 Mt Gambier, Morphett and Todd should have changed hands and Newland should not have changed; in 1985 Mt Gambier should have changed hands and Adelaide should not have changed; in 1993 Giles should have changed hands; and in 1997 Frome, Stuart and Mawson should have changed hands and Mitchell and Florey should not. Clearly, the uniform swing is not a particularly accurate model for predicting how many seats will change hands with a given average Statewide swing, nor is it a particularly accurate model for predicting which seats will change hands with a given average Statewide swing. 2. Proportionate swings? It is difficult to compare the performance of individual seats across Table 14 because of the boundary changes which occurred - Stuart in 1977 for example was not much like Stuart in Still, some seats did seem to swing more than others - Newland moved to the ALP in 1979 and then back to the Liberal Party in 1982 with an above-average swing on each occasion. More recently Bragg, Bright, Coles, Elizabeth, Fisher, Florey, Hart, Hartley, Heysen, Mitchell, Napier, Ross Smith, Torrens and Waite all swung to the ALP (in 1993) and then back (in 1997) by above-average amounts. From Table 14 it does appear that the seats which swung most were often the most likely seats to have swung back again at the next election. It is possible to calculate the hypothetical two party preferred results which would have occurred if there had been a proportionate swing back: that is, if the seats which had swung most from 1977 to 1979 had then swung back most in 1982, and so on. The results from three consecutive elections held on the same boundaries are required to construct this "swing and swing back" series, so it is possible to compare actual and hypothetical swings for (on 1976 EDBC boundaries), (on 1983 EDBC boundaries) and (on 1994 EDBC boundaries). Table 15A shows the hypothetical results for an average Statewide swing to the ALP of 6.0% in 1982, if it had been true that the swing from 1979 to 1982 in each seat had been proportionate to the swing which had occurred in that seat from 1977 to With such a hypothetical proportionate swing, in 1982 the ALP would have won 26 seats and the Liberal Party 21, and these would have been the same seats as the ALP and Liberal Party actually did win in 1982, with the exception of Todd (which was actually retained by the Liberal Party, but under a hypothetical proportionate swing would have fallen to the ALP). While a proportionate swing hypothesis for 1982 does not predict the two party preferred result in each seat very accurately at all, it does give a reasonable idea of the seats that would be won; that is. it does seem to be a reasonable approximation of the actual swing pattern for 1979 to By comparison, the uniform swing hypothesis was not at all useful for predicting which seats would change hands in 1982 (it predicted 7 seats would change hands, but in fact only 3 of those 7 changed hands and another one swung unexpectedly). Page 22 of 41 : Dated February 20,1998

26 () MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA If a proportionate swing model works quite well for , does it work for ? Table 15B shows the same exercise for that period. If the 5.3% Statewide swing away from the ALP in 1989 had been experienced in each seat in proportion to the size of the swing in that same seat from 1982 to 1985, the results would have been as shown in the 'Hypothetical 1989' column. The ALP would have won 22 seats (it actually won 24) and the Liberal Party would have won 25 seats (it actually won 23). The hypothetical proportionate swing generates the wrong results in 1989 for Henley Beach, Todd and Unley, which would hypothetically have changed hands to Liberal (in fact they stayed Labor), and also for Mt Gambier, which would hypothetically have changed hands to Labor (in fact it stayed Liberal). The hypothesis that the swing in a given seat at one election is likely to be proportional to the swing in that same seat at the previous election was reasonable for explaining the results in 1982 but not very helpful at all in relation to the results in Was 1997 any better? Table 15C shows the actual and hypothetical results for If the 9.4% Statewide swing towards the ALP in 1997 had been experienced in each seat in proportion to the size of the swing in that same seat from 1989 to 1993, the results would have been as shown in the 'Hypothetical 1997' column. The ALP would have won 27 seats (it actually won 21) and the Liberal Party would have won 20 seats (it actually won 26). The hypothesis that a seat which swings in one election will swing back at the next election in proportion to that first swing, was a reasonable description of what actually happened in 1982 but was really not useful at all in 1989 nor in Characteristics of the individual seats? When the 1994 EDBC checked its redistributed seats against the voting figures for 1993, it needed to check that the redistribution would be fair. A fair redistribution would give 24 seats to the party or group which won 50.1 % of the two party preferred vote, so the EDBC aimed to "ensure that the Labor Party would win 14 more seats under the 1994 redistribution with a swing in its favour of 11 %" (EDBC 1994 Report; para 8.11). The pendulum at Figure 1 shows the Commission's own estimates of the new seats. With a uniform swing of 11 % the ALP would expect to win another 14 or perhaps 15 seats (Florey is on 11% swing-to-iose), giving the ALP 24 or 25 seats. But the Commission argued that because they were rural seats Stuart and Frome would be unlikely to change hands even with an average Statewide swing much higher than the swing required in those particular seats (Stuart would need a swing to the ALP of 9% and Frome 8.2%). The ALP submission to the 1994 Commission had included a table showing that over the period from 1977 to 1991, rural seats had swung less readily than metropolitan seats (EDBC 1994 Transcript of Public Hearings, vol. 1; p38). The Commission took up this point (EDBC 1994 Report section 8). I do not have access to the ALP's table but I imagine that it would look something like Table 16. Table 16 confirms that, on average, country seats have not swung as far as the average seat across the State. The swing across the State at the 1979 election was 8.4 percent to LIB, but in country seats this was dampened, and was only felt as a swing of 4.9% to LIB. In 1982, the swing of 6.0% to the ALP Statewide was again dampened in country areas where it was expressed as a swing of only 2.1 % to the ALP. In 1985 the Statewide swing was 2.1 % towards the ALP but in country areas it was expressed as a swing of only 1.8% to the ALP. Page 23 of 41 : Dated February 20, 1998

27 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA Then in 1989, country areas expressed a swing of 5% to LIB compared to a swing of 5.1 % to LI B across the State. In 1993 the Statewide swing was 8.9% to LIB but in country areas this movement was dampened and a swing of only 4.9% to LIB was recorded in country seats. Most recently, in 1997 the Statewide swing was 9.4% towards the ALP but seats in country areas recorded a swing of only 7.5% to the ALP. Table 16 confirms that at all of the last six elections, the average swing recorded in country seats has been lower than the average swing recorded in seats in metropolitan areas or over the State as a whole. Most of the country seats over this period have been safe seats for the Liberal Party (or the National Party). It is possible that the result observed in Table 16 is not because country seats somehow swing less than metropolitan seats, but has another explanation - perhaps safe seats do not swing as much as marginal seats. Perhaps the group of "All metropolitan seats" includes a higher proportion of seats that swing because they are marginal seats. Table 17 shows the swing recorded in marginal, fairly safe and safe seats at each election from 1979 to Although I have normally followed the convention of referring to the swing at an election as a swing to or away from the ALP, in Table 17 it is clearer to express the swing as being either towards the ALP or towards LIB. I have also included both estimates for the swing from 1989 to 1993; i.e. the estimate for the swing from 1989 to 1993 when the seats are delimited using the boundaries which applied in 1989 and also the estimate for the swing from 1989 to 1993 when the seats are delimited using the boundaries which applied in Table 17 does not confirm any hypothesis that safe seats swing less. ALP safe seats swung more than the Statewide average swing in 1979, 1982, 1989 and again in 1997, and LIB safe seats swung more than the Statewide average swing in Can we expect that any individual country seat will swing less than the average swing for metropolitan seats or for the Statewide swing? Table 18 shows that most country seats do swing less than the State average, but this is not true all the time. In 1982 the average swing across the State was 6.0% towards the ALP, but in Kavel the swing was 7.6% to the ALP. In 1985 the average swing across the State was 2.3% towards the ALP but in country seats the results were very varied. Alexandra recorded a swing of 9.8% towards the ALP, Custance 3.7%, Flinders 3.4%, Goyder 5.6% and Light 5.7%. (And Whyalla - an ALP seat - recorded a swing of 8% away from the ALP.) In 1989 the average swing across the State was 5.3% towards the Liberal Party. Most country seats swung less than that, but Flinders was right on the State average, Mt Gambier swung by 12.6% to LIB, Stuart by 8.4% to LIB and Whyalla by 6.1% to LIB. In 1993 only Finniss recorded a swing above the State average of 8.9% to LIB. Expressed in terms of the 1991 boundaries the swing in Finniss was 9.0% to LIB, or on the 1994 boundaries it was 9.8% to LIB. In 1997 the average swing across the State of 9.4% to the ALP, was equalled in Hammond and exceeded in Finniss (10.2% towards the ALP). Table 18 shows 105 swings for country seats between 1979 and 1997; of these cases 92 swung less than the Statewide swing and 13 swung by the same amount or more. Is a probability of 88% good enough to prove a rule? Page 24 of 41 : Dated February 20, 1998

28 6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA My conclusions from Tables 16, 17 and 18 are that it is fair to expect that at any given election the swing in country areas will be lower than in metropolitan areas, but it should be borne in mind that not all seats will behave as if they are aware of the rule. The common idea is that safe seats swing least and marginal seats swing the most; this was not confirmed by Table 17. Table 17 is also a fairly clear proof that uniform swings have not occurred in South Australian elections over the past 20 years. It is difficult to see any clear pattern in the results from Table 17, but it may be possible to infer that the swings gained in the marginals are actually unlikely to be the greatest swings of a given electoral campaign, not because these seats are inherently less likely to swing, but because they usually contain the median seat and are therefore the sites of the most attention from both of the major parties (and often from minor parties and independents as well). The boxed outlines of Table 17 (mostly in the centre of the table) show the swing for that sector of the pendulum in which the median seat was located before the election: for example in 1977 the ALP fairly safe seat of Brighton became the median seat. By definition, the median seat will always belong to the party which won the election most recently. In 1977 the ALP won 27 of the 47 seats - including the median, Brighton - and therefore won government. GOing into the 1979 election then, the task for the ALP was to retain all of its seats, of which those from Brighton down to the most marginal seat required the most protection (see Figure 2 again). Brighton would fall with a swing in that seat of 8.2% to LIB. The task for the Liberal Party was to win all of the ALP seats at least as far up the pendulum as Brighton, but in order to make sure of the result the next most marginal seats after the median (Henley Beach, Newland, Unley, NorwOOd) would also have been targeted. The LIB campaign achieved a Statewide average swing of 8.5% to LIB, felt most strongly not in the most marginal seats (the swing to LIB in the single ALP marginal - Morphett - was only 5.7%), but in the group of fairly safe ALP seats (which included Brighton) and also in the safest ALP seats. The result was that the Liberal Party won government in All of the seats which went into the 1979 election as LIB seats stayed LIB, and the swing in these seats was lower than the average swing for that election. This pattern of swings is consistent with the idea of a strong LIB campaign especially in the ALP marginals and the fairly safe ALP seats, and a strong ALP defence of its most marginal seats. Going into the 1982 State election the Liberal Party had been in government and the ALP would have been aiming for the median seat which was then Todd (LIB marginal requiring a swing of 4.7% to change hands). The ALP achieved a 6% Statewide swing back to the ALP which was expressed most forcefully in the LIB marginal seats and also in those seats the ALP already held. The ALP took government. In 1985 the Liberal Party had been in Opposition and needed to win either Unley or Todd (both about 52% ALP after the redistribution). The 2.3% Statewide swing to the ALP had its biggest impact in the ALP marginals and in the safe LIB seats. The ALP retained power. To win the 1989 State election the Liberal Party required the median seat of Unley (marginal ALP, requiring a swing of 5.1 % to LIB to change hands). The Statewide average swing of 5.3% to LIB which they achieved was experienced least forcefully in the marginal ALP seats and most forcefully in safe ALP seats and in marginal or safe LIB seats. In effect. the required swing to LIB was achieved but in the wrong seats, and not enough seats changed hands. In 1993 the Liberal Party required just one more seat. Unley (the median) was a marginal ALP seat requiring a swing of under 1 % to change hands. The eventual 8.9% Statewide swing to LIB was expressed most forcefully in the marginal ALP seats and in the Liberal seats other than the very safest group. The Liberal Party took government. Page 25 of 41. Dated February 20, 1998

29 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA In 1997 there were so many marginal and fairly safe LIB seats that the median seat was a safe Liberal seat. The eventual swing of 9.4% to the ALP was expressed most strongly in ALP seats. The Liberal government retained power. Table 17 also does show quite clearly that when there has been a Statewide swing to the ALP, then the ALP marginals have swung more than the LIB marginals, and when there has been a Statewide swing to LIB then LIB marginals have swung more than the ALP marginals; there seem to be a tendency for marginal seats to become less so. Is this evidence of the ability of hard-working Members in marginal seats to make their seats safer? The implications for predicting the outcome of an election are unsettling: marginals can be expected to swing more if they are moving up the pendulum than if they are moving down the pendulum. THEORETICAL APPROACHES TO SWINGS 1. Fischer and Charnock Fischer's 1994 paper for the Electoral Districts Boundaries Commission proposes a method for checking whether a redistribution is fair, based on the probability that a given seat will swing. His method requires us to assume that "the swing in the two party preferred vote between elections is normally distributed with a standard deviation of 2.5 percentagepoints" (Fischer 1994:p6). Fischer's paper does not provide evidence that this assumption can be made in relation to South Australian State elections. Firstly, have swings in South Australian State elections really been normally distributed? Figures for swings at each State election since 1977 have been provided earlier (Tables 13A and 13B). These data use Jaensch's and my own estimates to take into account the redistributions between the 1982 and 1985 elections and the 1989 and 1993 elections. Table 19 looks at several of the statistical characteristics of the data shown in Table 13A and 13B. These characteristics have been chosen because they indicate whether we should accept the assumption that swings at State elections in South Australia are generally normally distributed, with a variation of 2.5 percentage points. Sprent (1977) states that: "Generally normality implies that, in all but very small samples, small deviations of observations from the sample mean are more common than large ones.... we may get a good picture of what normality implies by comparing the magnitude of all deviations from the sample mean with the mean absolute deviation. This is obtained by dividing the sum of all the magnitudes of deviations from the mean by the number of observations... As a rough rule of thumb the normality assumption is reasonable if about half, or perhaps a few more, of the deviations have a magnitude less than (the mean absolute deviation) and none are more than three or four times (that amount).... Another indication of normality is that the numbers of positive and negative deviations from the sample mean are about the same." (Sprent, 1977: p87). Therefore, Fischer's assumption that seat-by-seat swings at State elections in South Australia are normally distributed with an average variability of 2.5 percentage points, would be acceptable if: Page 26 of 41. Dated February 20, 1998

30 6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA the mean absolute deviation was about 2.5 percentage points - in fact Table 19 shows that over the six periods it ranged from 2.0 to 3.2; about 23 or 24 of the 47 deviations from the mean in each swings column fall below the mean - Table 19 shows that they have done so (ranging from 22 below the mean in the 1979 to 1982 swing, to 31 below the mean in the swing from 1982 to 1985); there are very few values (ideally none at all) falling above 3 or 4 times the mean absolute deviation. Table 19 shows that there were five falling above 3 times the mean absolute deviation, and only one fell above 4 times the mean in all of the 47 seats over the six elections; and finally roughly half of the deviations in each series were positive and roughly half negative - and indeed they were. On the basis of these results, it is at least arguable that Fischer's assumption of normality with a 2.5% variation in relation to Federal swings can be applied to South Australian State elections over the past twenty years. In particular, the assumption would seem to apply even to the swings which were extraordinarily large. To look at whether individual seats would be likely to change hands with a given Statewide swing, Fischer's 1991 paper outlined the process, although in this paper on Federal elections he assumed that "swings within an election will be normally distributed with a standard deviation of 3.5 percentage points" (Fischer 1991, p302, my emphasis). "To measure whether a seat will change hands, consider the seat, named K, held by the government with a two party preferred margin of 4.8 per cent. Suppose that there is no overall swing against the government. When there is a 3.5 percentage point standard deviation of the swing from seat to seat, we see that K is 4.8/3.5=1.37 standard deviations from changing hands to the opposition. There is thus an 8.5% chance of K swinging to the opposition (since the probability of being 1.37 standard deviations or more above the mean of a standard normal distribution is 0.085). If there is a 2 per cent overall swing against the government around Australia, with a 3.5 percentage-point standard deviation as before, K requires a 2.8 percentage point swing over and above the national swing to change hands. The probability of the swing being the additional 2.8 percentage points or greater, is 21.2 per cent (the probability of being 2.8/3.5=0.8 standard deviations from the mean, assuming normality as before). To find out what would happen on average if there were a 2 per cent overall swing against the government, we perform this calculation for all seats. " (Fischer, 1991: p302). To see how well this method would predict which seats are likely to change hands, it seemed reasonable to apply Fischer's method to the South Australian State seats as they existed just before the last two State elections. Table 20 shows the results of this exercise. I have calculated the probability of each seat changing hands, given an election with a swing of 8.9% to LIB (for 1993) and an election with a swing of 9.4% to the ALP (for 1997). The swing-to-iose figures for the 1993 table are my estimates of the 1989 voting results applied to the 1991 boundaries, and the swing-to-iose figures for the 1997 table are my estimates of the 1993 voting results applied to the 1994 boundaries. Table 20 shows that, according to Fischer's method, with an 8.9% swing to LIB in 1993, 14 ALP seats had a probability of 50% or more of swinging to the Liberal Party. These seats were Lee, Giles, Wright, Hanson, Elder, Torrens, Kaurna, Mitchell, Peake, Mawson, Florey, Norwood, Newland and Unley. (Perhaps not surprisingly, these were the same seats which would be predicted to swing if a simple uniform swing of 8.9% to LIB is applied to the 1989 pendulum.) The list is roughly, but not entirely, correct. In 1993 all of the above seats changed hands except Giles, which remained with the ALP. In addition, Reynell became a Liberal seat, although Fischer's method gives it a probability of changing hands of only 48.4%. Page 27 of 41 : Dated February 20, 1998

