Research Report 2 - Redistribution Forward Estimates 1984-1999

Updated: 15 January 2013
Download Research Report 2

Redistribution Forward Estimates 1984–1999

Author: Rod Medew

Date: December 2003

Contents

Foreword

In December 2003 a number of officers from the AEC, Australian Bureau of Statistics (ABS) and acknowledged academics met in Canberra to discuss this paper. As a result of discussions a project between the AEC and ABS has commenced to develop a more accurate model of projecting enrolment. The project is continuing.

Research Report – Enrolment Projections

Abstract

A unique feature of the Australian redistribution process is the use of forward enrolment projections which is specified in Section 63A of the Commonwealth Electoral Act 1918 (CEA). Electoral boundaries are drawn on current enrolment figures, which can deviate +/- 10% from the State average. In addition, the redistribution committee use projected data; these projections are based on what the enrolment of each Census Collection District is likely to be in three and a half years, the midpoint of the redistribution period. The methodology employed is an algorithm used by the Australian Bureau of Statistics named the cohort-component methodology and as a projection mechanism has a high degree of confidence amongst demographers.

The accuracy of these projections can only be assessed after the fact. It must be remembered that the projections are estimates and therefore can never be entirely accurate and contain an element of uncertainty. The most obvious indicator of accuracy is the difference between the forecast enrolment and the actual enrolment. In terms of redistribution the accuracy of the real figures, while important, is of less concern than the deviation, which has until recently been +/- 2.5% (now +/- 3.5%) of the State average at the time, that is 3.5 years from the date of the redistribution.

It should also be noted that a major performance indicator of redistribution is malapportionment. The CEA specifies that a redistribution must occur if more than one-third of the Divisions in a State deviate from the State average by more than 10% for a period of more than 2 months. To date no redistribution has been triggered as a result of this criterion.

This paper compares the actual enrolment deviations of each redistributed Division with the projected deviation and observes that in many cases the deviation has been well above or below the estimated deviation and recommends that the AEC seek advice from the Australian Bureau of Statistics into an alternative methodology for projecting enrolment figures.

Enrolment Projections

Purpose

The purpose of this report is to analyse the accuracy of enrolment projections used in the redistribution process.

Background

The conduct of redistributions is governed by the Commonwealth Electoral Act 1918 (CEA). In addition the Constitution plays a part in apportioning the electoral districts among the States and Territories. This paper focuses on one redistribution criterion, in particular enrolment projections.

At a federal level electoral districts have been drawn on a criteria of equal population for some time. The theoretical and legislative basis for substantially equal populations is simple and convincing – each person's vote should have equal weight in the election of representatives. This is achieved by ensuring that at the time of the redistribution no one division deviates by more or less than 10% of the State average.

Redistributions at a federal level changed in 1984 when an additional mathematical principle was supplemented to the criteria for determining electoral boundaries. In 1984 the Commonwealth Electoral Act 1918, s.25S(3) was changed so that the redistribution committee would need to take into account the following:

  • As far as practicable ensure that 3 years and 6 months after the State or Territory had been redistributed, the number of electors enrolled in each Electoral Division in the State or Territory be equal
  • Community of interests within the proposed Electoral Division, including economic, social and regional interests.
  • Means of communication and travel within the proposed Electoral Division.
  • The trend of population changes within each State and Territory.
  • The physical features and area of the proposed Electoral Division.
  • The boundaries of existing Divisions in the State or Territory

The inclusion of a requirement to have all divisions equal at some time in the future was a result of certain recommendations from the Joint Select Committee on Electoral Reform. The Committee's view was that a better method of ensuring equality in enrolments was to be achieved by having areas of high population growth start well below the quota and areas of population decline start above the quota, and at some point in time the two would converge. The date of three years and six months was chosen at it represented the mid point of the seven year redistribution cycle.

The requirement for Redistribution Committees to ensure that each Division was as far as practicable equal in enrolment numbers 3 years and 6 months from the time of redistribution required that the redistribution committee have access to projected enrolment figures. The planning and design of such projected figures depends greatly on the availability of reliable, accurate, relevant and recent data and the ability to analyse and interpret these data. Since the accuracy of any population projection can only be evaluated after the event, this report attempts to evaluate the accuracy of forward enrolment projections by comparing the estimate to the actual figures obtained from the enrolment system.

