Nội dung

ACCOUNTING FOR POPULATION AGEING IN TAX MICROSIMULATION MODELLING BY SURVEY REWEIGHTING* LIXIN CAI, JOHN CREEDY and GUYONNE KALB University of Melbourne This paper investigates the use of sample reweighting, in a behavioural tax microsimulation model, to examine the implications for government taxes and expenditure of population ageing in Australia. First, a calibration approach to sample reweighting is described, producing new weights that achieve specified population totals for selected variables. Second, the performance of the Australian Bureau of Statistics’ (ABS) weights provided with the 2000–2001 Survey of Income and Housing Cost (SIHC) was examined and it was found that reweighting does not improve the simulation outcomes for the 2001 situation, so the original ABS weights were retained for 2001. Third, the implications of changes in the age distribution of the population were examined, based on population projections to 2050. A ‘pure’ change in the age distribution was examined by keeping the aggregate population size fixed and changing only the relative frequencies in different age-gender groups. Finally, the effects of a policy change to benefit taper rates in Australia were compared for 2001 and 2050 population weights. It is suggested that this type of exercise provides an insight into the implications for government income tax revenue and social security expenditure of changes in the population, indicating likely pressures for policy changes. I. Introduction The aim of this paper is to investigate the use of sample reweighting, in a behavioural tax microsimulation model, to examine the implications for government taxes and expenditure of population ageing in Australia. Tax microsimulation models are based on large-scale cross-sectional surveys containing substantial information about the characteristics of individuals and households. Each household has a sample weight provided by the statistical agency responsible for collecting the data, and these weights are used to ‘gross up’ from the sample in order to obtain estimates of population values. This applies to aggregates such as income taxation, the number of recipients of a particular social transfer or the number of people in a particular demographic group. In addition, the weights are used in the estimation of measures of population inequality and poverty. The possibility therefore arises of adjusting the sample weights to reflect anticipated changes in the population age structure. Such a change in population structure could have important policy implications. Consider, for example, an increase in the proportion of individuals over 65, who in principle are eligible for the Age Pension and traditionally have low labour force participation rates, relative to the proportion of individuals between 25 and 64. If age-specific participation rates do not change, an increase in the proportion of individuals over 65 gives rise to higher Correspondence: Assoc Prof Guyonne Kalb, Melbourne Institute of Applied Economic and Social Research, University of Melbourne, Victoria 3010, phone +61 3 8344 2095, email: g.kalb@unimelb.edu.au * We should like to thank the Department of Family and Community Services for funding this research and an anonymous referee for helpful comments. The views expressed in this paper are those of the authors and do not represent the views of the Minister for Family and Community Services, the Department of Family and Community Services or the Commonwealth Government. © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 2006 ACCOUNTING FOR POPULATION AGEING 19 government expenditure and lower government revenue. Revised weights, based on the changed population structure, can be used to estimate implications for labour force participation and government expenditures, on the assumption that other characteristics remain unchanged. That is, at the individual level, the outcomes remain unchanged. What changes is the weight assigned to each of the individuals’ outcomes when calculating the aggregate effects. Any population age structure could be assessed, allowing the effects of alternative future scenarios to be evaluated. Other possibilities of adapting individual characteristics could also be considered at the same time, including for example the potential effects of real wage growth or of changes to the tax and benefit regime. Such changes are of course likely to arise partly as a response to the pressures of population ageing, so it is useful to be able to examine the precise nature of those pressures. The microsimulation approach combined with reweighting contrasts with a popular method of examining population ageing, which combines population projections with age-specific per capita expenditures on a range of benefits in order to obtain projected social expenditures. These are typically combined with GDP projections based on age-specific labour force participation and unemployment ratios, along with productivity growth assumptions. While accounting frameworks of this type have proved useful, they necessarily lack the kind of policy modelling, detail and heterogeneity available in microsimulation models.1 The