31 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA A similar result appears for the 1997 part of Table 20. Assuming a swing of 9.4% to the ALP and applying Fischer's method, 12 LIB seats had a probability of 50% or more of changing hands: Lee, Hanson, Kaurna, Reynell, Wright, Elder, Peake, Torrens, Norwood, Frome, Stuart and Mawson. Once again, these are the same seats we would expect to change hands if a uniform swing of 9.4% to the ALP had been applied to the 1993 pendulum. And once again the list is not quite right: Lee, Hanson, Kaurna, Reynell, Wright, Elder, Peake, Torrens and Norwood all swung to the ALP but Frome, Stuart and Mawson did not. In addition, there were a few changes which Fischer's method would not predict. Mitchell was a fairly safe LIB seat going into the 1997 election; it required a swing of 9.5% and Fischer's method gives it a probability of changing hands with a Statewide swing of 9.4% to the ALP, of 48.4% - a little unlikely. Mitchell recorded a swing of 10.3% and did change hands. Florey was a safe LIB seat with a swing-to-iose going into the 1997 election of 11.0%. The probability of Florey changing hands with a Statewide swing of 9.4% to the ALP, was % - highly unlikely. Florey recorded a swing of 12.3% and did change hands. Sitting between Florey and Mitchell on the pendulum going into the 1997 election, Colton was a safe LIB seat, requiring a swing-to-iose of 10.7%. Fischer's method gives it a probability of changing hands with a Statewide swing of 9.4% to the ALP, of 30.15% - unlikely but not impossible. It recorded a swing of 6.6%, and remained in LIB hands. Here we encounter a problem with the application of Fischer's method to the results of actual electoral contests. This problem relates to the assumption in such statistical abstractions, that each seat will be equally likely to swing by a given amount, and that the swing in one seat is independent of the swing in another seat. That is, we are unable to take account of our "real world experience" that, although there are always exceptions, country seats do generally swing less than city seats; that a marginal ALP seat will swing more if the Statewide swing is in favour of the ALP than if the swing is in favour of the Liberal Party (and a marginal LIB seat will swing more if the Statewide swing is in favour of the Liberal Party than if the swing is in favour of the ALP); that a party will concentrate its resources on winning the most marginal of its opponents' seats, and also that a party may be able to concentrate its resources to produce a sizeable swing in one seat but may not be able to stretch to the neighbouring seat as well. My feeling is that Fischer's method is probably no more useful than a uniform swing theoretical framework. A recent paper by Charnock proposes a similar probability model linking voting swings and seat changes. Charnock's model differs from Fischer's in that he proposes separate statistical functions for coalition seats and ALP seats, but he still assumes both that the distribution of swings in each group follows a normal distribution and that "swings in electorates are statistically independent random variables" (Charnock 1993: p341). Charnock applied his model to Federal swings from 1990 to His model predicted roughly the correct number of seats changing hands (11 to the coalition and 4 to the ALP compared to the actual result of 10 to the coalition and 4 to the ALP) but in the process he pointed to a major problem - the extreme variability of possible outcomes. ".. it is also extremely important to note the amount of random variation which could be reasonably expected. Overall, as stated in the previous paragraph, the model suggests an expected gain by the coalition of 7 seats, but with a standard deviation (for the net gain) of 3.2 seats. This implies that, for example, there was approximately a 90% chance of the coalition having a net gain of from 2 to 12 seats (inclusive). I would suspect that this is a larger degree of variability than most people would have anticipated. Variability such as this must, however, be recognised as a direct consequence of acknowledging a role for random fluctuations of swings (and hence in net seat changes)." Page 28 of 41 : Dated February 20, 1998

32 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA (Charnock 1993: p344) While the variability would be lower with a standard deviation of 2.5 rather than the 3.2 and 3.7 Charnock was obliged to use for Federal statistics, it would still be difficult to validate the fairness of any redistribution on the basis of a result which could range between 2 and 12 seats. The work by Fischer and by Charnock applying probability theory to electoral results has not yielded a theoretical framework which provides any more certainty than MacKerras's pendulum and the uniform swing. 2. The Cube Rule Political analysts and commentators often refer to the Cube Rule, which specifies the number of seats which should be won by a party gaining a given percentage of the vote, assuming fair boundaries and single-member electorates. It takes into account the observed fact that a party may win 50% of seats with 50% of the two party preferred vote, but will win more than 60% of seats with 60% of the two party preferred vote and may win almost all of the seats with much less than all of the vote. "The cube law is an attempt to provide a formula which reproduces the exaggeration of the winning party's majority... It states that the ratio between the percentages of seats won by the two parties is roughly the same as the ratio between the cubes of the percentage of votes cast for the two parties. If V and v are the percentages of votes cast and Sand s are the percentages of seats won by the two parties then the cube rule may be written as an equation: = V 3 S v 3 (Soper and Rydon, 1958: Note 6). The ALP submission to the House of Assembly Select Committee on the Constitution (Electoral Redistribution) Amendment Bill, described the rule in more detail. "... the Cube Law is a special case of a wider principle linking seats (S), electorates (E) and votes (V) in electoral systems whether based on single-member electorates (S = E) or multimember electorates (E = ns where n > 1). The general rule is SA = rva T SB LVB J It can be demonstrated that in the Anglo-Saxon single-member electorate (S = E) system, n = log V log E If n = 3, we have a reformulation of the Cube Law. It can be shown that the Cube Law approximates mathematical accuracy when the number of seats in the National Assembly is a cube root of the nation's population: S = 3.yp or the cube root of 2 times the voting population: S = 3.y[2P v l. On this basis, South Australia with a voting population of , should have a Parliament of 123 members, that is for a given voting population the "ideal" or "Cube Law" Assembly size can be calculated as above (=123). Since the actual Assembly size in South Australia is much less than this (=47) it follows that!qgjl will be greater than 3. log E In fact it is equal to log 941,368 = = = n, so the "Cube Law" for the log House of Assembly in South Australia becomes: Page 29 of 41 : Dated February 20,1998

33 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA SA= rvaf 5 SB LVBJ The point is that one can expect some mild over-amplification in South Australia's electoral results provided that malapportionment or other factors do not distort the position." (ALP 1990 : footnote 5.) Now that there are electors on the South Australian roll (as at October 1997) n is even bigger, at But in practice, the simpler cube (3.0) is used rather than the more precise 3.5 or Does the Cube Rule work in South Australian elections, or do we have so few seats that the errors are too high? Table 21 shows the number of votes actually won by the ALP and Liberal Party, at the State elections from 1977 to 1997, and the number of seats which each party won as a result. It also shows how many seats each party would have won at those elections if the Cube Rule had applied - that is if the boundaries had been statistically "fair". Table 21 shows since the Boundaries Commission has been required to take voting into account, the number of seats won by each party at elections have been exactly right. Over the longer term, the number of seats which the parties won in 1977, 1982, 1985, 1993 and 1997 all followed the Cube Rule very closely, but in 1979 and 1989 the Cube Rule would not have predicted the number of seats won by each party very well at all. That lack of alignment in 1979 and 1989 could be because the Cube Rule doesn't apply very well in South Australia, or it could be because the distribution of boundaries at the time of those two elections was not ideal. We know that the boundaries in 1989 were not ideally placed. These boundaries were seen before the 1989 election to present a problem, when Mr Griffin argued that under the boundaries as they then stood, and as they would apply at the election expected within months, the Liberal Party required 52% of the total State-wide two party preferred vote for House of Assembly seats in order to win government. The 1991 report of the EDBC accepted that the Liberal Party had been disadvantaged by those boundaries, and estimated that the disadvantage was of the order of one seat. The Cube Rule would imply a disadvantage of 3 seats (Table 21 shows that the Liberal Party should have won 26 seats rather than 23 with its 52% of the two party preferred vote), but it is possible that the Commission judged that the Liberal Party lost two seats because of the different population size of many of the electorates, and one because of the differential concentration of Liberal supporters in several seats. The Cube Rule does seem to be useful even in the context of the relatively small number of State seats here in South Australia. It gives us a fairly accurate idea of how many seats a party would win with a given percentage of the vote (and, by implication, a given swing) but has the disadvantage of not showing us which seats a party would win with a given percentage of the vote, or which seats would change hands with a given swing. Accepting that, we can calculate the number of Assembly seats which should be won by each party with a given percentage of the two party preferred vote, and this is shown in Table 22. I wanted to look at the smallest percentage of the vote which either of the two major parties would need to win in order to win a first - and then each subsequent - seat.. Of course winning a seat needs to be seen as a contest, and winning a first seat requires that a party wins one more than the number of votes required to win half a seat - i.e. enough votes to prevent the other party from winning all of the seats. Table 22 shows, for example, that to win its first seat a party would need to win 18.08% of the two party preferred vote plus one vote. Rounding up to 18.1 % would take that Page 30 of 41 : Dated February 20, 1998

34 () MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA required extra vote into account, so 18.1 % of the two party preferred vote across the State would give a party one seat, 24.3% would ~ive it two seats, 27.7% would give a party three seats, and so on. In order to win the 241 seat, a party requires one more vote than the number of votes required to win 23.5 seats; Table 22 shows that this is 50% of the two party preferred vote across the State plus one vote. Finally, Table 22 shows that a party would need to win almost 82% of the vote to win every seat. In 1997 the Liberal Party and the 3 non-labor Members won 51.5% of the two party preferred vote; they should have won 26 seats, which indeed they did. With a swing of, say, 5% away from the Liberal Party at the next election the Cube Rule would predict that the Liberal Party and the 3 non-labor Members (on 46.5% of the two party preferred vote) would win 19 seats. The 1994 Commission seems to have accepted that the Cube Rule is a useful guide to the optimal distribution of seats around the pendulum, under the circumstances of a fair electoral distribution. In late 1995, when the Commonwealth Parliament's Joint Standing Committee on Electoral Matters was enquiring into the process of Federal redistributions, one of the Commissioners (Mr Andy Becker, then South Australia's State Electoral Commissioner) appeared before the Committee as a private citizen. He was asked about the fairness test. "Naturally, of course, these things are new to us, and I think probably new universally. Most of the systems that we have seen in the past are top-up-systems to try to ensure that groups with 50 per cent plus or more of the vote, do get more than 50 per cent of the seats. It was an interesting exercise and one of the things for which we were not particularly well geared in South Australia, largely because of the lack of information. In the 1991 redistribution we did not have two party preferred figures on each polling place: we only had them by district. We had to make a lot of assumptions regarding declaration voting and how we should apply those across the state. The outcome of that redistribution, according to pundits, was about 0.4 per cent still in favour of the Labour Party, and I do not know that I could argue against that. I think when you are getting that close it is probably pretty hard to say exactly what the situation is. But given the situation at that stage, it probably was a pretty fair redistribution, and I think that the fairness test, as best we could apply it, operated. In 1993 we had an unbelievable landslide against the Labor Party, with the State Bank fallout and all that sort of thing. But the interesting thing about that was the cube law approach. The cube law stood fairly well firm... That is the only test you can apply. If it had not come out that way, I dare say we probably would have been saying it was such a flood that we could not predict that and therefore we cannot say whether or not the fairness test was right or wrong. I say at this stage - and I think Colin Hughes would agree - that we got it right the first time, and I think Malcolm MacKerras says we got it right the first time. And the cube law has shown that it does operate in the case of the South Australian elections, that the fairness test that was applied in 1991 was effective. The next question of course is: on the figures that we got at the last election, which was the landslide, will the redistribution stand up? So that remains to be seen, and we will not be seeing that until Depending on the swings and whatever, we might have to apply the cube law or it might become so simple that we would have the situation we had in 1989 and it would be very easily tested against the total vote outcome. If somebody has got 50 per cent of the vote plus one, and they have got 24 seats, then the fairness test has been shown to have worked." (Committee on Electoral Matters, 1995: EM 135.) At the State election of 1997, the ALP won 48.5% of the two party preferred vote across the State and fell just short of winning 22 seats. In the end, the ALP won 21 seats, the Liberal Party 23 and non-labor Members won 3 seats. The Cube Rule does seem to have applied in Page 31 of 41 : Dated February 20, 1998

35 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA THE CUBE RULE AND THE DEMOCRATS An interesting question arises in considering the Cube Rule table. Could the Cube Rule provide a measure of the fairness of a redistribution to the Australian Democrats, whose candidates won 9.1 % of first preferences at the State election of 1993 and 16.4% in 1997? The Cube Rule table (Table 22) is a theoretical table, based on two parties, not necessarily the ALP and the Liberal Party. It suggests that a party which won 18% or more of the two party preferred vote should win at least one seat if an electoral distribution is unbiased, so the question is whether the Australian Democrats - with 16.4% of the first preference vote in would have won 18% or more of the DEM : OTHER two party preferred vote (and hence one or more seats) in For Table 21 the results for seats in which the final two candidates were DEM and LIB or NAT and LIB were converted to results on an ALP: LIB two party preferred basis. In the same way then it should be possible to convert the results in all of those seats contested by the Democrats, to DEM : LIB or DEM : ALP two party preferred results. What is required is an indication of how ALP votes would be distributed between DEM and LIB candidates if the ALP candidate is to be excluded during the count; and an indication of how LIB votes would be distributed between DEM and ALP candidates if the LIB candidate is to be excluded during the count. In fact, at the 1997 State election, there were three safe LIB seats in which there were only three candidates and the final two candidates represented the Liberal Party and the Democrats, so we do have an indication of how ALP votes would be distributed if the ALP candidate was to be excluded. In Finniss, the ALP candidate's 2583 votes were allocated 9.3% to LIB and 90.7% to DEM. In Heysen, the ALP candidate's 3938 votes were allocated 8.3% to LIB and 91.7% to the Democrat candidate. In Waite, the ALP candidate's 3270 votes were allocated 14.8% to LIB and 85.2% to DEM. In 1993, there had been two seats where the ALP candidate's first preferences were allocated: Davenport where the ALP first preferences went 11.0% to the LIB candidate and 89.0% to the DEM candidate, and Waite where the ALP candidate's first preferences were allocated 42.6% to the DEM candidate and 57.4% to the Liberal sitting Member. With such a difference between these two seats in terms of the percentages of their ALP votes which were allocated to the Democrats, it was only possible to make tentative remarks about the LlB:DEM two party preferred results around the State. The 1997 results make Waite in 1993 look atypical. In 1997, there was also one safe ALP seat - Napier - where there were only three candidates and the final two candidates were ALP and DEM - i.e. the LIB candidate's first preference votes were distributed. In Napier the LIB first preferences were allocated 16.1% to DEM and 83.9% to LIB. It would not be unreasonable to calculate adem: ALP two party preferred vote in safe ALP seats on the assumptions that LIB votes were distributed in each of those safe ALP seats, and that they were distributed in the proportion which we found in Napier. Similarly, in safe LIB seats it would not be unreasonable to calculate adem: LIB two party preferred vote on the assumptions that ALP votes were distributed in each of those safe LIB seats, and that they were distributed in the proportions which we found in Finniss, or Heysen, or in Waite. As for estimating adem: OTHER vote in seats which are not safe ALP or safe LIB seats, there are no indicators in the 1997 State election results; my feeling is that the pattern of allocating ALP preferences in Finniss, Heysen and Waite, and the pattern of allocating LIB preferences in Napier could al/ be used to give outer limits of a reasonable estimate. Page 32 of 41 : Dated February 20, 1998