The first redistributions that required projected enrolment figures 3 years and 6 months from the redistribution date occurred in 1984. Constrained by scarce resources, the Redistribution Committees of 1984 sought the best information available to them, and concluded that the assessments of the Divisional Returning Officers provided the soundest estimates, relying as they did on their local knowledge and on inquiries directed to public authorities with knowledge of probable developments in housing and population movements.

In addition the Redistribution Committee made some changes to the 1984 DRO estimates and regarded the final figures sufficient at the time to create all the Divisions as near as practicable as being equal 3.5 years from the time the redistributions took place.

For the 1989 redistributions the CEA was changed in connection with redistribution estimates. The CEA qualified the acceptable deviation that would be allowed in estimating enrolment figures 3 years and 6 months. The CEA 1918, 66(3) prescribed that:

"In making the proposed redistribution, the Redistribution Committee -

  • Shall as far as practicable, endeavour to ensure that, if the State were redistributed in accordance with the proposed redistribution, the number of electors enrolled in each Electoral Division in the State would not 3 years and 6 months after the State had been redistributed, be less than 98% or more than 102% of the average divisional enrolment of that State at the time."

Consequently the 1989 Redistribution Committees were under a duty to "as far as practicable, endeavour to ensure" that in 3 years and 6 months time enrolments in proposed Divisions would be approximately equal, subject to a variation of 2% above or below the average enrolment for the State.

In pursuing the objective of approximate equally of enrolments in 3.5 years time the Committees sought the best information available, including enrolment trends in recent years and modification of this data by DROs.

In 1992 the AEC developed a computer program to project enrolment figures for each Census Collection District (CCD) with in the State. This was necessary as the Redistribution Committee had begun using computer systems to draw new electoral boundaries using CCDs as basic building blocks.

The algorithm used by the AEC relied on growth rates from Statistical Local Areas (SLA) and applied to all CCDs within the SLA. The underlying assumption was that each CCD within the SLA had a similar growth rate.

The computer generated enrolment figures were occasionally modified by DROs on the basis of their local knowledge and approved by the Australian Electoral Officer of the particular State and also the Redistribution Committee. The figures were published and made available to political parties and the general public for use in submissions.

In 1995 the Joint Standing Committee on Electoral Matters (JSCEM) conducted an inquiry into the effectiveness of the redistribution process. The inquiry received suggestions from the Australian Bureau of Statistics (ABS) that an increased role for the ABS in enrolment projections might improve the accuracy of the forecasts. The ABS submission argued that past growth rates might not be an appropriate indicator of future growth rates, particularly SLAs that are in rapid growth or decline.

The JSCEM recommended that the AEC and ABS form a working party to determine the most effective methodology for enrolment projections. Another recommendation was that after the AEC had agreed on the final projected enrolment projections, the projections be forwarded to the ABS for an opinion to be published in the volumes of the AEC enrolment projections.

As a result of the above recommendation the ABS developed a methodology to predict future enrolment figures. The algorithm employed uses a demographic technique named cohort-component methodology, which is accepted as the most accurate population projection methodology. The ABS projections were again based on CCD level and given to DROs to modify where necessary based on their local knowledge. This system has been in place for all redistributions since 1997.

The JSCEM also recommended that the permissible variation for projected enrolments be increased from 2% to 3.5%. The 3.5% requirement has applied to redistributions from 1999.

Measuring Accuracy

In determining if the projected population is accurate it is important to develop a basis for measuring accuracy and at the same time identify benchmarks for determining equality. The latter is important as the basis for projecting enrolment figures is to ensure that all electoral districts are equal within a 2% variation (now 3.5%). The purpose of the projections is to maintain divisional enrolments as equal as possible throughout the redistribution. To this end what we are actually interested in is not the accuracy of the predictions but the deviations from the quota.

The following indicators have been used in measuring the accuracy of the population projections.

Absolute Difference. The absolute difference is obtained by subtracting the actual enrolment from the projected enrolment of the State in absolute terms.

The Quota. The quota is determined by dividing the total enrolment of the State or Territory by the number of electoral divisions within the State or Territory. Thus if a State's enrolled population is 1025840 and there are 12 divisions the quota would be 1025840/12 = 85486.

Deviation from the Quota. The deviation from the quota examines the degree to which the electoral divisions vary from the quota in absolute terms and expressed as a percentage. The methodology adopted in this paper classifies divisions deviating from the projected quota by greater than 2% and those deviating from the projected quota by more than 4%

Mean Difference. The mean difference is the arithmetic mean of the absolute difference. That is the absolute difference divided by the number of electoral divisions.