microsimulation model used here is the Melbourne Institute Tax and Transfer Simulator (MITTS). This is a behavioural tax microsimulation model allowing detailed examination of the potential effects on government direct tax revenue and expenditure of policy reforms to the tax and transfer system.2 The database is the Australian Survey of Income and Housing Costs (SIHC), a large-scale cross-sectional survey of about seven thousand households, with each household having a sample weight provided by the Australian Bureau of Statistics (ABS). This paper begins by describing a calibration approach to sample reweighting which achieves specified population totals for selected variables, subject to the constraint that there are minimal adjustments to the weights. A formal statement of the problem of obtaining ‘minimum distance’ weights and a general approach to the solution are described in Section II. Section III applies the approach to the SIHC, and considers whether – before examining population ageing – the SIHC needs to be reweighted for tax simulation purposes.3 Reweighting to allow for population ageing is examined in Section IV, which makes use of ABS population projections. The analysis abstracts from changes in population size and concentrates purely on changes in the age structure. Section V reports simulation results of a tax policy change with the revised weights, reflecting the aged population structure. Brief conclusions are in Section VI. II. The Calibration Approach This section discusses methods of calibration. Subsection II.a provides a general statement of the problem of minimising the overall distance between two sets of weights, subject to a set of calibration conditions. Subsection II.b examines a class of distance functions giving rise to a convenient structure. An iterative solution procedure is presented in subsection II.c. 1 On this type of modelling, see Alvarado and Creedy (1998). For further details of the MITTS model, see Creedy et al. (2002). 3 The ABS weights are calibrated to provide correct aggregates at the population level with regard to for example age and household composition, but not necessarily to provide a good representation of benefit recipient groups, such as for example sole parents on Parenting Payments Single or families receiving NewStart Allowance. Given the importance of these groups in tax microsimulation, reweighting might be necessary if the ABS weights are not adequate to represent these groups correctly. 2 © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 20 AUSTRALIAN ECONOMIC PAPERS MARCH a) Statement of the problem For each of K individuals in a sample survey, information is available about J variables; these are placed in the vector:  xk ,1   .    xk =  .   .     xk , J  (1) For present purposes these vectors contain only the variables of interest for the calibration exercise. Most elements of xk are likely to be 0/1 variables. For example xk , j = 1 if the kth individual is in a particular age group, and zero otherwise. The sum ∑ kK=1 xk , j therefore gives the number of individuals in the sample who are in the age group. The sample design weights, provided by the statistical agency responsible for data collection, are sk for k = 1, . . . , K. These weights can be used to produce estimated population totals, tx| s, based on the sample, given by the J-element vector: K ∑ sk xk t x| s = (2) k =1 The calibration approach can be stated as follows. Suppose that other data sources, for example census or social security administrative data, provide information about ‘true’ population totals, tx. The problem is to compute new weights, wk, for k = 1, . . . , K which are as close as possible to the design weights, sk, while satisfying the set of J calibration equations: tx = K ∑ wk xk (3) k =1 It is thus necessary to specify a criterion by which to judge the closeness of the two sets of weights. In general, denote the distance between wk and sk as G(wk, sk ). The aggregate distance between the design and calibrated weights is thus:4 D= K ∑ G(wk , sk ) (4) k =1 The problem is therefore to minimise (4) subject to (3), for which the Lagrangean is: L= K J k =1 j =1  K  ∑ G(wk , sk ) + ∑ λ j  tx, j − ∑ wk xk , j  (5) k =1 where λ j for j = 1, . . . , J are the Lagrange multipliers. b) A class of distance functions Suppose that G(wk, sk) is such that the differential with respect to wk can be expressed as a function of wk /sk, so that: w  ∂G(wk , sk ) = g k  ∂wk  sk  (6) 4 Some authors, such as Folsom and Singh (2000) write the distance to be minimised as ∑kK=1 sk G ( wk , sk ) , but the present paper follows Deville and Särndal (1992). © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 2006 ACCOUNTING FOR POPULATION AGEING 21 The K first-order conditions for minimisation can therefore be written as: w  g  k  = xk′ λ  sk  (7) Write the inverse function of g as g −1, so that if g (wk /sk) = u, say, then wk /sk = g−1(u). From (7) the k values of wk are expressed as: wk = sk g −1 ( xk′ λ ) (8) If the inverse function, g−1, can be obtained explicitly, equation (8) can be used to compute the calibrated weights, given a solution for the vector, λ. The Lagrange multipliers can