36 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA Table 23 shows the actual 1997 voting figures for each seat at that stage of the count just before the Democrat candidate was excluded. (I have allocated the OTHER votes on a 50:50 basis to the DEM and LIB columns.) The right hand columns of Table 23 show estimates of the ALP: DEM two party preferred vote in safe ALP seats if the LIB candidates had been excluded and the LIB first preferences distributed in the same way as occurred in Napier; the ALP: DEM or LIB : DEM two party preferred vote in those seats which were not safe ALP seats or safe LIB seats, if the LIB or ALP first preferences were distributed as occurred in Napier, or in Finniss, or in Heysen, or in Waite; and finally the LlB:DEM two party preferred vote in safe LIB seats if the ALP candidates had been excluded and the ALP first preference votes had been distributed in the same way as occurred in Finniss or in Heysen or in Waite. Using Table 23 it is possible to estimate a range within which adem : OTHER two party preferred result for the whole of the State after the 1997 State election might be found. The only indicator we have for estimating DEM : LIB two party preferred figures for safe ALP seats is Napier, but for seats which are not safe ALP or safe LIB seats there are four possibilities, and for safe LIB seats there are three. There are therefore 12 possible DEM OTHER estimates for Table 24 shows all of these estimates. Table 24 shows that an estimate of adem: OTHER two party preferred result for the 1997 State election would fall somewhere between DEM 45.6% : OTHER 54.4% and DEM 47.6% : OTHER 52.4%. With somewhere between 45.6% and 47.6% of the two party preferred vote Statewide, the Cube Rule (see Table 22) suggests that the Australian Democrats could have expected to win between 17 and 20 seats. Given that the Democrats did not win even one seat in 1997, these results are startling: how should they be regarded? Firstly, because the Democrat vote relies so heavily on the distribution of ALP and LIB preferences I think that we can only see the Democrat vote as part of a coalition. Secondly, any ALP ticket is likely to put the Democrats ahead of the Liberal Party (as any Liberal ticket is likely to put the Democrat candidate ahead of an ALP candidate). But my feeling is that this practice would change if the Australian Democrats looked likely to win many seats. Finally, the problem for the Democrats has been, in the past, that their support has been spread over several electorates. In 1993 there were only three seats in which the Democrats gained more than even 15% of the first preferences: Davenport (26.5%), Ross Smith (16.5%) and Waite (18.2%). In 1997, there were 31 seats in which the Democrats gained 15% or more of first preference votes, and indeed in 11 of these seats the DEM candidate recorded over 20% of first preferences. With 16% of all first preferences across the State in 1997 it could be argued that the Democrats must have support - in their own right - at least equal to the 18.1 % of the State's two party preferred vote which the Cube Rule requires before a seat can be fairly won. In their submission to the 1994 EDBC, the Australian Democrats did argue that the Commission should create a seat around Davenport and Waite which would be winnable for a candidate from their party (P. Black, 1994 EDBC Transcript of Public Hearings, p ). The Commission rejected this idea on the grounds that the relationship between votes and seats in the House of Assembly is not based on proportional representation, but on the ability of a party to concentrate its vote in an area to the extent necessary to win a seat. Page 33 of 41 : Dated February 20, 1998

37 () MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA "It would not be right, in our opinion, to create an electoral district in order to give a party representation in the House of Assembly that it could not ordinarily expect to get under the normal operation of our single-member electorate system. That would be to contrive a result on something like proportional representation grounds that the Parliament has plainly rejected." (EDBC 1994 Report: para 9.2). An alternative view could be that the Commission leaves itself open to accusations that it has deliberately split the Democrat vote between electorates. THE SITTING MEMBER ADVANTAGE In their submissions to the 1991 and 1994 Commissions (and in the public hearings), the ALP representatives argued that: a sitting Member has an advantage over a non-sitting candidate, increasing the two party preferred vote for the sitting Member's party at the next election by 1 % to 2%; the 1993 election had resulted in a Parliament with many more Liberal than ALP sitting Members; and therefore at the next election the Liberal Party would receive an advantage from having so many sitting Members. (J.Hill for the ALP, 1994 EDBC Transcript of Public Hearings, pp46, 48, 55.) In their submissions to both the 1991 and 1994 Commissions, the ALP compared the ALP:L1B two party preferred vote for the House of Assembly with the ALP:L1B two party preferred vote for the Legislative Council lodged at the same booths. The difference, according to the ALP, should be seen as a measure of the personal vote for the local Member representing that area in the House of Assembly. The ALP argued that "It's Labor's contention,... that this, this is the personal vote and in particular the Sitting Member advantage, is a relevant matter to be considered by the commission under section 83(2) when drawing boundaries." (J.Hili for the ALP, 1994 EDBC Transcript of Public Hearings, p55.) At the Commission's request, I commented on the methodology of calculating the personal vote tables. The ALP's methodology is to reduce the Assembly and the Council contests in a certain geographic area - a House of Assembly electorate - to a two party contest, ALP versus LIB. My conclusion from this methodology was that, because of the way that the personal vote advantage had been calculated: "... (t)he sum of the personal votes of the Liberal candidate and the ALP candidate must always equal zero. That means that a candidate who is very popular and gains a good personal vote will necessarily have to win that vote away from the other candidate. In one seat where there are two excellent candidates and in another seat where there are two hopeless candidates, the personal vote table will show one excellent candidate and one hopeless candidate with the same personal vote." (J. Newton, 1994 EDBC Transcript of Public Hearings, p ) That is, the personal vote calculations can only show the popularity of a Member compared to his or her opponent, not compared to any other Member. It is important to make the distinction between a Member's personal vote, and the advantage which sitting Members may have over non-sitting candidates who may not be blessed with party resources or the high media profile which may accrue to a sitting Member. To look only at a sitting Member's advantage over a non-sitting Member, it is necessary to compare the average level of personal vote gained by the group of Sitting Members who win their seats Page 34 of 41 : Dated February 20, 1998

38 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA again, with the average level of personal vote gained by non-sitting candidates who win their seats. At Table 25 is a summary of the "personal vote" calculations for sitting Members and new Members at the State elections of 1985, 1989, 1993 and I have excluded those seats where the final two candidates were not ALP and LIB candidates, because it is difficult to know just how to interpret the personal vote calculations in these contests. (For example, in Chaffey the final 2PP contest was actually NAT: LIB in 1997 but we have rethrow figures of ALP: LIB 29.0%: 71.0%. My estimate of the way that the voters of Chaffey voted for the ALP and Liberal Parties for the Legislative Council was ALP: LIB 33.9%: 66.1 %, giving LIB a higher vote in the House of Assembly of 4.9%. But it would be true to say that this 4.9% advantage relates as much to the ALP: NAT contest as it does to the ALP: LI B contest.) I have excluded those contests where the ALP and LIB candidates were both sitting Members (because neither sitting Member would have an advantage). Finally, it is important to that both contests are against candidates of a similar strength, so I have excluded the personal votes of new MPs who win their seats away from sitting Members. Table 25 shows that successful sitting MPs have, on average, had a personal vote at a State election of somewhere between 1 and 5 percent; new Members who did not have to compete against a sitting MP had much more varied personal vote scores ranging from 3 percent to minus 2 percent. When the average sitting MP's personal vote is compared to the average personal vote for a new MP who did not face a sitting Member, the advantage to the sitting Member has varied from almost nothing to 4.4 percent. It is very difficult to draw conclusions about the different levels of sitting Member advantage which accrue to Members of the ALP or the Liberal Party, and this is particularly so for new LIB MPs because most new LIB MPs have arrived in the Parliament after having beaten a sitting ALP Member (and were therefore excluded from the analysis). For the ALP, the difference between the personal vote of an ALP sitting Member and a new ALP Member who did not face a sitting MP candidate, has varied from almost nothing in 1985 to about 1 percent in 1989 and 1993 and then 4.4 percent in The 1997 data do not change the conclusions I drew from the 1985 to 1993 figures: "... 1 do not believe that the data support any hypothesis that the higher personal vote of ALP Members (sitting and new) is a result of their party being in government. It may be explained just as readily by different campaign techniques... Similarly I do not believe that the data would support any average figure to quantify the sitting Member advantage. The figures have been too variable and there have been too few new Members over the period for which I have data, to give any sort of credibility to such a figure. In addition, the figures for Ministers show that it is quite possible to be disadvantaged by being a sitting Member (for too long, for example)." (J. Newton, written submission to the 1994 EDBC, listed as Exhibit 107.) The 1994 EDBC decided against taking into account a sitting member advantage or a personal vote, for many reasons, including that: it is not possible to know which of the members who won seats at a given election would stand for those seats at the next election; the opposing parties can be expected to campaign more strongly in an electorate with a sitting member; the idea that the difference between the 2PP vote for the Council and for the Assembly is only due to the local Member is unproven; and Page 35 of 41 : Dated February 20, 1998

39 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA the size of a Member's personal vote depends on the calibre of the opposing candidate. (EDBC 1994 Report: para 7.6). My own feeling about a personal vote which a Member may build up over time in his or her own constituency is that it should perhaps be seen as akin to a campaign skill, albeit built up over an extended period rather than just the period of a campaign. A personal vote or a sitting Member advantage belongs to the Member, not the electorate, and I do not believe that it is the sort of thing that the Commission should take into account. THE NUMBER OF MARGINAL The Commission's job must be to set the seats around the pendulum in such a way that a party cannot win the median seat without also winning the majority of the vote, but it must not be so difficult to win the median seat that the party would need more than a small majority of the Statewide vote. Given that some seats will not swing as easily as others, and that we cannot necessarily predict which seats these will be, it seems prudent to me to have a grouping of seats around the median seat. But the 1994 Report states that the Commissioners "agree with the Labor Party that it is undesirable to have too much bunching of seats on the pendulum, especially in the few percentage points above and below the median seat. We have adjusted the political balance in a number of instances to make the percentage steps between the closely marginal seats more even." (1994 EDBC Report: para 8.13). Perhaps the difference between the Commission's attitude and mine is a matter of judgement about how much grouping is too much bunching. In the end the Commission cannot prevent a result like that of 1989, in which the Liberal Party received the majority of the two party preferred vote across the State but not in the right seats. No amount of reasoned diligent work by the Commission will be able to take the relative strength of the Parties' campaigns into account - and nor should it. It is the Commission's job to create a level playing field for the candidates and the parties, and both the 1991 and the 1994 Commissions have worked extraordinarily hard to do this. In 1993 the result was so much in favour of the Liberal Party that it was difficult to say how many seats they would have won with only a 50.1 % share of the two party preferred vote, but in 1997 the result was much closer. Looking back, both the 1993 and the 1997 results can be judged now to have been fair results; each party's vote translated into the number of seats which the Cube Rule would have predicted. The Act does not require the Commission to "ensure, as far as practicable" that the party which wins 65% of the vote will win 65% of the seats, and the party which wins 35% of the vote should win 35% of the seats. This could only be achieved by proportional representation or by a top-up system such as the West German and now the New Zealand system. Page 36 of 41 : Dated February 20, 1998

40 6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA SUMMARY AND CONCLUSIONS "Exercise of judgement based upon wide familiarity with the situation, and great caution in the statement of conclusions from the data, are the beginning of wisdom in statistical work." E.E. Day, 1930, Statistical Analysis, MacMillan, New York, p382. After each State election in South Australia, an Electoral Districts Boundaries Commission is required to examine State electorates with a view to redistribution. In deciding whether boundaries need to be amended, the Commission is required to take into account many factors, including "as far as practicable" making sure that the boundaries are fair to all parties "so that, if candidates of a particular group attract more than 50 per cent of the popular vote..., they will be elected in sufficient numbers to enable a government to be formed." (Constitution Act (SA) s.83(1 ).) This fairness requirement is unique to South Australia. It imposes an obligation on the Commission to attempt to draw boundaries which will be fair at a future election, but whether a given redistribution will actually meet the requirements of section 83(1) will only ever become apparent once that future election has been held. In the past, the Commission has checked a proposed set of boundaries against past election figures. This checking exercise has its own problems. The 1991 Commission checked its proposed boundaries against the 1989 election results and found that they would have given government to the Liberal Party which had won 50% or more of the vote; my own estimate at the time, and those of the ALP and the Liberal Party, was that although several of the seats would be very marginal, if the new boundaries had been in place in 1989 the Labor government would have been returned with 24 seats. (Jaensch gave both parties 23 seats and found Newland too hard to call.) The different estimates arose, I believe, from different methods for distributing preferences, as well as from different decisions when booth results needed to be split between new electorates. The difficulty with the distribution of preferences which existed in 1991 (and which has since been removed), was that figures showing the distribution of preferences were available for an electorate as a whole, but not for each booth. In the absence of any better guide, it was common practice to assume that the pattern of distribution of preferences in a group of booths would be the same as that for the electorate as a whole. Now that the State Electoral Office produces actual two party preferred counts for each booth it is possible to calculate the error, which this assumption contained. In general, I suspect that two party preferred estimates for a given electorate made with data prior to 1993 need to be treated as if they had a margin of error of up to 0.5 percentage points (page 9). This means that, while I would not argue with the Commission's conclusion that its 1991 boundaries would have given 24 of the 47 House of Assembly seats to the Liberal Party, I also believe that the other publicly available estimates made at the time - which gave 24 seats to the ALP - are quite defensible. While checking a proposed set of boundaries against past election figures is no guarantee of the fairness of those boundaries at a future election, it is at least a good guide, and so the checking exercise itself should contain as little room as possible for dispute or error. That is Page 37 of 41 : Dated February 20, 1998

41 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA why the State Electoral Office's action at the 1993 State election, making available a two party preferred count for each booth, is a vital step. Not only has it made more accurate calculations possible but it has also reduced the scope for disagreement (between parties or political analysts) about the proposed seats. There is one other source of possible distortion, albeit of smaller proportions, namely declaration votes. While they have generally been included in checking exercises in the past, the availability of two-party preferred figures for each booth now makes it tempting to concentrate on booth figures and ignore the (less-numerous) declaration votes. Including or excluding declaration votes generally changed the winning party's 2PP vote by 0.1 percentage points in 1993 (Table 8A) and 0.3 percentage points in 1997 (Table 8B). However, there were quite striking variations between seats, of up to +/-1.0 percentage points in 1993 and up to +/-1.6 percentage points in Excluding declaration votes from a redistribution checking exercise would therefore mean that the estimated 2PP for any new seat would need to be treated as if it had a margin of error of up to 1.6 percentage points. For most purposes it would seem better to avoid that margin of error simply by including declaration votes in the exercise. The difficulty is in knowing how best to allocate declaration votes to booths. After reviewing several methods (Tables 12A to 120) I believe that the most accurate method is to allocate ALP declaration votes according to the distribution across the electorate of all ALP voters who lodged their votes at booths, and allocate LIB declaration votes across the electorate according to the distribution of LIB voters who lodged their votes at booths (page 16). This method reflects the observed difference between declaration voters and ordinary voters but sees ALP declaration voters as more like other ALP voters than like LIB declaration voters (and LIB declaration voters as more like other LIB voters than like ALP declaration voters). It also produces correct two party preferred percentages for the seat as a whole. It is clear that the State Electoral Office's provision of two party preferred figures for each booth, and a more accurate method of including declaration votes in a checking exercise, can now be used to deliver the best possible estimates of how the voters in proposed electorates voted at previous elections. How voters in the new electorates will vote at the next election is more difficult to see. Looking at the results of the past six State elections, spanning the past 20 years, it is quite clear that there has not been a uniform swing at any South Australian State election over this period, and the seats which changed hands were not always the ones we would predict on the basis of a uniform swing (Table 14 and Figures 2 to 5). Applying the swing-to-iose information for each new seat (which is produced by the checking exercise) to a pendulum, can give us a fairly good guide to how many seats would change hands with a given swing, but probably not which seats would change hands. As an alternative, a hypothesis that the swing in a given seat at one election is likely to be proportional to the swing in that same seat at the previous (or next) election was not very helpful (Tables 15A to 15C). Finally, attention to the characteristics of individual seats or groups of seats was not infallible. A hypothesis was made that rural seats swing less readily than metropolitan seats. This was found to be true in general terms: it is true that at any given election rural seats have, on average, swung less than metropolitan seats (Table 16). But individual rural seats have not all learnt to abide by the rule, so it is quite possible that a given rural seat will swing more than expected (Table 18). Page 38 of 41 : Dated February 20, 1998