Mean Deviation. The mean deviation is the arithmetic mean of the sum of the deviations from the quota.

Maximum Deviation. The maximum deviation from the quota from any division in the State or Territory

Minimum Deviation. The minimum deviation from the quota for any Division in the State or Territory

Divs > 2.0. The total number of Divisions which deviate by more than 2% from the quota. This measure is further broken down into 4 categories. The number of Divisions deviating between 2.% and 3.0%, 3.1% and 4%, 4.1% and 5% and those divisions deviating by more than 5%.

Dw Index. This is an index of variation, ranging from 0 to 1. If all Divisions are within the 2% tolerance then the index is set at 0. The index is weighted according to the size of the deviation. The further the index is from 0 the more severe the deviations are. While there will inevitability be difference in the projected and actual enrolment figures, the more significant figure is the deviation from the quota. The variance should be within a 2% range (now 3.5%) at the 3 year and 6 month period. Therefore the number of divisions that fall outside this range is perhaps the most significant benchmark of success in forecasting enrolment projections.

1984 Redistributions

The Commonwealth Electoral Legislation Amendment Act 1983 made it necessary for all States and the ACT to be redistributed in 1984. The Representation Act 1983 increased the number of Senators from 10 to 12 and consequently the representation in the House of Representatives of all States except Tasmania was altered.

In all States and the ACT enrolment was projected to December 1987. There was no mathematical methodology employed for these projections, rather DROs were asked to project enrolment based on their own local knowledge. In addition the Redistribution Committees of 1984 exercised their power to make some changes to the DRO estimates. The results of the redistributions for all States and the ACT can be found in Attachment 1A – 1D of this document.

Below is a summary of the 1984 redistribution.
State Projected Enrolment No. Divs Mean Deviation Divs > 2% % Divs > 2% Dw
NSW Dec 1987 51 2.58 25 49.0 .28
VIC Dec 1987 39 3.04 21 53. .35
QLD Dec 1987 24 3.49 14 58.3 .42
SA Dec 1987 13 1.61 5 38.4 .09
WA Dec 1987 13 0.91 4 30.7 .09
TAS Dec 1987 5 0.55 0 0 0.0
ACT Dec 1987 2 2.91 2 100 .25

There was no definition of equality prescribed for the 1984 redistributions but since a 2% tolerance was put into place after 1984, this criterion has been used to measure the effectiveness of the 1984 redistributions.

The above table demonstrates that the enrolment estimates were far from satisfactory. A total of 49% of the divisions in NSW deviated by more than 2%, Victoria had 53% of the States divisions deviating by more than 2% and Queensland was the worst positioned State with 58% of the States division's deviating by more than 2%.

Of more concern are those divisions, which were well above or below 2% tolerance. The table below identifies all those divisions that were above or below 4% deviation from the quota (twice the 2%). The States of New South Wales, Victoria and Queensland all recorded deviations greater than 4%. Queensland appears to have been the worst predicted State with 9 of the 24 divisions deviating by more than 4% from the quota. One possible explanation is the unexpected high growth of Queensland and unusually high internal migration.

State Division Projected Deviation Actual Deviation
NSW Cowper +0.16 +5.21
NSW Cunningham +0.53 -4.33
NSW Gilmore -0.40 -4.07
NSW Greenway +0.70 -5.45
NSW Newcastle -0.89 -5.00
NSW Phillip -2.58 +5.59
NSW Prospect -0.26 +7.64
NSW Riverina-Darling -0.08 -6.16
NSW Throsby +0.82 -7.60
NSW Werriwa -0.86 +8.59
QLD Fadden +0.78 -6.9
QLD Fisher +0.20 +5.83
QLD Forde +0.38 -6.99
QLD Herbert +0.36 +5.25
QLD McPherson -0.49 +14.17
QLD Moreton +0.16 -4.17
QLD Oxley -0.48 -5.15
QLD Rankin +0.76 -5.46
QLD Wide Bay -0.21 -5.25
VIC Batman -0.2 -4.8
VIC Bruce +0.3 -5.2
VIC Holt 0.1 -5.2
VIC Jagajaga -0.7 -7.2
VIC Scullin 0.00 -9.5

1989 Redistributions

In 1989 two States (Victoria and Western Australia) were redistributed. The enrolment projections were for the first time calculated using Local Government Areas and the Redistribution Committees were under obligation to "as far as practicable, endeavour to ensure" that in 3 years six months time enrolments in proposed Divisions should be approximately equal subject to a variation of 2% above or below the State average. The AEC used computer trends from the enrolment system and used an algorithm to extrapolate the projected enrolment figure. Divisional Returning Officers subsequently modified these figures.