be obtained by post-multiplying (8) by xk, summing over all k = 1, . . . , K and using the calibration equations, so that: tx = K K k =1 k =1 ∑ wk xk = ∑ sk g −1( xk′ λ )xk (9) Finally, subtracting t x| s = ∑kK=1 sk xk from both sides of (9) gives the nonlinear equations: t x − t x| s = K ∑ sk {g −1( xk′ λ ) − 1}xk (10) k =1 where sk {g −1 ( xk′ λ ) − 1} is a scalar and the left hand side is a known vector. An iterative procedure to solve these equations is given in Appendix A. c) The Deville and Särndal distance function The solution procedure requires only an explicit form for the inverse function g −1(u), from which its derivative can be obtained. Hence, it is not necessary to start from a specification of G(w, s).5 Deville and Särndal (1992) suggested the use of an inverse function g −1(u) of the form:6 g −1 (u) = rL (rU − 1) + rU (1 − rL )exp αu (rU − 1) + (1 − rL )exp αu (11) where rL and rU are the lower and upper limit of the allowed proportionate change in weight with rL < 1 < rU and: α= rU − rL (1 − rL )(rU − 1) (12) Thus g −1(−∞) = rL and g−1(∞) = rU, so that the limits of w/s are r L and rU. Hence the new weights are kept within the range, rLsk < wk < rUsk, without the need to make checks during computation.7 5 Deville and Särndal (1992) discuss the use of a normalisation whereby g−1′(0) is set to some specified value, but this is not necessary for the approach. 6 Singh and Mohl (1996), in reviewing alternative calibration estimators, refer to this ‘inverse logit-type transformation’ as a Generalised Modified Discrimination Information method. Folsom and Singh (2000) propose a variation on this, which they call a ‘generalised exponential model’, in which the limits are allowed to be unit-specific. In practice, they suggest the use of three sets of bounds for low, medium and high initial weights. 7 In implementation, the limits are not fixed. They are adjusted to make the range as small as possible, conditional on a solution being found. © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 22 AUSTRALIAN ECONOMIC PAPERS III. MARCH The Survey of Income and Housing Costs (SIHC) This section checks the performance of the ABS weights provided with the SIHC against an extensive set of calibration conditions, and reports revised weights and MITTS totals. The most recent dataset available is for 2001, which is used here. The calibration conditions include demographic variables, such as age, family composition, unemployment by age and income support recipiency. For the first three variables, population information is taken from census data (ABS, 2002), while information on the last variable is obtained from administrative data on income support payments. Details of the calibration conditions are given in tables presented in Appendix B. A comparison between SIHC and census numbers on the age distribution of males and females reveals that the SIHC appears to have too many people in the lower age groups and too few in the higher age groups, particularly in the highest age group for women.8 With regard to family composition, except for the group of sole parents with dependent and nondependent children, all groups appear to be over-represented in the SIHC.9 Furthermore, the ABS weights understate the number of unemployed men in all but the 15–19 and 35–44 age groups. In contrast, the ABS overstates the number of unemployed women in all but the 20 –34 age groups. The numbers of income support recipients are taken directly from the observed values in the SIHC, according to self-reported responses. There are both under- and overestimates of particular subgroups. The iterative approach described in Section II and Appendix A was applied using the calibration conditions listed in Appendix B. Lower and upper bounds of 0.68 and 1.87 were obtained after experimentation to find the smallest possible range. Figure 1 presents the distribution of the ratio of the new weight to the ABS weight. Relatively few people have a new weight that is more than 1.4 times as large as the ABS weight. Around 50 per cent of all observations are reweighted by a factor between 0.85 and 1.20. Before considering the performance of MITTS with these new weights, it is useful to compare a few summary measures resulting from calculations using the old and the new weights. First, consider the simulated number of income support recipients based on the two sets of weights, shown in Table I. The reweighting has had little effect on most types of income support recipients. The two main exceptions are disability support pensioners, which show an improvement, and age pensioners, where the difference between actual and simulated numbers becomes bigger. The latter is caused by the reweighting on age, putting additional weight on the older age groups. The MITTS model overestimates the proportion of older persons eligible for the Age Pension as a result of the lack of information on assets held by households. People over 60 are amongst those most likely to have built up assets in the form of superannuation or other investments.10 The aggregate expenditures for a range of benefits produced directly by the SIHC for the old and new weights may be compared; that is, the actual benefits reported as being received by individuals in the SIHC are used. Comparisons are shown in Table II, which reports estimated expenditure obtained directly from the SIHC when aggregated using the ABS weights and when 8 This is probably caused by the fact that the SIHC excludes people in institutions or people living in remote areas, whereas these groups are included in the Census. An alternative reweighting could be based on total numbers from the Census excluding these groups if possible. 