42 6 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA The common belief that marginal seats swing wildly and safe seats swing very little, was also shown to be misleading (Table 17). Indeed, the bracket of marginal seats usually contains the median seat which is the target of both parties, and so it is likely that the contest within this bracket of seats will usually produce quite a low net swing for seats within the bracket (page 23). It is also clear that the average size of the swing in the bracket of marginal seats varies according to the direction of the swing (Table 17). A Statewide swing to the ALP has consistently produced larger swings in marginals that the ALP already holds than in LIB marginals, and vice versa. This would be consistent with the idea that the contest in the bracket of seats which contains the median seat can be expected to be intense, and the swings in that bracket lower. My conclusion was that marginals can be expected to swing more if they are moving up the pendulum than if they are moving down the pendulum (page 25). While it is interesting to look at how the seats swing, it would be nane to assume that the size and direction of swing is somehow contained in the nature of each seat. That would mean ignoring the role of the Parties' campaigns, or attributing swings in individual seats to luck. Clearly, which seats swing - and by how much - is a function not only of the characteristics of the seat itself but also of the pressure applied by the parties' campaigns. Fischer has proposed a method for checking whether a redistribution is fair, based on the probability that a particular seat will swing. The method relies on an assumption that each seat will be equally likely to swing by a given amount, and that the swing in one seat is independent of the swing in another seat. However, when this method is used to apply an 8.9% swing to the South Australian seats as they stood just before the 1993 election it was wrong in three cases, and when used to apply a 9.4% swing to the seats as they appeared just prior to the 1997 State election, it was wrong in five cases (Table 20). Fischer's method is not particularly helpful. Charnock has proposed a more refined probability model linking voting swings and seat changes, but it relies on the same assumptions as Fischer's model. Charnock has also pointed out the extreme variability of possible outcomes under a probability model (page 27). An alternative method which has often been held up as showing the fairness of a redistribution, is the Cube Rule, which specifies the number of seats which should be won by a party gaining a given percentage of the vote, assuming fair boundaries and single-member electorates. Indeed, it does seem to apply in the South Australian case: the number of seats which the parties won in 1977, 1982, 1985, 1993 and 1997 followed the Cube Rule very closely, but in 1979 and 1989 the Cube Rule would not have predicted the number of seats won by each party very well at all (Table 21). It is probably not a coincidence that these two elections have been recognised as having the largest numbers of electorates going into the election well above or below quota. The Cube Rule does seem to be useful even in the context of the relatively small number of State seats here in South Australia. It gives us a fairly accurate idea of how many seats a party would win with a given percentage of the vote (and, by implication, a given swing) but has the disadvantage of not showing us which seats a party would win with a given percentage of the vote, or which seats would change hands with a given swing. While the Cube Rule is generally considered in terms of an ALP:LlB two party contest, it could be applied to a DEM:LlB or ALP:DEM contest. At the State election of 1997 the Australian Democrats won 16.4% of first preferences across the State. With this level of support it could be argued that the Democrats must have support - in their own right - at least equal to the 18.1 % of the State's two party preferred vote which the Cube Rule requires before a seat can be fairly won. Page 39 of 41 : Dated February 20, 1998

43 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA As a member of a coalition, I estimate that at the State election of 1997 the Australian Democrats could have received somewhere between 45 and 48% of the DEM:OTHER two party preferred vote (Table 24). While DEM candidates still were not able to win a single House of Assembly seat in 1997, they ran second in six seats when first preferences were counted, and then when second and subsequent preferences were counted out DEM candidates were runners-up in seven seats. These seats were Davenport, Finniss, Heysen, Kavel, Napier, Schubert and Waite. Some of these seats share boundaries. The Commission has considered the argument that it should create a seat which the Democrats have a chance of winning, and has rejected that view on the grounds that the Act does not require the Commission to "contrive a result on something like proportional representation grounds that the Parliament has plainly rejected" (page 32). An alternative view could be that the Commission leaves itself open to accusations that it has deliberately split the Democrat vote between electorates. In both 1991 and 1994 the Commission deliberated over the question of whether to take into account an electoral advantage which may accrue to a party with many sitting members. Personal vote calculations made by comparing the Assembly and Council contests within a given area, can only show the popularity of a Member compared to his or her opponent, not compared to any other Member involved in a different contest. Examination of personal vote data for the 1985, 1989 and 1993 State elections showed that the average personal vote of a sitting MP has generally been higher than the average personal vote of a successful non-sitting candidate (Table 25). Figures for Liberal Party sitting MPs are not as clear, but in the case of ALP sitting Members, the advantage to a Sitting MP ranged from about 1 % to 3%. I do not believe that the data support an argument that a sitting Member advantage of a given size exists and must be taken into account by the Commission. The Commission's job must be to set the seats around the pendulum in such a way that a party cannot win the median seat without also winning the majority of the vote, but it must not be so difficult to win the median seat that the party would need more than a small majority of the Statewide vote. Given that some seats will not swing as easily as others, and that we cannot necessarily predict which seats these will be, it seems prudent to me to have a grouping of seats around the median seat. But in the end the Commission cannot prevent a result like that of 1989, in which the Liberal Party received the majority of the two party preferred vote across the State but not in the right seats. It is the Commission's job to create, as far as possible, a level playing field for the candidates and the parties. Methods by which the Commission can check its proposed boundaries will always be vital to the Commission's own credibility, and I hope that this exercise will be recognised as a contribution towards improving the methodology. At the same time the Commission needs, I feel, to make it clear to concerned individuals and parties that not only is there really no way of predicting how voters will swing in particular seats, but also that the fairness requirement does not require the Commission to take into account those factors in an election which are under the parties' (or candidates') control, such as the relative strength of the parties' campaigns, their concentration in marginal or winnable seats, or which candidates the parties choose. Page 40 of 41 : Dated February 20, 1998

44 o MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA REFERENCES Constitution Act (SA) Australia. Parliament. Joint Standing Committee on Electoral Matters, 1995, Official Hansard Report, 4 October Australian Labor Party (SA Branch), 1990, Submission to the House of Assembly Select Committee on the Constitution (Electoral Redistribution) Amendment Bill, ALP (SA Branch), Adelaide. Charnock, D., 1993, "Voting swings and seat changes: a stochastic model", Australian Journal of Political Science (1993) vol 28: pp Day, E.E. 1930, Statistical Analysis, MacMillan, New York. Fischer, A., 1991, "Swings and Gerrymanders", Electoral Studies, Vol 10 (4), December 1991, pp Fischer, A., 1994, The Redistribution of Electoral Boundaries, copy courtesy of the 1994 EDBC. Hughes, 0., 1984, "Uniform swing revisited: Further comments on MacKerras", Politics, vol. 19 (2), November Jaensch, D., 1977, The 1977 Elections in South Australia: a statistical analysis, Flinders University Politics Department Occasional Monograph No.6, Adelaide. Jaensch, D., 1979, The 1979 Elections in South Australia a statistical analysis, Flinders University Politics Department Occasional Monograph No.7, Adelaide. Jaensch, 0, 1983, The 1982 Elections in South Australia a statistical analysis, Flinders University Politics Department Occasional Monograph No.8, Adelaide. Karmel, P.H. and Polasek, M., 1970, Applied Statistics for Economists, Sir Isaac Pitman (Aust) Ply Ltd, Melbourne. MacKerras, M., 1973, "The swing: variability and uniformity", in Mayer, H., Robertson for the Australasian Political Studies Association, Sydney. Labor to Power: Australia's 1972 Election, Angus & Newton, J, 1992, The 1991 Electoral Redistribution in South Australia: 1989 voting statistics transferred to the new House of Assembly boundaries, Parliamentary Library of South Australia Research Service Information Paper No.11, Parliamentary Library, Adelaide. Rydon, J., 1973, "Swings and Predictions: the analysis of Australian electoral statistics", in Mayer, H, Labor to Power: Australia's 1972 Election, Angus & Robertson for the Australasian Political Studies Association, Sydney. Soper, C.S. and Rydon, J., 1958, "Under-Representation and Electoral Prediction", Australian Journal of Political History, Vol. 4(2). South Australia. Electoral Districts Boundaries Commission (EDBC), 1991, 1991 Report, Government Printer, Adelaide. South Australia. Electoral Districts Boundaries Commission (EDBC), 1994, 1994 Report, Government Printer, Adelaide. South Australia. Parliament. House of Assembly Select Committee on the Constitution (Electoral Redistribution) Amendment Bill, 1990, Report, SA Parliamentary Paper 157 of South Australia. Parliament. South Australian Parliamentary Debates (SAPO), The Parliament, Adelaide. South Australia. State Electoral Department, 1978, General Elections 1977: Statistical Returns, SA Parliamentary Paper 144 of South Australia. State Electoral Department, 1980, Periodical and General Elections 1979: Statistical Returns, SA Parliamentary Paper 132 of South Australia. State Electoral Department, 1982, Periodical and General Elections 1982: Statistical Returns, SA Parliamentary Paper 145 of South Australia. State Electoral Department, 1986, General Elections 1985: Statistical Returns, SA Parliamentary Paper 145 of South Australia. State Electoral Department, 1991 (a), Periodical and General Elections 1989: Statistical Returns, SA Parliamentary Paper 161 of South Australia. State Electoral Department, 1991 (b), Referendum on the (Electoral Redistribution) Act 1990, South Australian Parliamentary Paper 167 of South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 1993 and By-elections 1994, State Electoral Office, Adelaide. South Australia. State Electoral Department, 1993, Everyday Electoral Terms, The Department, Adelaide. Sprent, P., 1977, Statistics in Action, Penguin, Melbourne. Page 41 of 41 : Dated February 20, 1998

45 PARLIAMENTARY LIBRARY OF SOUTH AUSTRALIA RESEARCH SERVICES MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA A revised version of Information Paper No.15, incorporating data from the 1997 State election. [TABLES] JENNI NEWTON INFORMATION PAPER 16

46 1998, Parliamentary Library of South Australia Not to be reproduced in part or whole without the written permission of, or acknowledgement to, the Parliamentary Library of South Australia. This information paper has been prepared by the Research Service of the South Australian Parliamentary Library. While all care has been taken to ensure that the material is both accurate and clearly presented, the responsibility for any errors remains with the author. February 1998 ISSN

47 ----.~ ~ INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA FIGURE 1: PENDULUM AFTER THE STATE ELECTION OF IF THE NEW HAD BEEN IN EXISTENCE (BOUNDARIES COMMISSION ESTIMATES) ALP SWING LIB 2PP TO LOSE 2PP Flinders 79.4 MacKillop 79.3 Bragg Chaffey ::-" 75.3 Finniss Hammond ---_.-,, Heysen 74.3 Waite Kavel,,---- f Goyder 72.4 Morphetl & Schubert Davenport 71.5 Gordon Fisher _ f---,,?8.2 Bright f Coles Newland ---_.-.- SAFE Lig~,,_ SAFE 14 ~.. 1!-delaide FAIRLY SAFE _ 57.4 Norwood Torrens Napier MARGINAL Hartley._- Price Unley 60.9 Florey Ramsay Colton MEDIAN Taylor Mitchell I _ 59.2 Mawson - Hart Stuart FAIRLY Frome SAFE._--,,- Spence Peake Giles MARGINAL ---,,- f Elder Wright 1--- f Reynell..._---- Elizabeth 52.7 Ross Smith & Playford Kauma Hanson Lee V...QJ.. 'Note that the actual swing to lose in any given seat would be the difference between 50.0% and the 2PP vote for that seat, plus either 1 vote or 0.1 %. SOURCE: EDBC 1994 Report, Appendix 17

48 INFORMATION PAPER 15: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 1: DISTRIBUTION OF PREFERENCES AT EACH BOOTH, LEE ELECTORATE, STATE ELECTION 1993 ACTUAL RESULTS 2PP IF PREFERENCES STATE ELECTION 1993 ACTUAL ALL DISTRIBUTED DIFFERENCE DISTRIBUTION IN LINE WITH THE BETWEEN FIRST TWO PARTY OF PREFERENCES DISTRIBUTION FOR THE THE METHODS PREFERENCES PREFERRED SEAT AS A WHOLE BOOTH NAME ALP LIB Others ALP LIB total to ALP to LIB All ALP LIB total ALP LIB ALP LIB (No) (No) (No) (No) (No) formals (%) (%) (%) (No) (No) formals (%) (%) (No) (No) % total formals LEE Albert Park Grange (s2) Royal Park Royal Park South Seaton Semaphore Park Semaphore Park South West Lakes Shore Woodville Central (s2) Woodville West Total booth votes Declaration votes LEE ELECTORATE NOTES: (s2) means that the Woodville Central booth operated for Lee and also for another electorate. The figures in the table above relate to votes lodged at the Woodville Central booth by electors who lived in Lee. SOURCE: Actual data are from figures published in: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections 1994, State Electoral Office, Adelaide. Estimated data are my calculations based on applying the distribution of preferences in the seat as a whole to the first preference votes in each booth.

49 INFORMATION PAPER 15: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 2: ESTIMATED AND ACTUAL TWO PARTY PREFERRED RESULTS FOR GROUPS OF BOOTHS, STATE ELECTION TWO PARTY PREFERRED RESULTS LEE Actual Estimated 1993 (actual count) (preferences allocated on proportional basis) STATE ALP LIB TOTAL ALP LIB ALP LIB TOTAL ALP LIB ELECTION (no.) (no.) (%) (%) (%) (no.) (no.) (no.) (%) (%) Semaphore Park Semaphore Park South West Lakes Shore Northern booths Albert Park Grange (52) Royal Park Royal Park South Seaton Woodville Central (52) Woodville West Southern booths NOTES: (52) means that the Woodville Central booth operated for Lee and also for another electorate. The figures in the table above relate to votes lodged at the Woodville Central booth by electors who lived in Lee. SOURCE: Actual data are from figures published in: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections 1994, State Electoral Office, Adelaide. Estimated data are my calculations based on applying the distribution of preferences in the seat as a whole to the first preference votes in each booth.

50 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 3' DISTRIBUTION OF PREFERENCES AT EACH BOOTH, STATE ELECTION 1993 ACTUAL RESULTS ACTUAL ESTIMATED 2PP STATE ELECTION 1993 DISTRIBUTION IF PREFERENCES DIFFERENCE BETWEEN THE FIRST TWO PARTY OF ALL FOLLOWED THE METHODS PREFERENCES PREFERRED PREFERENCES AVERAGE FOR SEAT (Actual minus Estimated) BOOTH NAME ALP LIB Others ALP LIB total to ALP to LIB ALP LIB total ALP LIB ALP LIB % ave. (No) (No) (No) (No) (No) formals (%) (%) (No) (No) formals (%) (%) (No) (No) formals ADELAIDE Adelaide Adelaide Hospital Adelaide South Adelaide West Broadview (sl) Collinswood Lower North Adel Nailsworth North Adel East Prospect Prospect South Walkerville Total booth votes Declaration votes ADELAIDE BRAGG Beaumont Beulah Park (sl) Burnside Glen Osmond Glenunga Kensington Park Linden Park Marryatville Rose Park (s2) Stonyfell (sl) Tusmore Total booth votes Declaration votes BRAGG BRIGHT Brighton Brighton South Hallett Cove Hallett Cove South Marino Nth Brighton (sl) Seacliff Sheidow Park (sl) West Brighton(sl) Total booth votes Declaration votes BRIGHT CHAFFEY Barmera Berri Cobdogla Glossop Loxton Loxton North CHAFFEY CONTINUES SOURCE: calculated from data published in: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections 1994, State Electoral Office, Adelaide.

51 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 4' BOOTHS AT WHICH THE ERROR WOULD HAVE BEEN 0.15 PER CENT OR MORE IN 1993 ACTUAL RESULTS ACTUAL ESTIMATED 2PP STATE ELECTION 1993 DISTRIBUTION IF PREFERENCES DIFFERENCE BETWEEN THE FIRST TWO PARTY OF ALL FOLLOWED THE METHODS PREFERENCES PREFERRED PREFERENCES AVERAGE FOR SEAT (Actual minus Estimated) Booth Name ALP LIB Others ALP LIB total to ALP to LIB ALP LIB total ALP LIB ALP LIB % ave. (No) (No) (No) (No) (No) formals (%) (%) (No) (No) formals (%) (%) (No) (No) formals CHAFFEY Loxton COLES Athelstone Athelstone West COLTON West Beach FLINDERS Kirton Point Port Lincoln Port Lincoln South Wudinna GILES Why Norrie NW Why Stuart South HART Taperoo HARTLEY Magill North HEYSEN Aldgate MACKILLOP Bordertown NAPIER Smithfield PEAKE Flinders Park Lockleys Lockleys North Torrensville (s 1) PLAYFORD Ingle Farm PRICE Queenstown WAITE Myrtle Bank SOURCE: calculations made for Table 3: full set available on request.