The following table summarises the two redistributions in 1989.
State Projected Enrolment No. Divs Mean Deviation Divs > 2.0 % Divs > 2.0 Dw
VIC Sept 1992 38 2.24 14 36.8 0.24
WA Sept 1992 14 1.62 3 21.4 0.17

The results for both these States in terms of projections improved considerably from the 1984 redistributions. The number of Divisions deviating by 2% fell from 21 to 14 for Victoria. The Redistribution Committee noted the difficulty with the 2% rule:

"The loss of one Division in Victoria, taking the States total down to 38 ensured that substantial changes to divisional boundaries would be necessary. Projected enrolment growth and decline over the 3½ years period is located unevenly across the State and greatly increases the pressure for major alternations in a number of areas. Having regard to the loss or increase in projected enrolment numbers necessary to approximate average enrolment in September 1992, there were three existing Divisions which would need to lose more than 10 000 "future" electors (i.e. lose 10 000 from their projected enrolments), Flinders (16 700), Lalor (15 600), and Burke (14 400), and four other Divisions (Corangamite, Indi, La Trobe and McEwen) which would need to lose between 5 000 and 10 000 "future electors". Twelve Divisions (Bruce, Chisholm, Deakin, Gellibrand, Goldstein, Henty, Higgins, Isaacs, Kooyong, Maribyrnong, Melbourne and Melbourne Ports) need to gain between 5 000 and 10 000. Only 3 of the existing Divisions could comply with the statutory requirements without alternation, and their continuation unchanged would be dependent on the needs of their neighbours to lose or gain electors. This is an aspect of the redistribution process frequently misunderstood. If Division A requires 5 000 electors and can get them best, and at times only from Division B and Division B also needs 5 000 electors, then a combined shortfall may have to be met from Division C even though that Division could be left alone because it already met the statutory requirements. Ripple effects may spread through a number of divisions before they are spent."

The difficulties in complying with the new requirement that all divisions deviate no more or less than 2% of the State average 3 years and 6 months from the redistribution date are manifested in the table set out below which identifies those divisions that were above twice the allowed deviation (4%).
State Division Projected Deviation Actual Deviation
VIC Calwell -0.5 +5.1
VIC Corinella -0.9 +7.0
VIC Dunkley -1.2 -4.9
VIC Flinders +1.6 -4.6
VIC Holt -0.6 -6.3
VIC Jagajaga -1.4 -6.4
VIC Scullin -1.1 +4.2
WA Brand +1.8 +5.1
WA Canning -0.8 -5.7

The Divisions of Jagajaga, Scullin and Holt in Victoria were well over the 2% tolerance in both the 1984 and 1989 elections. However in general terms the accuracy of the projections appears to have improved for both Victoria and Western Australia. In 1984 53% of the Division in Victoria were outside the 2% range compared with only 36% for the 1989 redistributions. Western Australia also improved with those divisions falling outside the 2% tolerance form 30.7% to 21.1%. Nonetheless in both States the number of Divisions that were outside 4% of the States average increased. It should be noted that in 1984 there was no 2% tolerance, the redistribution committee were simply asked that as far as practicable the divisions be equal in 3 years and 6 months.

1992 Redistributions

In 1992 New South Wales, Queensland, South Australia and Tasmania were redistributed. Due to a redistribution for Queensland occurring in 1994 there is little point in comparing actual enrolment figures for this State as the figures in the year 1994 will reflect the new electoral boundaries and thus overtake the projected figures.

In 1992 the AEC began using Census Collection Districts (CCDs) as the basis of the projections and also using the CCDs as building blocks for the Divisions. Accordingly the AEC developed an algorithm to determine the projected enrolment at each CCD. Central to this algorithm was the growth rate for each Statistical Local Area (SLA). The SLA's in Australia are generally equivalent with Local Government Areas and consequently consist of a number of CCDs. The Australian Bureau of Statistics (ABS) supplied the AEC with growth rates for all CCDs over a 5 year period.