9 The number of families in the group ‘other types of family’ is omitted from the calibration conditions to avoid singularities. 10 Some alternative approaches are reported in Cai, Creedy and Kalb (2004). For example, observed benefit receipt was used as a requirement for taking up of Age Pension, and people with eligibility for benefits under $10 per week were assumed not to take up these benefits. Using observed eligibility for the Age Pension instead of assets (which are not observed) improves the simulation of the number of recipients, but does not improve the estimated expenditure. © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 2006 Figure 1. Table I ACCOUNTING FOR POPULATION AGEING 23 Ratio of new weight to ABS weight Actual and simulated numbers (in ’000s) of income support recipients Parenting Payment (single & couple) Sickness Allowance Widow’s Allowance AUSTUDY/ABSTUDY NewStart Allowance Mature Age Allowance Youth Allowance Special Benefit Partner Allowance Age Pension Disability Support Pension Wife’s Pension Widow B Pension Carer’s Payment Total Actual from FaCS1 (1) Simulated using ABS weights (2) Simulated using new weights (3) Difference between (1) and (2) Difference between (1) and (3) 639 11 36 42 541 39 393 12 90 1 786 624 78 9 57 4 357 674 21 1 135 660 47 674 232 215 1 935 575 101 41 33 5 344 602 22 0 150 690 47 671 246 212 2 094 615 93 38 33 5 513 −35 −10 35 −93 −119 −8 −281 −220 −125 −149 49 −23 −32 24 −987 37 −11 36 −108 −149 −8 −278 −234 −122 −308 9 −15 −29 24 −1 156 Note: 1, The source for column 1 is FaCS (2003). aggregated using the revised weights. There seems to be a slight overall improvement resulting from using the new weights. However, when examining particular payment types separately, for some types the amount is much further from the actual amount than before the reweighting, whereas for other types an improvement is evident. Finally, the performance of MITTS with regard to expenditures using the different sets of weights is illustrated in Table III. Table III is similar to Table II but compares the simulated expenditure based on the reweighted SIHC data with the simulated expenditure based on the original SIHC data. Comparing Tables II and III, it can be seen that the difference between the actual expenditure and the simulated expenditure is smaller than the difference between the actual expenditure and the expenditure observed from the SIHC. However, the reweighting does not improve the simulated expenditure. In fact, the difference between actual and simulated expenditure for 2001 is quite small with the initial weights, although there are a few exceptions. Regarding the Widow’s Allowance and the Widow B Pension it seems that the two payments cannot be separated as they should, but in aggregate the simulated amount paid on these is quite close to the actual © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 24 Table II AUSTRALIAN ECONOMIC PAPERS MARCH Actual and estimated expenditure on income support Actual from FaCS 1 ($m) (1) SIHC using ABS weights ($m) (2) SIHC using new weights ($m) (3) Diff. between (1) and (2) Diff. between (1) and (3) Parenting Payment (single & couple) Sickness Allowance Widow’s Allowance AUSTUDY/ABSTUDY NewStart Allowance Mature Age Allowance Youth Allowance Special Benefit Partner Allowance Age Pension Disability Support Pension Wife’s Pension Widow B Pension Carer’s Payment 5 327.0 95.9 330.2 255.6 4 918.3 353.1 2 121.6 113.8 717.1 15 571.8 5 837.4 680.0 75.3 478.3 4 911.3 212.7 402.5 n/a2 3 466.2 329.1 1 446.2 150.5 605.2 14 233.9 5 182.7 491.7 n/a2 605.2 4 303.7 223.0 369.1 n/a 3 858.9 304.9 1 521.1 164.9 668.2 15 681.1 5 656.3 445.0 n/a 582.3 415.7 −116.8 −72.3 1 023.2 −127.1 −39.0 1 452.1 24.1 675.4 −36.6 111.9 1 337.9 654.7 188.3 1 059.5 48.2 600.5 −51.1 48.9 −109.4 181.1 235.0 Total 36 875.4 32 037.2 33 778.5 −127.0 −104.0 4 507.4 2 765.8 Notes: 1, The source for column 1 is FaCS (2001), 2, AUSTUDY and Widow B Pension cannot be identified from the SIHC data. Table III Actual and simulated expenditure on income support Simulated using Actual from FaCS(1) ($m) (1) ABS weights ($m) (2) new weights ($m) (3) Diff. between (1) and (2) Diff. between (1) and (3) Parenting Payment (single and couple) Sickness Allowance Widow’s Allowance AUSTUDY/ABSTUDY NewStart Allowance Mature Age Allowance Youth Allowance Special Benefit Partner Allowance Age Pension Disability Support Pension Wife’s Pension Widow B Pension Carer’s Payment 5 327.0 95.9 330.2 255.6 4 918.3 353.1 2 121.6 113.8 717.1 15 571.8 5 837.4 680.0 75.3 478.3 5 037.5 181.7 4.9 947.1 4 268.1 221.2 2 475.7 1 910.2 1 493.2 15 865.0 5 133.8 792.9 398.2 274.1 4 454.7 