52 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 5: ESTIMATED 2PP VOTE FOR THE THREE MOST MARGINAL, 1989 STATE ELECTION RESULTS APPLIED TO THE 1991 BOUNDARIES ALP estimates Liberal Party estimates Jaensch estimates Parliamentary Library (Newton) estimates NEWLAND ALP 50 to 53 ALP : 50.0 ALP 50.6 UNLEY ALP 50 to 53 ALP 51.2 ALP 50.5 ALP 50.4 COLTON ALP 50 to 53 LIB 50.9 LIB 52.0 LIB 50.9 SOURCE: Newton, J., 1992, The 1991 Electoral Redistribution in South Australia: 1989 voting statistics transferred to the new House of Assembly boundaries, Research Service Information Paper No.11, Parliamentary Library of South Australia, Adelaide.

53 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 6: DECLARATION VOTES AS A PROPORTION OF ALL VOTES IN EACH SEAT, STATE ELECTION 1993 STATE ELECTION ADELAIDE BRAGG BRIGHT CHAFFEY COLES COLTON CUSTANCE 8.7 seat did not continue DAVENPORT ELDER ELIZABETH EYRE 13.0 seat did not continue FINNISS FISHER FLINDERS FLOREY FROME GILES GORDON GOYDER HAMMOND new seat 16.2 HANSON HART HARTLEY HEYSEN KAURNA KAVEL LEE LIGHT MACKILLOP MAWSON MITCHELL MORPHETT NAPIER NEWLAND NORWOOD PEAKE PLAYFORD PRICE RAMSAY REYNELL RIDLEY 11.9 seat did not continue ROSS SMITH SCHUBERT new seat 12.8 SPENCE STUART new seat 17.2 TAYLOR TORRENS UNLEY WAITE WRIGHT AVERAGE, ALL SOURCES: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections 1994, State Electoral Office, Adelaide. Official but as-yet-unpublished 1997 State election data from the State Electoral Office

54 .. INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 7 A: EFFECT OF DECLARATION VOTES ON THE TOTAL FORMAL VOTE IN EACH SEAT, 1993 SWING TO LOSE PERCENT AGE OF VOTES EFFECT OF WHICH WERE FORMAL INCLUDING ALP LIB DECLARATION ORDINARY ALL DECLARATION (%) (%) VOTES VOTES VOTES VOTES PRICE RAMSAY '---- HART TAYLOR SPENCE ELIZABETH ,----"-~ PLAYFORD GILES ROSS SMITH NAPIER AVERAGE, ALP LEE HANSON REYNELL KAURNA ELDER WRIGHT PEAKE FROME ,, f-----,-- EYRE TORRENS NORWOOD MITCHELL MAWSON r----'- '-,------, -,----~ _- FLOREY COLTON UNLEY HARTLEY ADELAIDE COLES LIGHT NEWLAND BRIGHT FISHER MORPHETT GORDON DAVENPORT (rethrow) GOYDER WAITE (rethrow) KAVEL FINNISS HEYSEN CUSTANCE RIDLEY MACKILLOP BRAGG CHAFFEY (rethrow) FLINDERS (rethrow) AVERAGE, LIB AVERAGE, ALL SOURCE: calculated from data published In: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections State Electoral Office, Adelaide.

55 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 7B: EFFECT OF DECLARATION VOTES ON THE TOTAL FORMAL VOTE IN EACH SEAT, 1997 SWING TO LOSE PERCENTAGE OF VOTES EFFECT OF WHICH WERE FORMAL INCLUDING ALP LIB DECLARATION ORDINARY ALL DECLARATION (%) (%) VOTES VOTES VOTES VOTES PRICE SPENCE HART TAYLOR NAPIER 21.2 (rethrow) RAMSAY ROSS SMITH ELIZABETH PLAYFORD GILES TORRENS LEE " PEAKE KAURNA HANSON REYNELL WRIGHT ELDER FLOREY MITCHELL NORWOOD AVERAGE, ALP HARTLEY STUART FROME COLTON UNLEY MAWSON ADELAIDE ~ "- "" "- BRIGHT LIGHT COLES NEWLAND "- FISHER " "-- HEYSEN (rethrow) WAITE (rethrow) GORDON (rethrow) MORPHETT DAVENPORT (rethrow) HAMMOND SCHUBERT (rethrow) KAVEL (rethrow) FINNISS (rethrow) GOYDER BRAGG MACKILLOP (rethrow) CHAFFEY (rethrow) FLINDERS (rethrow) AVERAGE, LIB AVERAGE, ALL SOURCE: calculated from official but as-yet-unpubhshed results of the State election 1997, from the State Electoral Office.

56 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 8A: EFFECT OF DECLARATION VOTES ON THE TWO PARTY PREFERRED VOTE IN EACH SEAT, 1993 SWING 2PP 2PP EFFECT OF TO ORDINARY ALL DECLARATION LOSE VOTES VOTES VOTES ALP LIB ALP LIB ALP LIB ALP LIB (%) (%) (%) (%) (%) (%) (%) (%) PRICE RAMSAY HART TAYLOR SPENCE ELIZABETH PLAYFORD GILES ROSS SMITH NAPIER A VERAGE, ALP LEE HANSON REYNELL KAURNA ELDER WRIGHT ~- PEAKE FROME EYRE TORRENS NORWOOD MITCHELL MAWSON FLOREY 10.5 ~---~ COLTON UNLEY HARTLEY ADELAIDE COLES LIGHT NEWLAND BRIGHT FISHER MORPHETT GORDON DAVENPORT (rethrow) GOYDER WAITE (rethrow) KAVEL FINNISS HEYSEN CUSTANCE 24.6 RIDLEY MACKILLOP BRAGG CHAFFEY (rethrow) FLINDERS (rethrow) AVERAGE, LIB AVERAGE, ALL SOURCE: calculated from data published In: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections 1994, State Electoral Office, Adelaide.

57 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 8B: EFFECT OF DECLARATION VOTES ON THE TWO PARTY PREFERRED VOTE IN EACH SEAT, 1997 SWING 2PP 2PP EFFECT OF TO ORDINARY ALL DEC LARA TlON LOSE VOTES VOTES VOTES ALP LIB ALP LIB ALP LIB ALP LIB (%) (%) (%) (%) (%) (%) (%) (%) PRICE SPENCE HART TAYLOR NAPIER 21.2 (rethrow) RAMSAY ROSS SMITH ELIZABETH PLAYFORD GILES TORRENS LEE PEAKE KAURNA HANSON REYNELL WRIGHT ELDER FLOREY MITCHELL NORWOOD A VERAGE, ALP HARTLEY STUART FROME COLTON UNLEY MAWSON ADELAIDE , ~--~-.~,-- BRIGHT LIGHT COLES NEWLAND FISHER _.---- HEYSEN (rethrow) WAITE (rethrow) GORDON (rethrow) MORPHETT DAVENPORT (rethrow) HAMMOND SCHUBERT (rethrow) KAVEL (rethrow) FINN ISS (rethrow) GOYDER BRAGG MACKILLOP (rethrow) CHAFFEY (rethrow) FLINDERS (rethrow) AVERAGE, LIB A VERAGE, ALL SOURCE: calculated from official but as-yet-unpublished results of the State election 1997, from the State Electoral Office.

58 ---~-~~- ---.~ ~-~.-- ~-.--.-, ~ ~--- INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 9A: DIFFERENCE BETWEEN DECLARATION VOTES AND ORDINARY VOTES IN EACH SEAT,1993 SWING 2PP 2PP DIFFERENCE EFFECT OF TO ORDINARY DECLARATION (DECLARATION DECLARATION LOSE VOTES VOTES MINUS ORDINARY) VOTES ALP LIB ALP LIB ALP LIB ALP LIB ALP LIB (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) ALP RETAINED THESE PRICE RAMSAY HART TAYLOR SPENCE ELIZABETH PLAYFORD GILES ROSS SMITH NAPIER AVERAGE, THE ALP RETAINED ALP LOST THESE TO LIB LEE HANSON REYNELL KAURNA ELDER WRIGHT PEAKE TORRENS NORWOOD MITCHELL MAWSON FLOREY UNLEY NEWLAND AVERAGE, THE ALP LOST TO LIB LIB RETAINED THESE FROME EYRE COLTON HARTLEY ADELAIDE COLES LIGHT BRIGHT FISHER MORPHETT GORDON DAVENPORT (rethrow) GOYDER WAITE (rethrow) KAVEL FINNISS HEYSEN CUSTANCE 24.6 RIDLEY MACKILLOP BRAGG CHAFFEY (rethrow) FLINDERS (rethrow) AVERAGE, LIB RETAINED AVERAGE, ALL SOURCE: calculated from data published In: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections 1994, State Electoral Office, Adelaide... ~ _.-

59 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 9B: DIFFERENCE BETWEEN DECLARATION VOTES AND ORDINARY VOTES IN EACH SEAT, 1997 SWING 2PP 2PP DIFFERENCE EFFECT OF TO ORDINARY DECLARA TION (DECLARATION DECLARA TION LOSE VOTES VOTES MINUS ORDINARY) VOTES ALP LIB ALP LIB ALP LIB ALP LIB ALP LIB (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) ALP RETAINED THESE PRICE SPENCE HART TAYLOR NAPIER 21.2 (rethrow) RAMSAY ROSS SMITH ELIZABETH PLAYFORD GILES AVERAGE, THE ALP RETAINED LIB LOST THESE TO THE A'LP TORRENS LEE ~ PEAKE KAURNA HANSON REYNELL WRIGHT ELDER FLOREY MITCHELL NORWOOD AVERAGE, LIB LOST TO THE ALP LIB RETAINED THESE HARTLEY STUART FROME COLTON UNLEY MAWSON f'.delaide _ ~--- BRIGHT LIGHT COLES NEWLAND FISHER HEYSEN (rethrow) WAITE (rethrow) GORDON (rethrow) MORPHETT DAVENPORT (rethrow) HAMMOND SCHUBERT (rethrow) KAVEL (rethrow) FINNISS (rethrow) GOYDER BRAGG MACKILLOP (rethrow) CHAFFEY (rethrow) FLINDERS (rethrow) AVERAGE, LIB RETAINED AVERAGE, ALL SOURCE: calculated from official but as-yet-unpubllshed results of the State election 1997, from the State Electoral Office.

60 INFORMATION PAPER 16 : MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE loa: TWO PARTY PREFERRED VOTE IN EACH SEAT IF THERE HAD BEEN THE SAME NUMBER OF DECLARATION VOTES IN EACH SEAT, STATE ELECTION RANK ALL VOTES LESS EQUALS ACTUAL APPLY ADD ORDINARIES GIVES US FINAL LESS EQUALS I ORDINARIES DECSEFFECT DECS2PP rom1decs FINAL COUNT 2PP ORDINARIES DECSEFFECT SWING ACTUAL ACTUAL ACTUAL ACTUAL HYPOTHETICAL ACTUAL HYPOTHETICAL ALL VOTES ACTUAL EXPECTED TO ALL ORDINARY EFFECT OF DECLARATION DECLARATION ORDINARY VOTES (actual ordinary votes ORDINARY EFFECT OF LOSE VOTES VOTES DEC VOTES VOTES VOTES (ind. ord. ticket votes) plus hypothetical dec votes) VOTES DEC VOTES ALP LIB ALP LIB ALP LIB ALP LIB ALP LIB ALP LIB TOTAL ALP LIB TOTAL ALP LIB TOTAL ALP LIB ALP LIB ALP LIB (%) 10/0) (%] (%L.[%) 10/0).roAl].(%1 (%) (%) (No) (No) (No) (No) (No.) (No.) (No.) (No.) (No.) (%) (%) (%) (%) (%) (%) PRICE RAMSAY HART TAYLOR SPENCE ELIZABETH I PLAYFORD GILES ROSS SMITH NAPIER TOTAL, ALP AVE, ALL ALP LEE HANSON REYNELL KAURNA ELDER WRIGHT PEAKE FROME EYRE TORRENS NORWOOD MITCHELL MAWSON FLOREY COLTON UNLEY HARTLEY ADELAIDE COLES LIGHT NEWLAND BRIGHT FISHER MORPHETT GORDON DAVENPORT (rethrow) GOYDER WAITE (rethrow) KAVEL FINNISS HEYSEN CUSTANCE 24.6 RIDLEY MACKILLOP BRAGG I CHAFFEY (rethrow) FLINDERS (rethrow) TOTAL, LIB AVE, ALL LIB TOTAL, ALL AVE, ALL 39J _ ~_.0.2_ SOURCE: Caculated from data published in: South Aus~aia State E1ectora Office, 1996, Statistica Returns for Genera Elections 11 December 1993 and By-elections 1994 State E1ectora Office, Adelaide.

61 INFORMATION PAPER 15: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE lob: TWO PARTY PREFERRED VOTE IN EACH SEAT IF THERE HAD BEEN THE SAME NUMBER OF DECLARATION VOTES IN EACH SEAT, STATE ELECTION i i ~...,.,..,,~... RANK ALL VOTES I L"I>I> I "'-IUALI> ORDINARIES DECS EFFECT SWING ACTUAL ACTUAL ACTUAL TO ALL ORDINARY EFFECT OF LOSE VOTES VOTES DEC VOTES ALP LIB I ALP LIB 1 ALP LIB 1 ALP LIB ('!o) ('!o) _.. ('!o) ('!o) ('!o) ('!o) ('!o) ('!o) PRICE SPENCE HART TAYLOR NAPIER 21.2 (rethrow) RAMSAY ROSS SMITH ELIZABETH PLAYFORD GILES TORRENS LEE ~ IT ~.-.. PEAKE KAURNA ~.--- HANSON REYNELL WRIGHT ELDER FLOREY MITCHELL NORWOOD TOTAL, ALP AVERAGE, ALP HARTLEY STUART FROME COL TON UNLEY MAWSON ADELAIDE 5._5 _. ~ _~ ~ BRIGHT LIGHT COLES NEWLAND I- FISHER.~_ ~ ~ ---.9~ HEYSEN (rethrow) WAITE (rethrow) GORDON (rethrow) MORPHETT DAVENPORT (rethrow) HAMMOND SCHUBERT (rethrow) KAVEL (rethrow) FINN ISS (rethrow) GOYDER BRAGG MACKILLOP (rethrow) CHAFFEY (rethrow) FLINDERS (rethrow) TOTAL, LIB AVERAGE, LIB TOTAL, ALL AVERAGE,ALL SOURCE: calculated from official but as-yet-unpublished results of the State election from the Slate Electoral Office. ACTUAL DECS2PP APPlY ADD ORDINARIES GIVES US FINAL LESS I EQUALS TO 2749 DECS FINAL COUNT 2PP ORDINARIES DECSEFFECT ACTUAL HYPOTHETICAL ACTUAL ~ HYPOTHETICAL ALL VOTES ACTUAL EXPECTED DECLARATION DECLARATION OR. DINARY. V. OTES. (actu.al ordinary vot.es ORDINARY EFFECT OF VOTES VOTES -.Jin.cl. om. ticket votes] Qlus hypojl!eti~dec vote l VOTES DEC VOTES ALP LIB ALP LIB TOTAL 1 ALP LIB TOTAL ALP LIB TOTAL I ALP LIB I ALP LIB 1 ALP LIB ('!o) ('!o) (No.) (No.) (No.) (No.) (No.) (No.) (N.Q.L (tio.) _(No.) _ ('!o) ('!o) ('!oj ('!ol ('!o) ('!o) ~ t--:::_0' _~ _~ _~ ~_ t--:::-0",-.5 i I I I ~_~~ ~...1I4L_ _.2_ TI:L~_~ ~ 2~_ I I I I _.1_

62 INFORMATION PAPER 16; MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 11A: ACTUAL 2PP VOTE AND EXPECTED 2PP VOTE IF THERE HAD BEEN THE SAME NUMBER OF DECLARATION VOTES IN EACH SEAT, STATE ELECTION SWING ACTUAL EFFECT OF EXPECTED TO DEC LARA TION EFFECT OF LOSE VOTES DEC VOTES ALP LIB ALP LIB ALP LIB (%) (%) (%) (%) (%) (%) PRICE RAMSAY HART TAYLOR SPENCE ELIZABETH PLAYFORD GILES ROSS SMITH NAPIER AVERAGE, ALP LEE HANSON REYNELL KAURNA ELDER WRIGHT "----- PEAKE FROME EYRE TORRENS NORWOOD MITCHELL MAWSON FLOREY " COLTON UNLEY HARTLEY ADELAIDE COLES LIGHT NEWLAND BRIGHT FISHER MORPHETT GORDON DAVENPORT (rethrow) GOYDER WAITE (rethrow) KAVEL FINNISS HEYSEN CUSTANCE 24.6 RIDLEY MACKILLOP BRAGG CHAFFEY (rethrow) FLINDERS (rethrow) AVERAGE, LIB AVERAGE, ALL SOURCE: Table 10A.