The projected number of electors in each CCD was initially determined by applying the ABS supplied growth rate for each SLA to all CCDs within each SLA. The formulae used to calculate the number of projected electors for a particular SLA was:

Up = Ea (1 + r) n

Where Up = Projected enrolment Ea = Current enrolment

  1. = average population growth rate for the SLA

n = time in years until projected date.

A number of assumptions were made in the course of calculating the projections.

  • Enrolment trends followed the population trends for Australian citizens aged 18 years and over.
  • The enrolment population would continue to grow or decline at the same the same average annual rate as for 18+ citizens grew during the period identified by the growth rate.
  • The growth rate applied to a given SLA would apply uniformly to all CCDs within the SLA.

These projections were then referred to the Divisional Returning Officers for review and modification.

In drafting the new boundaries, the Redistribution Committee made use of a new computer system, which was developed by the CSIRO and the AEC. The system displayed all the CCDs with each CCD containing details of current enrolment and projected enrolment. In addition the system displayed the SLA and State electoral boundaries as well as the existing federal electoral boundaries.

By using the mouse to move CCDs from one electoral boundary to another the Redistribution Committees were able to instantly evaluate the proposed electoral boundaries.

Proposed electoral boundaries
State Projected Enrolment No. Divs Mean Deviation Divs > 2.0 % Divs > 2.0 Dw
NSW Mar 1995 50 2.32 28 56.0 .24
SA Mar 1995 12 3.12 8 66.6 .37
TAS Oct 1995 5 1.89 2 40.0 .15
The number of Divisions where the deviation was greater or less than twice the allowable deviation is as follows.
State Division Projected Deviation Actual Deviation
NSW Berowra +0.5 +4.6
NSW Charlton 0.0 +4.0
NSW Fowler -0.5 +5.8
NSW Gimore -0.7 -5.0
NSW Mitchell +1.2 -4.1
NSW North Sydney +1.3 +5.4
SA Bonython +0.8 -9.7
SA Mayo -1.1 +4.1
SA Sturt -0.6 -5.7

1994 Redistributions

In 1994 Victoria, Queensland and the Australian Capital Territory were redistributed. However, both the A.C.T and Queensland projected figures to 1998. In 1997 both these States were redistributed again, effectively overwriting the projected figures. Only Victoria is analysed.

Once again the AEC used its algorithm to predict the enrolment and the redistribution committee used the ITA system to draft boundaries.
State Projected Enrolment No. Divs Mean Deviation Divs > 2.0 % Divs > 2.0 Dw
VIC Jun 1998 37 2.79 21 56.7 .32
The total number of Divisions greater than the 4% is shown below.
State Division Projected Deviation Actual Deviation
VIC Batman +0.28 +6.17
VIC Burke +1.88 -4.21
VIC Calwell +0.84 +4.53
VIC Corangamite -1.85 -4.35
VIC Goldstein +1.86 +5.69
VIC Hotham +1.82 +6.60
VIC Isaacs +1.79 -5.27
VIC La Trobe -0.58 -5.29
VIC Melbourne +0.61 +4.47
VIC Wills -0.47 +4.89

1997 Redistributions

In the redistributions prior to 1997, the AEC had employed an algorithm to predict enrolment growth for each CCD by applying a modified component interest formulae to each of the CCDs within a SLA.

In a submission to the JSCEM reviewing the redistribution process, the ABS suggested that the Bureau had developed algorithms that could improve the reliability of projections. The JSCEM recommended the ABS and AEC work together on enrolment projections.

The ABS supplied enrolment projections using a cohort component methodology. The approach involves the computation of separate age-sex groups on the basis of separate allowances. The base population is applied with survival rates, birth rates, migration rates and other demographic information. The final product is a population projection, which in this case are enrolment figures at a CCD level.

These projections were once again submitted to the DROs where they were asked to examine and modify if necessary the figures in light of their local knowledge. The projections are also subject to possible modification by the Redistribution Committee.

The ITA system, which the redistribution had previously used was no longer supported and consequently de-commissioned. The AEC developed its own system based on MAPINFO software. The system named Electoral Boundary Mapping System (EBMS) displayed the CCD, LGA and electoral boundaries. Each CCD also contained the current and projected enrolment, enabling the redistribution Committees to move individual or groups of CCDs from one boundary to another and see the results.

The three States that were redistributed in 1997 were Queensland, Western Australia and the Australian Capital Territory.