193.6 3.5 1 000.6 4 562.9 244.3 2 463.7 2 036.1 1 492.4 17 401.7 5 610.6 729.9 360.5 272.4 289.5 − 85.8 325.3 − 691.5 650.2 131.9 −354.1 −1 796.4 −776.1 −293.2 703.6 −112.9 −322.9 204.2 872.3 − 97.7 326.7 −745.0 355.4 108.8 −342.1 −1 922.3 −775.3 −1 829.9 226.8 − 49.9 −285.2 205.9 Total 36 875.4 39 003.6 40 826.9 −2 128.2 −3 951.5 Note: 1, The source for column 1 is FaCS (2001). amount. Similarly, adding the NewStart Allowance and the Partner Allowance seems to smooth out differences between actual and simulated amounts. There remain AUSTUDY and Special Benefit, both of which are overestimated in MITTS. The Special Benefit has strict requirements, which cannot easily be tested in MITTS because not all necessary information is available in the SIHC. For AUSTUDY, the recipient needs to © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 2006 ACCOUNTING FOR POPULATION AGEING 25 undertake a qualifying study and again this information is not available in the SIHC. From the lack of improvement – indeed deterioration – in simulated expenditures after reweighting, the conclusion is drawn that reweighting the base data for simulations of policy in the current time period cannot be recommended.11 IV. Population Ageing The previous section showed that reweighting the base sample for the current time period is unlikely to improve the outcome of simulations. This section explores the use of MITTS in combination with reweighting to examine the implications of population ageing. Projected population distributions by age and gender for 2050 from the ABS (2003) are used to reweight the population in the 2000/01 SIHC. However, to avoid the effects of changes in population size, it is assumed that the total population size does not change: only the proportion in each subgroup is used. The calibration conditions in the reweighting exercise then consist of the reallocated population totals by age and gender. Three series of projections for 2050 are presented in ABS (2003). Series B results in a medium-sized stable population, based on a fertility rate of 1.6 babies per woman, a net overseas migration of 100,000 persons and a life expectancy at birth of 84.2 for men and 87.7 for women. Series A presents a larger population based on a fertility rate of 1.8 babies per woman, a net overseas migration of 125,000 persons and a life expectancy at birth of 92.2 for men and 95.0 for women. Series C presents a declining population size based on a fertility rate of 1.4 babies per woman, a net overseas migration of 70,000 persons and a life expectancy at birth of 84.2 for men and 87.7 for women. The distributions across age-gender groups are also different for the three scenarios. Figure 2 presents the age-gender distribution in the three population projection series. It shows that the proportion of young individuals is lowest in series C and highest in series A. The proportion of older individuals is highest for series C and lowest for series B.12 Finally, series B has the largest proportion of the population in the working age category.13 Therefore, series A, B and C are referred to in this paper respectively as the young, medium and old population projections. Figure 2 shows that the relative proportion of older persons versus younger persons is expected to change between 2001 and 2050. Due to different assumptions, there are some differences between the three projections provided by the ABS, but generally the three alternatives are relatively close to each other, especially when compared with the current situation. All three alternatives are based on plausible assumptions about the fertility and mortality rates. To provide an insight into the effect of this changed population composition, the SIHC can be reweighted to reflect the composition of the 2050 Australian population before running a simulation. By using the three alternatives, the sensitivity of changes in government expenditure and revenue to the alternative population scenarios can be analysed. Given the current fertility and mortality rates, it is clear that the current population composition must change. Therefore, a simulation based on the current population structure is not satisfactory. 11 A wide range of alternatives, including the imposition of take-up conditions and basing the calibration on numbers calculated by MITTS, based on entitlement according to reported characteristics, are discussed in Cai, Creedy and Kalb (2004). 12 The proportion of the population aged 65 and over for series A, B and C is 28.03, 26.94 and 29.45 per cent, respectively. 