63 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 11 B: ACTUAL 2PP VOTE AND EXPECTED 2PP VOTE IF THERE HAD BEEN THE SAME NUMBER OF DECLARATION VOTES IN EACH SEAT, STATE ELECTION SWING ACTUAL EFFECT OF EXPECTED TO DECLARATION EFFECT OF LOSE VOTES DEC VOTES ALP LIB ALP LIB ALP LIB (%) (%) (%) (%) (%) (%).~ PRICE SPENCE HART TAYLOR NAPIER 21.2 (rethrow) RAMSAY ROSS SMITH ELIZABETH PLAYFORD GILES TORRENS LEE PEAKE KAURNA HANSON REYNELL WRIGHT ELDER FLOREY MITCHELL NORWOOD AVERAGE, ALP HARTLEY STUART FROME COLTON UNLEY MAWSON ADELAIDE ~-~-~ BRIGHT LIGHT COLES NEWLAND FISHER ~--..- HEYSEN (rethrow) WAITE (rethrow) GORDON (rethrow) MORPHETI DAVENPORT (rethrow) HAMMOND SCHUBERT (rethrow) KAVEL (rethrow) FINNISS (rethrow) GOYDER BRAGG MACKILLOP (rethrow) CHAFFEY (rethrow) FLINDERS (rethrow) AVERAGE, LIB AVERAGE, ALL SOURCE: Table 10B.

64 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 12A: COMPARING METHODS OF ALLOCATING DECLARATION VOTES A. Pro-rata allocation of declaration votes 2PP TOTAL 2PP all carrying the 2PP value for all declaration votes, ALP LIB VOTES ALP LIB of ALP:L1B 52.9:47.1% (No) (No) (No.) (%) (%) FOR REFERENCE ACTUAL RESUL TS, PEAKE Total formal ordinary votes Total formal declaration votes II 2034 I Total, all formal votes, PEAKE Allocate dec votes Apply 2PP vote Add ordinary pro-rata of 52.9:47.1 votes Total votes for each booth Actual %of all Allocate to declaration (Actual results) formal formal declaration votes votes ordinary votes ALP LIB ALP LIB ALP LIB TOTAL ALP LIB votes pro rata JNol (No) (No) (No) (No) (No) (No.) (%) (%) Brooklyn Park Brooklyn Pk Sth Cowan dill a (s2) Flinders Park Hindmarsh Kidman Park Kidman Pk Sth Lockleys Lockleys North Thebarton Torrensville (s1) Total formal ordinary votes II 2034 I Total formal declaration votes I 2034 ]1 Total, all formal votes SEGMENT 1 Brooklyn Park Brooklyn Pk Sth Cowandilla (s2) Total formal votes in this segment SEGMENT 2 Flinders Park Lockleys North T orrensville (s 1 ) Hindmarsh Thebarton Total formal votes in this segment SEGMENT 3 Kidman Park Kidman Pk Sth Lockleys Total formal votes in this segment SEGMENT SEGMENT SEGMENT TOTAL FORMAL VOTES ALL SEGMENTS ACTUAL RESULTS, PE)KE, SOURCE: Calculated from data published In: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections State Electoral Office, Adelaide.

65 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 12B: COMPARING METHODS OF ALLOCATING DECLARATION VOTES B: The Commission's method: 2PP 2PP pro-rata allocation of declaration votes, all carrying TOTAL ALP LIB ALP LIB the average 2PP value for all ordinary votes. VOTES (No) (No) (%) (%) FOR REFERENCE" ACTUAL RESULTS, PEAKE, 1993 Total formal ordinary votes Total formal declaration votes Tota/, all formal votes, PEAKE SEGMENT 1 Brooklyn Park Brooklyn Pk Sth Cowandilla (s2) Total formal ordinary votes, this segment % of formal ordinary votes in Peake 18.7 So add 18.7% of formal declaration votes in Peake 381 Apply 2PP percentages of ordinary votes Total formal votes in this segment SEGMENT 2 Flinders Park Lockleys North Torrensville (s1) Hindmarsh Thebarton Total formal ordinary votes, this segment % of formal ordinary votes in Peake 47.4 So add 47.4% of formal declaration votes in Peake 964 fa,pply 2PP percentages of ordinary votes Total formal votes in this segment SEGMENT 3 Kidman Park Kidman Pk Sth Lockleys Total formal ordinary votes, this segment % of formal ordinary votes in Peake 33.9 So add 33.9% of formal declaration votes in Peake 689 Apply 2PP percentages of ordinary votes Total formal votes in this segment SEGMENT SEGMENT SEGMENT TOTAL FORMAL VOTES ALL SEGMENTS ACTUAL RESULTS, PEAKE, SOURCE: Calculated from data published In: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections 1994, State Electoral Office, Adelaide.

66 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 12C: COMPARING METHODS OF ALLOCATING DECLARATION VOTES C. Pro-rata allocation of declaration votes 2PP TOTAL 2PP according to the proportion of ALP or LIB ALP LIB VOTES ALP LIB votes lodged at the booth (No) (No) (No.) (%) (%) FOR REFERENCE ACTUAL RESUL TS, PEAKE Total formal ordinary votes Total formal declaration votes I Tota/, all formal votes, PEAKE Actual % of2pp Allocate Add ordinary votes: 2PP results ALP or LIB vote declaration votes Total votes for each booth at each booth pro rata ALP LIB ALP LIB ALP LIB ALP LIB TOTAL ALP LIB (No) (No) (%) (%) (No) (No) (No) (No) (No.) (%) 1%) Brooklyn Park Brooklyn Pk Sth Cowandilla (s2) Flinders Park Hindmarsh Kidman Park Kidman Pk Sth Lockleys Lockleys North Thebarton Torrensville (s1) Total booth votes Declaration votes I I I I PEAKE SEGMENT 1 Brooklyn Park Brooklyn Pk Sth Cowandilla (s2) Total formal votes in this segment SEGMENT 2 Flinders Park Hindmarsh Lockleys North Thebarton Torrensville (s1) Total formal votes in this segment SEGMENT 3 Kidman Park Kidman Pk Sth Lockleys Total formal votes in this segment SEGMENT SEGMENT SEGMENT TOTAL FORMAL VOTES ALL SEGMENTS ACTUAL RESULTS, PEA1KE, SOURCE: Calculated from data published In: South Australia. State Electoral Office, 1996, Statistical Returns for General Elections 11 December 1993 and By-elections 1994, State Electoral Office, Adelaide.

67 INFORMATION PAPER 16 : MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 120: COMPARING METHODS OF ALLOCATING DECLARATION VOTES: SUMMARY TABLE ACTUAL RESULTS, Declaration votes not allocated to booths 2PP TOTAL 2PP TOTAL ALP LIB VOTES ALP LIB VOTES (No) (No) (No.) (%) (%) (%) Segment Segment Segment Total formal votes in all segments Total formal declaration votes TOTAL, ALL FORMAL VOTES, PEAKE TABLE 12A: pro-rata allocation of declaration votes, all at ALP:LlB 52.9%: 47.1 % 2PP TOTAL 2PP ALP LIB VOTES ALP LIB (No) (No) (No.) (%) (%) Segment Segment Segment Total formal votes in all segments Total formal declaration votes TOTAL, ALL FORMAL VOTES, PEAKE TABLE 12B: The Commission's method: pro-rata allocation of declaration votes, at each booth's 2PP% 2PP TOTAL 2PP ALP LIB VOTES ALP LIB (No) (No) (No.) (%) (%) Segment Segment Segment Total formal votes in all segments Total formal declaration votes TOTAL, ALL FORMAL VOTES, PEAKE TABLE 12C: Pro-rata allocation of declaration votes according to the proportion of ALP or LIB votes taken at the booth 2PP TOTAL 2PP ALP LIB VOTES ALP LIB (No) (No) (No.) (%) (%) Segment Segment Segment Total formal votes in all segments Total formal declaration votes 0.0 TOTAL, ALL FORMAL VOTES, PEAKE SOURCE: Tables 12A to 12C.

68 State election 1977 (actual results) ALP LIB J%) (%) INFORMATION PAPER 16: MAKING SURE THAT ELECTORAl BOUNDARIES ARE FAIR IN SOUTH AUSTRAliA TABLE 13: TWO PARTY PREFERRED RESULTS AT SOUTH AUSTRALIAN STATE ELECTIONS, 1977 TO EDBC boundaries 1983 EDBC boundaries 1991 EDBC boundaries 1994 EDBC boundaries State election 1979 (actual results) ALP LIB (%) J%) State election 1982 (actual results) ALP LIB (%) (%) State election 1982 (Jaensch estimates) ALP LIB (%) J%) State election State election 1985 (actual results) 1989 (actual results) ALP LIB ALP LIB (%) (%) (%) (%) State election 1989 State election 1993 (Newton estimates) (actual results) ALP LIB ALP LIB (%) (%) (%) (%) State election 1989 (Newton estimates) ALP LIB (%) (%) State election State election 19 M (EDBC estimates) 1997 (actual results) ALP LIB ALP LIB (%) (%) (%) (%) Adelaide Albert Park Alexandra Ascot Park Baudin Bragg Briggs Bright Brighton Chaffey Coles Colton Custance Davenport Elder Elizabeth Eyre Finniss Fisher Flinders Florey Frome Giles Gilles Glenelg Gordon Goyder Hammond Hanson Hart Hartley Hayward Hen ley Beach Heysen Kauma Kavel Lee Light MacKiliop Mallee Mawson Mitcham Mitchell MorpheU MtGambier not available 21.8 (E) 78.2 (E) 22.0(E) 78.0 (E) (E) 74.6 (E) (E) (E) (E) 78.8 (E) (E) 45.7 (E) CONTINUES (E) (E) (E) (E) (R) 72.2 (R) (R) (R) (R) 63.9 (R) (R) 23.1 (R) CONTINUES (R) (R) (R) (R) CONTINUES (R) (R) (R) \ ! 55.8) : 44.9: : 52.8) 76.9: 48.1: 554: : 67.3: 705) 451: 28.0: : 464) 668: 41.9: 602: ) CONTINUES 64' :29.0 (R) 71.0 (R) 68.8: :36.2 (R) 63.8 (R) 53.5) : 34.1 (R) 65.9 (R) 70.?! )21.5 (R) 78.5 (R) 60.9: : :377 (R) 62.3 (R) 72.5) i ) : i 74.5:379 (R) 62.1 (R) 524: )35.0 (R) 65.0 (R) 51.0: : (R) 70.9 (R) )

69 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA Murray Murray Mallee Napier Newland Norwood Peake Playford Price Ramsay Reynell Ridley Rocky River Ross Smith Salisbury Schubert Semaphore Spence Stuart Taylor Todd Torrens Unley Victoria Waite Walsh Whyalla Wright 1976 EDBC boundaries State election 1977 State election 1979 State election 1982 (actual results) (actual results) (actual results) ALP LIB ALP LIB ALP LIB (%) (%) (%) (%) (%) (%) (E) (E) 70.0 (E) 30.0 (E) (E) 32.0 (E) 1983 EDBC boundaries State election 1982 State election 1985 State election 1989 (Jaensch estimates) (actual results) (actual results) ALP LIB ALP LIB ALP LIB (%) - (%) (%) (%) (%) (%) (R) 72.2 (R) (R) (R) (R) EDBC boundaries State election 1989 (Newton estimates) ALP LIB (%) j%) ~B ~4 ~4 4~1 ~2 34~ 3&1 ~.1 m State election 1993 (actual results) ALP LIB (%) (%) (R) ~] ~4 ~4 ~6 ~J ~~ 4~1 ~.2 7~ (R) EDBC boundaries State election 1989 State ele";m State election 1997 (Newton estimates) (EDBC estimates) (actual results) ALP LIB ALP LIB ALP LIB (%) (%) (%) (%) (%) (%) ,J l 45.9l 40.6l 341l \ 54.9l \ , 43.6l 71.1 (R) 28.9 (R) 66.6l l l l (R) 64.9 (R) 42.3: l i 74.3\37.8 (R) 62.2 (R) State-wide 534(El 46.6 (E) (E) 55.1 (E)! 50.9 (El 49.1 (El (Rl 468 (Rl : 47.9 (Rl 52.1 (Rl ! 39.1 (R) 60.9 (R) ! :485 (R) 51.5 (R) NOTES: Results marked (R) are from rethrows conducted by the State Electoral Office. Results marked (E) are Jaensch's estimates of the ALP: LIB two party preferred vote. Note that Jaensch did not calculate an estimated ALP: LIB two party preferred vote for Flinders at the 1977 State election. SOURCES: Jaensch, D, 1977, The 1977 Elections in South Australia: a statistical analysis, Flinders University Politics Department Occasional Monograph No.6, Adelaide. Jaensch, D, 1979, The 1979 Elections in South Australia: a statistical analysis, Flinders University Politics Department Occasional Monograph No.7, Adelaide. Jaensch, D, 1983, The 1982 Elections in South Australia: a statistical analysis, Flinders University Politics Department Occasional Monograph No.8, Adelaide. Newton, J, 1992, The 1991 Electoral Redistribution in South Australia: 1989 voting statistics transferred to the new House of Assembly boundaries, Research Service Information Paper No. 11, Parliamentary Library of South Australia, Adelaide. SA. State Electoral Department, 1978, General Elections 1977: Statistical Returns, SA Parliamentary Paper 144 of SA. State Electoral Department, 1980, Periodical and General Elections 1979: Statistical Returns, SA Parliamentary Paper 132 of SA. State Electoral Department, 1982, Periodical and General Elections 1982: Statistical Returns, SA Parliamentary Paper 145 of SA. State Electoral Department, 1986, General Elections 1985: Statistical Returns, SA Parliamentary Paper 145 of SA. State Electoral Department, 1991, Periodical and General Elections 1989: Statistical Returns, SA Parliamentary Paper 161 of SA. State Electoral Office, 1996, Statistical Returns for General Elections 1993 and By-elections 1994, State Electoral Office, Adelaide. SA. Electoral Districts Boundaries Commission, 1994, 1994 Report, The CommisSion, Adelaide electoral results are official but as-yet-unpublished data from the State Electoral Office.

70 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 14: SWINGS AT STATE ELECTIONS, 1977 to 1997 (swing to ALP 1976 EDBC boundaries 1983 EDBC boundaries 1991 EDBC boundaries 1994 EDBC boundaries SWING to ALP SWING to ALP SWING to ALP SWING to ALP 1977 to to to to to to to 1997 (actual results) (estimated) (actual) (estimated) (estimated) (estimated) (percentage points) (percentage points) (percentage points) (percentage points) Adelaide Albert Park Alexandra Ascot Park Baudin Bragg Briggs Bright Brighton Chaffey Coles Colton Custance Davenport Elder Elizabeth Eyre Finniss Fisher Flinders not available Florey Frome Giles Gilles Glenelg Gordon Goyder Hammond Hanson Hart Hartley Hayward Henley Beach Heysen Kaurna Kavel Lee Light MacKiliop Mallee Mawson Mitcham Mitchell Morphett MtGambier Murray Murray Mallee Napier Newland Norwood Peake Playford Price Ramsay Reynell Ridley -4.4 Rocky River Ross Smith Salisbury Schubert Semaphore Spence Stuart Taylor Todd Torrens Unley Victoria Waite Walsh Whyalla Wright State-wide SOURCE: calculated from Table 13.