The following table details the results of the projections.
State Projected Enrolment No. Divs Mean Deviation Divs > 2.0 % Divs > 2.0 Dw
QLD Jun 2001 27 3.69 17 62.9 .41
WA Jun 2000 14 2.27 9 64.2 .21
ACT Mar 2001 2 0.67 0 0 0.0
Those divisions where the deviation from the quota was greater or lesser than 4% are outlined below.
State Division Projected Deviation Actual Deviation
QLD Blair -1.6 -10.9
QLD Brisbane -0.8 +10.9
QLD Capricornia +1.6 -5.5
QLD Griffith -0.9 +7.0
QLD Herbert +1.9 +4.7
QLD Hinkler -1.8 -5.6
QLD Longman +1.4 -4.2
QLD McPherson +1.0 +8.7
QLD Wide Bay -0.6 -5.9
WA Cowan -0.5 -5.7

1999 Redistributions

The 1999 redistributions were affected by new legislation introduced in Section 63A which enabled the AEC to determine an earlier projection time, which was normally three years and six months, where in the opinion that a further redistribution may be required sooner than seven years.

Section 66(3)(a) of the Act was also changed and required the Committee to "as far as practicable, endeavour to ensure that, if the State or Territory were redistributed in accordance with the proposed redistribution, the number of electors enrolled in each Electoral Division in the State or Territory would not, at the projected time determined under section 63A, be less than 96.5% or more than 103.5% of the average divisional enrolment of that State or Territory at that time."

The AEC continued to employ the services of the ABS and also use the new EBMS systems. Only one State was redistributed in 1999, the State of South Australia. On the basis of population projections the Redistribution Committee was of the opinion that a further redistribution may be required in SA sooner than the mandatory seven years and applied Section 63A(4) of the Act which enabled the projection date of the enrolment to be 30 June 2001.

The table below summarise the SA redistribution.
State Projected Enrolment No. Divs Mean Deviation Divs > 2.0 % Divs > 2.0 Dw
SA Jun 2001 12 1.26 1 8.3 0.02

No divisions had a variation from the quota of more or less than 4%.

1999–2000 Redistributions

Redistributions were conducted in New South Wales and Tasmania in 1999–2000. There was no change in the Redistribution provisions between the 1999 Redistribution and the 1999–2000 Redistributions.

The table below summarises the 1999–2000 Redistributions.
State Projected Enrolment No. Divs Mean Deviation Divs > 2.0 % Divs > 2.0 Dw
NSW Jun 2003 50 2.54 12 24.0  
TAS Jun 2003 5 1.96 1 20.0  
Those divisions where the deviation from the quota was greater or lesser than 4% are shown below.
State Division Projected Deviation Actual Deviation
NSW Bradfield +2.22 +5.17
NSW Cunningham -1.09 -5.29
NSW Eden-Monaro -0.28 4.57
NSW Lyne +2.01 +5.29
NSW Mitchell +2.86 +7.16
NSW Newcastle +1.85 +5.28
NSW Reid +2.71 -10.31
NSW Sydney -1.79 +4.12
NSW Watson -2.46 -4.33

Summary

The following tables summarise the redistributions between 1984–2000.

NSW
Year Mean Deviation > 2% (3.5%) > 4% Dw Method
1984 2.58 49.0 19.6 .28 a
1992 2.32 56.0 12.0 .24 c
1999–2000 2.54 24.0 18.0 .11 c
VIC
Year Mean Deviation > 2% (3.5%) > 4% Dw Method
1984 3.04 53.8 12.8 .35 a
1989 2.24 36.8 18.4 .24 b
1994 2.79 56.7 27.0 .32 c
QLD
Year Mean Deviation > 2% (3.5%) > 4% Dw Method
1984 3.49 58.3 37.5 .42 a
1997 3.69 62.9 33.3 .41 c
WA
Year Mean Deviation > 2% (3.5%) > 4% Dw Method
1984 0.91 30.7 - .09 a
1989 1.62 21.4 14.2 .17 b
1997 2.27 64.2 7.1 .21 c
SA
Year Mean Deviation > 2% (3.5%) > 4% Dw Method
1984 1.61 38.4 - .09 a
1992 3.12 66.6 25.0 .37 b
1999 1.26 8.3 - .02 c
TAS
Year Mean Deviation > 2% (3.5%) > 4% Dw Method
1984 0.55 0 0 .00 a
1992 1.89 40.0 0 .15 b
1999–2000 1.96 20.0 0 .05 c
ACT
Year Mean Deviation > 2% (3.5%) > 4% Dw Method
1984 2.91 100 - .25 a
1997 0.67 0 0 0.00 c

Conclusion

A useful indicator of the success of the projection estimate is the number of divisions within a particular State that deviate by a large proportion from the quota. Previously in this paper the metric used for this indicator has been 4% or twice the allocable deviation.