13 The proportion of the working age population for series A, B and C is 56.71, 59.00 and 58.49 per cent, respectively. © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 26 AUSTRALIAN ECONOMIC PAPERS Figure 2. MARCH Gender and age distributions of current and projected populations In all three projections, the proportion of older persons has increased relative to the younger age groups. Assuming that people’s behaviour remains similar to current behaviour of comparable individuals, the effect on expenditure and revenue can be simulated. Under the same assumption, behavioural responses to policy changes can be simulated as well. The reweighting procedure is discussed in Subsection IV.a, followed by the microsimulation results using the alternative weights in Subsection IV.b. a) Reweighting procedure As mentioned above, the calibration conditions are constructed from the population projections for 2050 by the ABS (2003). Figure 2 also presents the age-gender distribution of the 2000– 2001 SIHC sample.14 From the graph, it is clear that there is a decrease in the younger age groups (up to about 54 years) and an increase in the proportion of older Australians. The patterns are similar for men and women. Only the proportion of older women is slightly higher than the proportion of older men for all current and projected populations. This is no surprise, given the longer life expectancy of women. The proportions presented in Figure 2 are used to calculate revised weights based on the original ABS weights. Given the low impact of the reweighting discussed in the previous section, the reweighting here is based only on the updated age and gender distribution. Figure 3 presents the distribution of the ratio of the new weights resulting from this procedure relative to the ABS weights. As anticipated, the weights in this section deviate more from the ABS weights than the earlier revised weights (which had a range for the ratio of new to old weights of 0.68 to 1.87). The substantial changes in the age structure of the population require some age groups to be weighted up and others to be weighted down. The minimum range that could be imposed in the D-S approach increased to [0.51, 3.25] for the 2050 reweighting using the medium population projection (series B). The upper boundary seems relatively more affected by the difference in age structure, which can be explained by the relatively sharp increase needed for the older age group compared with the smaller decrease of the other groups which is spread across a larger age range. Comparing the restrictions on the bounds that can be achieved in the different scenarios shows that the range is narrowest for the medium population scenario. This may be explained 14 Up to age 14, only age can be observed in the SIHC; gender is not available for this group. © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006. 2006 Figure 3. ACCOUNTING FOR POPULATION AGEING 27 Ratio of new weight for 2050 population structures and ABS weight by the fact that although series B has a higher proportion of people over 50 than the young population, the proportion of people over 75 is lower. This means that this relatively small group of people over 75 needs to have a larger increase in their weights in series A and C. The effects on wage and salary income distributions of reweighting are shown in Figure 4 for the three population series. This figure shows for each income level the reweighted frequency minus the initial frequency. In each case there is little change in the proportion of persons on very high wages, but the proportion on medium wage and salary incomes, in particular, has decreased. This has mostly gone to an increase in the proportion of people who have no wage and salary income. This is not shown in Figure 4, but the differences at zero income for series A to C respectively are 13.17, 12.08 and 13.55. The income from wage and salary distribution is further from the 2001 distribution in series A and C compared with B. A possible explanation for this is that the younger population has a larger proportion of children whereas the older population has a larger proportion of potentially retired people. The medium-aged population, B, on the other hand, has the highest proportion in the working-age category, resulting in a larger proportion of the population on non-zero wage and salary income. b) Population ageing, taxes and expenditure This section examines the effect of population ageing on government expenditure and revenue, if the changed demographic structure of the population were realised in 2001. Table IV presents the results for the medium population projection (series B). As expected, a larger proportion of people pay income tax, as there are fewer children and dependent adolescents, but at a lower level given the lower income of retirees. This results in a decrease in the revenue from taxation. Similarly the Medicare levy decreases and rebates increase. On the expenditure side, the number of people on pensions increases substantially, while the number of people on allowances and family payments decreases. The Age Pension sees the largest increase, in line with the ageing population and a smaller increase is observed for the Disability Support Pension, which also tends to be received by older individuals. When eligibility for the Age Pension is based on observed receipt in the SIHC (to account for the lack of information on assets), the expenditure on the Age Pension becomes smaller, as shown in the last two columns in Table IV. However, the relative increase in the expenditure due to population ageing, when comparing to the expenditure in 2001 based on observed eligibility for Age Pension (not presented in Table © Blackwell Publishing Ltd / University of Adelaide and Flinders University 2006.