71 INFORMATION PAPER 16 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA SAFE PENDULUM AFTER THE STATE ELECTION OF l(.;..;.AC.;;;TUAL RESU.=..L T;,.;;S;L.)... i SWING I ALP TO LOSE LIB ~ 131+ I I ACTUAL ~29 RESULTS EDBC BOUNDARIES L =- :: ~~ce 7:GL_.125 [ 24 Stuart 73.9 'Mlyalla Elizabeth 72.4 Ross Smith & Semaphore Napier 709 _ 20 Salisbury 695 ~~.@! 19 I Peake Florey & Ptayford Baudin Adelaide _6QL 116 I~ 1 15 t----- Albert Park I-- Gilles Ascot Park 'CM"'itcsh"'ell'-i Hartley 60.4 Norwood 60.2 Un~y_60.1 Newland 59~B [,0_ t -----~~each ~9.3 9 FAIRLY SAFE MEDIAN SEA T Brighton~ ' MaWSOn&T~~~-t ~ --I ACTUAL 28 RESULTS 1976 EDBC BOUNDARIES I I 74.6 Goyder (Jaensch est. of 2PP) 74.3 Flinders (NATvs ALP) Davenport 73.9 Mallee I 21.-_ 2O~nd;a t--=66.8~k~av-el-- -I _--I '-- 17' t 65.4Vic~ 15~~ '1--_.' _ Mitcham (Jaensch est. of 2PI} 60.7 Light 60.4 Fisher _lol Q&_f.1UI@)' 59.6 Chaffey 59.5 Glenelg 9 1 ~---- :L~2 R~iver --=-== 55.9 Eyre ~~n"," _~ -I STATEAVERAGE MARGINALJ 12 ~ 21~ Torr_""'-- 1 1\ 514 Cdes & Mt Gambier -- V I Morphett SAFE FAIRLY SAFE PENDULUM AFTER THE STATE ELECTION OF 1979,... :(-.;AC.;;;IUAL RESU,~p:.:;S:L.)... ALP TO LOSE I I---'-ACTUAL RESULTS EDBC BOUNDARIES I- _ , ~ce 70LI20 I- _ _ RossSmith...6!.4.jI8 17 Stuart 66.9 'Mlyalla Sema.. hore Jaensch es. t. 0.!2PP) 63.5j t-- Salisbury I 11 Elizabeth & Price 60.5 L.a. Napier I Adelaide5~1 ~ke 57.8 _ 7 Baudlri-55. Gilles 55.4 t--- Hartlel'!. 'I~_ ~ _ 5 Mitchell Albert Park 53.9 Florey ~5R_12 MARGINAL Ascot Park 51.7 LIB 82.5 Mitcham (Jaensch est. of 2PP Davenport Mall.. (Jaensch est. of 2PP) I 28 I 78.2 Flinders (Jaensch est. of 2PP) Goyder 261- _~_I 25~ 75.3 Alexandra 74.3 Kavel I J...~ 201~isher, Vict~ Glenelg & Light Murray 15 t ---;;64. <"5 LHi::an:;son:;;;; Chaffey Rocky River CeJes_ , Torrens 59:9 Eyre :Fs~I=:~d- -~ 55.6 Mt Gambier 55.3 Morphell STATEAVERAGE 54.7 Brighton 4J ~,~.Jodd MEDIAN~ :r~~aws~-=~_--=i 1-- -\ ~ )_.51R_Henley Beach NOTE: The actual swing to lose in any given seat will be the difference between 50.0% and the 2PP vote for thai seat, plus either 1 vole or 0,1%. SOURCE: Table 13. I 50.1 Norwood SAFE FAIRLY SAFE MARGtNAL PENDULUM AFTER THE STATE ELECTION OF 1982 r- (~A~CIUAL RESU,;;.LT;..;;S )... ALP TO LOSE LIB I I I---- ACTUAL _ RESULTS EDBe BOUNDARIES Spence ~~"'"'126 I---- Ross Smith 75,6 I C ~S"'"a"'lisb"'u:::ry;-c7n2-.; Elizabeth Semaphore t (Jaensch est. of 2PP) & Stuart Peake 69.5 _ Na ier 'Mlyalla (Jaensch est. of 2PP) I- 117 Prioe Baudin 65.5 _ Albert Park r Playford 64.7 M ~--~A~dwel~ai de~&~fnlo=re=y~6~3~.9~ _~. I Mitchell Hart~60.0 Grli8s Ascot Park 59.3 Norwood ~_---I Mawson & Unley 56.6 _--15 MEDIAN SE:;nleJe~~:~~ ~t--14 _ I-- ---_ I Mall.. (Jaensch est. of 2PP) I Flinders (Jaensch est. of 2PP) ~~~~~_ Davenport Goyder 24t Mitcham (Jaensch est. of 2PP) I Alexandra '1 66.7Kavel 66.6 Victoria Bragg = Chaffey Li ht 141~--_ Glenelg Eyre Fisher 60.6 Murray 60.3 Rocky River...1Il Hanson I Torrens Mt Gambier 51.5 Morphell 51.4 Todd Coles STATEAVERAGE 509 V Bllghton ~ ----.l SAFE FAIRLY SAFE MARGINAL.., l e c!:< to' "

72 -----~ _. SAFE vvaf~~ SAFE -----~ ~----- ~ ~ ~-- -~---- ~ ~~ ~- _~~ ~- ~-~~- --~- INFORMA non PAPER 16 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA PENDULUM AFTER THE STATE ELECTION OF 1982 PENDULUM AFTER THE STATE ELECTION OF 1985 _ ,."""", ' ""...,~ SWING SWING ALP TO LOSE LIB ALP TO LOSE LIB ~ _29 29 JAENSCH JAENSCH ACTUAL ACTUAL -~~- ESTIMATES ESTIMATES RESULTS RESULTS ON THE BASIS OF ON THE BASIS OF 1963 EDBC BOUNDARIES 1963 EDBC BOUNDARIES THE 1963 EDBC THE 1963 EDBC ~~ ~ _~ BOUNDARIES BOUNDARIES ~ - -~. ~a & Semaphore & Price ~ Flinders & Alexandra ~ Semaphore (IND LAB, rethrow) & Price 74.3 Ramsav & Sluart & Elizabelh Ramsav Napier Bragg ~----~-- Elizabelh (IND LAB, relhrow) 72.2 Spence Davenport Stuart Flinders (NAT, relhrow) Napier _~~ ~ Custance ~ ~~_~ Pla)'!ord ~- Ross Smith Murray Mallse Ross Smith _.~_6:8 Victoria & Goyder & MurraLM_allse Sp<lf1ce Vicloria Briggs 67.4 ~ 67.5 Bragg ---._~ke ~ Whyalla ~- ~-- SAFE Briggs 66_._ Chaffe! ~_~ Maw~ Cuslance <!nd Chaffey Peake Eyre Eyre & light Alexandra -- PENDULUM AFTER THE STATE ELECTION OF ,..., , SWING ALP TO LOSE LIB f-- 29 ACTUAL ACTUAL RESULTS RESULTS 1963 EDBC BOUNDARIES 1963 EDBC BOUNDARIES f Flinders (NAT, rethrow) f---~-----~ ~~ ~ Victoria 72.6 Murray-Mall.. isemaphore (IND LAB, relhrowi MI Gambier Chaffey 70.1 Custance Bragg -- f--- ~ Eyre f-~~ ~~ Napier Ramsay Goyder SAFE Price Alexandra 65.9 Davenport 65.5 Kav Heysen -- SAFE Pla),!ord & Albert Park Gilles Davenport (IND LIB, rethrow) Baudin 63.7 Baudin Mitchell ~rark62.9 Gilles _~~~~ ~~_~ ~_ Hartley Gcyder ~ ~ Kav" Mawson ~ -----~- 11 Hartley & Mitchell ( 60 Mitcham & Kaval MI. Gambier.. _..._._._--_._._ Coles 59.3 Light 56.6 Heyson FAIRLY FAIRLY 56.4 Coles SAFE Waislc_:iL_ ~-~ Florey Milcham --~ _.._._- 57 Morphell ~_!:Ienle! Beach Norwood & Florey 56 Norwood 56.3 I odd "...._---_ MEDIAN SEA T Unlet 55.0 ~ Morphell Henley Beach Adelaide & Hanson ~--~.~ ~ _~ ~~ ~_~ 3 53 MI. Gambier STATEAVERAGE Hayward 52.6 MEDIANSEAT MARGINAL Unle! & Todd Fisher MARGINAL --.~--~----~ Bright 51.6 Newland il Newland_~ ~!'! Fisher STA-tEAVERAGC50~ 1 V V 0 Adelaide Hayward ~noon--- I NOTE: The actual swing to lose in any given seat Will be the difference between 50.0% and Ihe 2PP vote for thai seat, plus either 1 vote or 0.1%. SOURCE: TABLE 13 Ross Smith & Spence Elizabeth (IND LAB, rethrow) f-~ ~ -- Sluart (rethrowl Coles 62.9 Lighl r---~~ Briggs Mitcham Whyalla Peake Morphell Pla)'!ord r ~- FAIRLY SAFE --- Albert Park Mawson 57.3 Gilles Walsh Hanson Baudin Hartley _. -- Mitch Henlev Beach Adelaide -~ Fisher Todd Norwood ~ MARGINAL Unley STATEAVERAGE --- MEDIAN SEAT Florey _._ Bright V ~ 50.9 Hayward ~Newland _ FAIRLY SAFE MARGINAL ":I H Gl C :;d trj W

73 INFORMATION PAPER 16 MAKING SURE THAT ELECTORAL BOUNDARJES ARE FAIR IN SOUTH AUSTRALIA PENDULUM AFTER THE STATE ELECTION OF '('-N_EWT-" O~~~ZIMr;iA-TE...;S..:..) -, ALP I TO LOSE LIB PENDULUM AFTER THE STATE ELECTION OF 1993 _ !:,;(A:.:.,;CTUAL RESU-=-LT...;S.:...)..., i SWING 1- ALP I TO LOSE I LIB I 301- ~ 1 I Flinders (rethrow) SAFE FAIRLY SAFE NEWTON ESTIMATES ON THE BASIS 28 OF THE 1991 EDBC BOUNDARIES t :: l-----i Hart ~ ~ I j Price ~{-l16 ~ Elizabeth 65~_-J15 t----;;--~c--,c =_-,-=~14 Ramsay & Spence & Taylor 63.9 R_ossSmith t-- -- ~_~ll1 110 Playford ,.,..~: r ,:ee I~NEWTON ESTIMATES ON THE BASIS OFTHE EDBC BOU~~._ Flinders 251- I MacKiliop " n.,_ ~70"'.7;-C"'h-aff;;-e-Cy ;;-& "'Ri"'dle:C y Custance Goyder Brag~ f 66.8 Kavel 16'1- _ Finniss 64_8 Heysen Davenport 62.8 Warte _._ ( 111.1Q 60.7 Light Morphett _ SAFE FAIRLY SAFE ~_ ~ ACTUAL 27 RESULTS 1991 EDBC BOUNDARIES I _ 1 23 I _~~~_I20 1--_ I f _---t 12 f Price Ramsay 59.9 I-~-- Hart 58.5 I 9 r- Taylor 58ti0 8 spence 57. Elizabeth 57_ ChaNey (rethrow) 78.7 Bragg 28 -~~7~7.~7~M;-ac~K~iII-op------i 27 ACTUAL RESULTS EDBC BOUNDARIES Ridley 74.5 Custance & Heysen 74.4 Finniss Kavel 73.9 Waite (rethrow) Goyder Davenport (rethrow) & Gordon 71.9 Morphett Fisher 2O~ ~ Bright Newland Ught 65.9 Coles Adelaide _2 Hartley 1~ "6~1<.5CU~nle~y I STATEAVEIl4GE Colton Florev MEDIAN SEA T 59.6 Mawson Mitchell 8\ Norwood 56.5 Eyre & Torrens SAFE FAIRLY SAFE MARGINAL Giles tkl.8 Wright 55.6 f-- Hanson 55'5_~5 Elder ~ens 54.S 4 Kaurna 53.2 Mitchell Peake 52.9 Mawson 52.4 Florey 52_0 Norwood _ Newland 50.6 Unley 50.4 MEiJ. v o ; AT Frome _5 Eyre 3~S-he-r Bright & Coles 52.4 Hartiey 21...!~1_ TATE A VEIl4 >E Adelai~ Colton MARGINAL tiqle: The actual swing to lose in any given seat will be the difference between 50,0% and the 2PP vote for that seat, plus either 1 vole or 0.1% _ Wnght Playford 52.7 Giles 52.4 f---- B9ssSmith , V o 55.7 Frome 55.6 Peake 51- ~ ---I Elder 52.8 Kaurna 51.2 Hanson & Reynell 51.1~ :;OUKCc: TABLE 13 MARGINAL "' >-- G c ~ t:r "'

74 PENDULUM AFTER THE STATE ELECTION OF J,;,;(N~EW:..;T.ON ESTIMr-:AT.:...;E;,;;;SJ...} -, r- SWING ALP TO LOSE LIB 130 I I PENDULUM AFTER THE STATE ELECTION OF ,l,;(E::.::D:;-BC ESTIMA.:..:TE;,;;;SJ...} -, SWING.- ALP TO LOSE LIB 30 I Flinders I 79.4 MacKiliop INFORMATION PAPER 16 MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA PENDULUM AFTER THE STATE ELECTION OF l(.:...:.ac:::.;tual RESU';=.LT.:.::S:I.,}... SWING ALP TO LOSE LIB I SAFE NEWTON ESTIMATES 29 ON THE BASIS OF THE EDBC BOUNDARIES 127 J f Hart 72.0 _.122 t=------~-_==l: f ~P::-:ri-ce--;;8~5.9n I Napier 85.6 Tayklr 65.1 '115 Ramsay 64.7 Spence Ross Smith Elizabeth NEWTON ESTIMATES ON THE BASIS OF THE 1994 EDBC BOUNDARIES :;:;-;c-=c-:::: Flinders 261-_. I 25 ~1~~~~ MacKiliop 241 " I 221t---:;-;-:o-:c Gordon 21~_::-::::--:: 70.7 Chaney Hammond j =-=~ 68.7 Bragg Goyder Schubert 66.8 l\avel Finniss Heysen Wail_e Davenport SAFE EDBC ESTIMATES ON THE BASIS OF THE 1994 EDBC BOUNDARIES _---125,, l _122 I t '.1:: Bragg 291-_J9 c O Chaney 28 EDBC ESTIMATES ON THE BASIS 27 OF THE 1994 EDBC BOUNDARIES Finniss Hammond 74.5 Heysen 74.3 Waite Kavel Goyder 72.4 Morphett & Schubert Daven 71.5 Gordon Fish_er -_-- ] 19'- _".. ---I Bright 171 ~ 66.8 Coles ' Newland 151" 65.2 Light Adelaide _ Har1Iey --- SAFE I ~128 28~ Flinders (rethrow) ---1 ACTUAL 27 RESULTS 1994 EDBC BOUNDARIES 26 J Price Spence 73.3 r 123 Hart 72.7 ~Ior 72.2 Napier (rethrow) J----- t Ra~Say 682 r: _ J Ro-ss-S~m-cithc--;;-;64, Elizabeth Playford , ACTUAL RESULTS 1994 EDBC BOUNDARIES I I I Chaney (NAT, rethrow) 70.9 MacKillop (IND LIB, rethrow) ~ r.68~.8~~~ag-g--- --I J Goyder t 63.8 Davenport (rethrow) Morphett 62.3 Gordon (IND, rethrow) 62.2 Waite (rethrow) Heysen (rethrow) SAFE FAIRLY SAFE Playford~~9 Lee c;Gccile"'-'S Reynell 57.6 Wright Elder Hanson 54.9 Torrens 54.5 Peake 54.1 MARGINALI-~a & Norwood 53.6 I- Mitchell I LiQht 59.8 Morphett :t~-~-----i 55.8 Coles 51_.55.4 Frome" 54.9 Stuart I Fisher STA TEA VERAGE I Florey 51.9 I Newland 51.0 I EDIANSEATHar1ley&Unley FAIRLY SAFE MARGINAL Price Ramsay Taylor 59.4 Hart 58.7 Spence ""_ 7 Napier _ 9Jles Elizabeth 52.7 f-~~smith & Playford \1 v Unley I 60.9 Florey ALSO STA TEA VERAGE Colton MEDIAN SEA T 59.4 Mitchell Mawson 58.9 Stuart 81~~ Norwood 56.5 Torrens 54.9 Peake 53.5 Elder 3 _.53.2 Wright 52.5 Reynell Kauma Hanso~ ' 51.0_Le_e FAIRLY SAFE MARGINAL ~... "iles Torrens r--- -~L-ee~ f 1--- Peake 57.0 J7 Kauma 55.9 Hanson I ~nell I.'!'!r:i9ht Elder Fisher Newland 57.8 Coles Light 56.1 Brigilt Adelaide 54.7 Mawson 54.5 Unley 54.0 Co~on 52.9 Frome MEDIAN SEA T 51.6 Stuart Florey 5..1~ 51.5 STATE AVERAGE Mitchell 50.9 Norwood Hartley FAIRLY SAFE MARGINAL "':I H gj Rl U1 NOTE: The actual s'ning to lose in any given seat will be the difference between 50.0% and the 2PP vote for that sea~ plus either 1 vote or 0.1%. SOURCE: TABLE 13

75 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 15A: ACTUAL AND HYPOTHETICAL PROPORTIONATE SWINGS, ACTUAL 1979 RESULTS HYPOTHETICAL 1982 RESUL TS ACTUAL 1982 RESULTS swing 2PP% State election proportionate 2PP% State election swing 2PP% State election to ALP swing to ALP to ALP ALP LIB ALP LIB ALP LIB (%) 2PP% 2PP% (%) 2PP% 2PP% (%) 2PP% 2PP% Spence Stuart Whyalla (a) Elizabeth Semaphore (b) Ross Smith Napier Salisbury Price Peake Playford Florey Baudin Adelaide Albert Park Gilles Ascot Park Mitchell Hartley Norwood Unley Newland Henley Beach Brighton Todd Mawson Morphett Coles MtGambier Torrens Hanson Eyre Rocky River Glenelg Chaffey Murray Fisher Light Mitcham (c) Bragg Victoria Kavel Alexandra Mallee(d) Davenport Flinders (e) Goyder (D State-wide SOURCE: Table 13A.