The following table identified those electoral divisions, which have deviated by more than 4% in more than one redistribution.
State Division Projected Deviation Actual Deviation
NSW (1984) Cunningham +0.53 -4.33
NSW (1999–2000) Cunningham -1.09 -5.29
NSW (1984) Gilmore -0.40 -4.07
NSW (1992) Gilmore -0.7 -5.00
NSW (1992) Mitchell +1.2 -4.1
NSW (1999–2000) Mitchell +2.86 +7.16
NSW (1984) Newcastle -0.89 -5.00
NSW (1999–2000) Newcastle +1.85 +5.28
QLD (1984) Herbert +0.36 +5.25
QLD (1997) Herbert +1.9 +4.7
QLD (1984) McPherson -0.49 +14.17
QLD (1997) McPherson +1.0 +8.7
QLD (1984) Wide Bay -0.21 -5.25
QLD (1997) Wide Bay -0.6 -5.9
VIC (1984) Batman -0.2 -4.8
VIC (1994) Batman +0.28 +6.17
VIC (1989) Calwell -0.5 +5.1
VIC (1994) Calwell +0.84 +4.53
VIC (1984) Holt +0.1 -5.2
VIC (1989) Holt -0.6 -6.3
VIC (1984) Jagajaga -0.7 -7.2
VIC (1989) Jagajaga -1.4 -6.4

All population projections are based on a set of assumptions that are rarely stated explicitly. In general terms the projections assume there will be no deep structural change from one event to another. Projections can never be entirely accurate due to many unforseen circumstances. In the case of the enrolment projections the actual projected figures are not really important, what is, is the deviation from the average enrolment. At the time of projection the figures have been projected so as to allow no one division to deviate by more or less than 2% (or 3.5% after 1997) of the State average. This criterion is what we can place some measure of accuracy on.

The first projections were made in 1984 with no agreed measure of accuracy. From 1989 until 1999 the allowable deviation from the State average for a division has been +/- 2%. In 1999 this was changed to +/- 3.5 %. It is this requirement that is used to measure the accuracy of projected enrolment.

The 1984 redistribution projections presented to the Redistribution Committees had no ABS input and were based solely on DRO input. In addition the 1984 redistributions used subdivisions as building bocks limiting the flexibility for small-scale modifications of redistributions. In the light of these constraints the 1984 redistribution projections must be seen as successful with the exception of a few divisions. Notably the Division of McPherson, which deviated by more than 14% of the State average.

The 1989 redistributions were aided by a report produced by the AEC, which identified enrolment figures for each electoral division on a monthly basis. These data were used to project figures, which were subsequently modified by DROs. A different approach was used in 1994 and 1997 where the AEC developed algorithms based loosely on the component interest formulae. This algorithm projected enrolment at the CCD level. Since 1997 the AEC employed the ABS to project enrolment for the purposes of redistribution. The actual enrolment figures show no evidence of any improvement in projections between the use of the AEC and the ABS algorithms.

In fact both algorithms demonstrate the difficulties in estimating enrolment figures especially at a low geographical level such as CCD. These difficulties have often resulted in divisions deviating by large values from the state average and consequently divisions being arguably malaportioned mid way in the redistribution cycle.

The following table uses the deviation index (Dw Index) for all States since 1984. A State where there are no deviations above 2% (or 3.5% after 1997) will record an index of 0.0 while a State with a large number of deviations larger than 2% (or 3.5%) will record a figure between 0.1 and 1.0.
State 1984 1989 1992 1994 1997 1999 1999–2000
NSW .28   .24       .11
VIC .35 .24   .32      
QLD .42       .41    
SA .09   .37     .02  
WA .09 .17     .21    
TAS .00   .15       .05
ACT .25       .00    

It is generally acknowledged that projecting small area populations is difficult. The AEC has employed a number of different strategies to date and perhaps it is time to revisit the methodology with an aim of improving the estimates. It may also prove to not to be feasible to make any improvements. Nonetheless it is suggested that the AEC, ABS and some academic institutions get together and discuss some of the issues with a view to determining the feasibility of improving the estimates.