76 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 15B: ACTUAL AND HYPOTHETICAL PROPORTIONATE SWINGS, ACTUAL 1985 HYPOTHETICAL 1989 ACTUAL 1989 Est. swing 2PP% State election proportionate 2PP% State election swing 2PP% State election to ALP swing to ALP to ALP ALP LIB ALP LIB ALP LIB (%) 2PP% 2PP% (%) 2PP% 2PP% (%) 2PP% 2PP% Price Semaphore (d) Whyalla Elizabeth (b) Ramsay Stuart (e) Spence Napier Ross Smith Peake Briggs Albert Park Playford Baudin Gilles Mawson Hartley Mitchell Walsh Florey Norwood Henley Beach Todd Unley Hayward Bright Newland Fisher Mt. Gambier Adelaide Hanson Morphett Coles Kavel Mitcham Heysen Eyre Light Chaffey Goyder Murray Mallee Victoria Custance Davenport (a) Bragg Alexandra Flinders (ct Statewide SOURCE: Table 13A.

77 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 15C: ACTUAL AND HYPOTHETICAL PROPORTIONATE SWINGS, ACTUAL 1993 (EDBC est.) HYPOTHETICAL 1997 ACTUAL 1997 Est. swing 2PP% State election proportionate 2PP% State election swing 2PP% State election to ALP swing to ALP to ALP ALP LIB ALP LIB ALP LIB (%) 2PP% 2PP% (%) 2PP% 2PP% (%) 2PP% 2PP% Price Spence Hart Taylor Napier Ramsay Ross Smith Elizabeth Playford Giles Torrens Lee Peake Kaurna Hanson Reynell Wright Elder Florey Mitchell NOlWood Hartley Stuart Frome Colton Unley Mawson Adelaide Bright Light Coles Newland Fisher Heysen Waite Gordon Morphett Davenport Hammond Schubert Kavel Finniss Goyder Bragg MacKiliop Chaffey Flinders State-wide SOURCE. calculated from Table 13.

78 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 16: TWO PARTY PREFERRED RESULTS AND SWINGS IN METROPOLITAN AND COUNTRY, 1977 TO 1997 (ACTUAL RESULTS) ALP LIB TOTAL ALP LIB TOTAL ALP LIB TOTAL ALP LIB TOTAL ALP LIB TOTAL ALP (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) ALL METROPOLITAN 2PP vote (No.) PP vote (%) LIB (No.) TOTAL (No.) ALP (No.) 1997 LIB (No.) TOTAL (No.) ALL COUNTRY 2PP vote (No.) PP vote (%) ALL 2PP vote (No.) PP vote (%) SUMMARY SWING IN ALL METRO SWING IN ALL COUNTRY SWING IN ALL I NOTES: The metropolitan seats were Adelaide, Albert Park, Ascot Park, Baudin, Bragg, Briggs, Bright, Brighton, Coles, Colton, Davenport, Elder, Elizabeth, Fisher, Florey, Gilles, Glenelg, Hanson, Hart, Hartley, Hayward, Henley Beach, Heysen, Kauma, Lee, Mawson, Mitcham, MitcheJl, Morphett, Napier, Newland, Norwood, Peake, Playford, Price, Ramsay, ReyneJl, Ross Smith, Salisbury, Semaphore, Spence, Taylor, Todd, Torrens, Unley, Waite, Walsh and Wright. The country seats were Alexandra, Chaffey, Custance, Eyre, Finniss, Flinders, Frome, Giles, Gordon, Goyder, Hammond, Kavel, Light, Mackillop, MaJlee, Mt Gambier, Murray, Murray MaJlee, Ridley, Rocky River, Schubert, Stuart, Victoria and WhyaJla. NOTES: In constructing the table I had to use estimated two party preferred figures for some seats at some elections PP figures for Goyder and Mitcham are Jaensch estimates, and PP figures for Flinders are ALP:NAT because Jaensch made no estimate of the ALP:LlB spl~ PP figures for M~cham, Semaphore, Flinders and MaJlee are Jaensch estimates. For 1985 onwards, the State Electoral Office has provided figures for actual rethrows of the results in Davenport, Elizabeth, Semaphore and Flinders (in 1985), Elizabeth, Semaphore, Flinders and Stuart (in 1989), and Davenport, Waite, Chaffey and Flinders (in 1993) and Davenport, Heysen, Napier, Waite, Chaffey, Finniss, Flinders, Gordon, Kavel, MacKiJlop and Schubert (in 1997).

79 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 17: SIZE OF THE SWING IN VARIOUS SECTORS OF THE PENDULUM. SWING 1977 TO 1979 SWING 1979 TO 1982 SWING 1982 TO 1985 SWING 1985 TO 1989 SWING 1989 to 1993 (1991 (1994 boundaries) boundaries) SWING 1993 TO 1997 SWING IN ALL 8.5 to LIB 6.0 to ALP 2.3 to ALP 5.3 to LIB 8.9 to LIB 9.4 to ALP ALP safe ALP 60% + fairly safe ALP 56% < 60% marginal ALP 50% < 56% ,,,...,...,,,,,,,..,,,,,,,,,,...,, Q3 6.8 ao a ""... "... "..,"""'''"''""... "... "... ""...,,...,," ""... "... """... ".. "... ",...,,... ".. "... "" marginal LIB 50% < 56% fairly safe LIB 56% < 60% safe LIB 60% + LIB ,... "'...,,,... "'...,, ""...,'""".,..,,,...,," 9.1 SOURCE: calculated from Figures 2,3,4, and 5, and Tables 15A, 158 and 15C.

80 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA I AtjLt: 1 ts: ~WINt;;~ A I ~ I A It: t:lt:{; IIUN~, l~rr to l~~r (SWIn~ to ALtJ) 1976 EDBC BOUNDARIES 1983 EDBC BOUNDARIES 1991 EDBC BOUNDARIES 1994 EDBC BOUNDRIES SWING to ALP SWING to ALP SWING to ALP SWING to ALP 1977 to to to to to to to 1997 (actual results) (estimated) (actual) (estimated) (estimated) (estimated) (percentage points) (percentage points) (percentage points) (percentage points) STATEWIDE SWING to ALP SWING IN COUNTRY Alexandra Chaffey Custance Eyre Finniss Flinders not available Frome Giles Gordon Goyder , Hammond Kavel Light MacKiliop Mallee Mt Gambier Murray Murray Mallee Ridley -4.4 Rocky River Schubert Stuart Victoria Whyalla SOURCE: Table 14.

81 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 19: MEASURES OF THE SWING PATTERN. STATE ELECTIONS 1977 TO 1993 Swing Deviation Absolute Swing Deviation Absolute Swing Deviation Absolute Swing Deviation Absolute Swing Deviation Absolute Swing Deviation Absolute 1977 from deviation 1979 from deviation 1982 from deviation 1985 from deviation 1989 from deviation 1993 from deviation to the from to the from to the from to the from to the from to the from 1979 mean mean 1982 mean mean 1985 mean mean 1989 mean mean 1993 mean mean 1997 mean mean (actual) swing swing (actual) swing swina lest.) swine swing (actual) swing swing (est.) swing swing (est.) swing swing Adelaide Albert Park Alexandra Ascot Park Baudin Bragg Briggs Bright Brighton Chaffey Coles Colton Custance Davenport Elder Elizabeth Eyre Finniss Fisher Flinders ~ Florey Frome Giles Gilles Glenelg Gordon Goyder Hammond Hanson Hart Hartley Hayward Henley Beach Heysen Kaurna Kavel Lee Light Mackillop Mallee Mawson Mitcham CONTINUES

82 INFORMATION PAPER 16 : MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA Swing Deviation Absolute Swing Deviation Absolute Swing Deviation Absolute Swing Deviation Absolute Swing Deviation Absolute Swing Deviation Absolute 1977 from deviation 1979 from deviation 1982 from deviation 1985 from deviation 1989 from deviation 1993 from deviation to the from to the from to the from to the from to the from to the from 1979 mean mean 1982 mean mean 1985 mean mean 1989 mean mean 1993 mean mean 1997 mean mean I (actual) swing swing (actual) swing swing (est.) swing swing (actual) swing swing (est.) swing swing (est.) swing swing Mitchell Morphett MtGambier Murray Murray Mallee Napier Newland Norwood Peake Playford Price Ramsay Reynell Ridley Rocky River Ross Smith Salisbury Schubert Semaphore Spence Stuart Taylor Todd Torrens Unley Victoria Waite :: Walsh Whyalla Wright Mean of swings Mean absolute deviation No. of observations below mean absolute deviation above 3 X mean absolute deviation above 4 X mean absolute deviation No. of +lie deviations from sample mean No of -ve deviations from sample mean I SOURCE: calculated from Table 14.

83 INFORMATION PAPER 16 : MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 20: FISCHER'S METHOD: PROBABILITY THAT A GIVEN SEAT WOULD CHANGE HANDS WITH A GIVEN SWING. HYPOTHETICAL ELECTION WITH AN 8 9% SWING TO THE ALP ALP SWING TO LOSE Swing No. of standard Area under Probability ACTUAL IN 1989 (Newton est.) required deviations probability of seat RESULTS above 8.9% from mean curve changing hands 1993 Hart n.a. 0.0 stayed ALP Price stayed ALP Elizabeth stayed ALP Ramsay stayed ALP Spence stayed ALP Taylor stayed ALP Ross Smith stayed ALP Napier stayed ALP Playford stayed ALP Reynell swu~ Lee swung Giles stayed ALP Wright swung Hanson swung Elder swung Torrens swung Kaurna swung Mitchell swung Peake swung Mawson swung Florey swung Norwood swung Newland n.a swung Unley n.a swung HYPOTHETICAL ELECTION WITH A 9 4% SWING TO THE ALP LIB SWING TO LOSE Swing No. of standard Area under Probability ACTUAL IN 1993 (Newton est.) required deviations probability of seat RESULTS above 9.4% from mean curve changing hands 1997 Lee n.a swung Hanson swung Kauma swung Reynell swung Wright swung Elder swung Peake swung Torrens swung Norwood swung Frome stayed LIB Stuart stayed LIB Mawson stayed LIB Mitchell swung Colton stayed LIB Florey swung Unley stayed LIB Hartley stayed LIB Adelaide stayed LIB Light stayed LIB Newland stayed LIB Bright n.a. 0.0 stayed LIB Coles n.a. 0.0 stayed LIB Fisher n.a. 0.0 stayed LIB Gordon n.a. 0.0 stayed LIB Davenport n.a. 0.0 stayed LIB Morphett n.a. 0.0 stayed LIB Schubert n.a. 0.0 stayed LIB Goyder n.a. 0.0 stayed LIB Kavel n.a. 0.0 stayed LIB Waite n.a. 0.0 stayed LIB Heysen n.a. 0.0 stayed LIB Hammond n.a. 0.0 stayed LIB Finniss n.a. 0.0 stayed LIB Chaffey n.a. 0.0 stayed LIB Bragg n.a. 0.0 stayed LIB MacKiliop n.a. 0.0 st<l}'ed LIB SOURCES: Table 13; Fischer 1991 at p.302; and Karmel and Polasek 1970 at Appendix 1.

84 INFORMATION PAPER 16 : MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 21: THE CUBE RULE APPLIED TO SOUTH AUSTRALIAN STATE ELECTIONS, 1977 TO ALP LIB STATE VOTES WON VOTES WON ELECTION WON EXPECTED WON EXPECTED ACTUAL under ACTUAL under (No.) (No). Cube Rule (No.) (No). Cube Rule SOURCES: Expected seat data are calculated from the formula SALP / Sus = [V ALP / VLlSP' reformulated to become SALP = 47x [Vus / VALPP divided by 1+ [VLlB / VALPj3 Calculated from data in: SA. State Electoral Department, 1978, General Elections 1977: Statistical Returns, SA Parliamentary Paper 144 of SA. State Electoral Department, 1980, Periodical and General Elections 1979: Statistical Returns, SA Parliamentary Paper 132 of SA. State Electoral Department, 1982, Periodical and General Elections 1982: Statistical Returns, SA Parliamentary Paper 145 of SA. State Electoral Department, 1986, General Elections 1985: Statistical Returns, SA Parliamentary Paper 145 of SA. State Electoral Department, 1991, Periodical and General Elections 1989: Statistical Returns, SA Parliamentary Paper 161 of SA. State Electoral Office, 1996, Statistical Returns for General Elections 1993 and By-elections 1994, State Electoral Office, Adelaide. NOTE: Totals for 1985 to 1997 include rethrow results for those seats where the final two candidates preferred were not ALP and LIB. Rethrow data were unpublished until 1997.

85 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 22: CUBE RULE - PERCENTAGE OF THE TOTAL VOTE REQUIRED TO WIN A GIVEN NUMBER OF Effective times Vs% = Seats SA SS=(47-SA) SAl Ss cube root of cube root of cube root of VA% (100-VA%) (Party A) SAl Ss SAl Ss SAl Ss SOURCE: Calculated from the formula SAlSs = [VA I Vsy reformulated to become VA = (100 x the cube root of SAlSs) I (1+ the cube root of SAlSB).

86 INFORMATION PAPER 16: MAKING SURE THAT ELECTORAL BOUNDARIES ARE FAIR IN SOUTH AUSTRALIA TABLE 23: CALCULATING AN ESTIMATED OEM: OTHER TWO PARTY PREFERRED VOTE, STATE ELECTION 1997 TWO PARTY FIRST PREFERENCES TOTAL ESTIMATED TWO PARTY PREFERRED VOTE PREFERRED ALP LIB DEM OTHER FORMA FOLLOWING ALP LIB VOTES NAPIER (%) (%) (No.) (No.) (No.) (No.) (No.) ALP DEM (No.) (No.) SAFE ALP PRICE SPENCE HART TAYLOR NAPIER RAMSAY ROSS SMITH ELIZABETH PLAYFORD GILES TOTAL, ALP SAFE FOLLOWING FOLLOWING FOLLOWING FOLLOWING NAPIER FINNISS HEYSEN WAITE WHICH ARE NOT SAFE ALP OR SAFE LIB ALP DEM LIB DEM LIB DEM I LIB DEM (No.) (No.) (No.) (No.) (No.) (No.) (No.) (No.) TORRENS LEE PEAKE KAURNA HANSON REYNELL I WRIGHT ELDER I FLOREY MITCHELL NORWOOD HARTLEY STUART FROME COLTON UNLEY MAWSON ADELAIDE BRIGHT LIGHT COLES NEWLAND FISHER TOTAL, NOT-SAFE FOLLOWING FOLLOWING FOLLOWING FINN ISS HEYSEN WAITE I LIB DEM LIB DEM LIB DEM SAFE LIB I (No.) (No.) I (No.) (No.) (No.) (No.) HEYSEN I WAITE GORDON MORPHETT DAVENPORT HAMMOND SCHUBERT KAVEL FINNISS GOYDER BRAGG MACKILLOP CHAFFEY I FLINDERS I TOTAL, SAFE LIB I I ! SOURCE. calculated from official but as-yet-unpubhshed data from the State Electoral Office.