Welfare Benefits In Kind and Income Distribution

Fiscal Studies (1993) vol. 14, no. 1, pp. 57-76 Welfare Benefits In Kind and Income Distribution MARIA EVANDROU,* JANE FALKINGHAM, JOHN HILLS and JULIAN LE GRAND I. INTRODUCTION This article explores the value to households in different income groups of benefits from public spending on education, the National Health Service and subsidies to local authority housing. Its results are drawn from secondary analysis of the 1987 General Household Survey (GHS). The paper compares these findings with those which the Central Statistical Office (CSO, 1990) derived from the 1987 Family Expenditure Survey (FES). As well as using the more detailed information given by the GHS on use of health services and higher education than by the FES, we also apply some different methodological approaches from the CSO, including the allocation of higher education for students living away from home to their households of origin and the use of estimates of economic housing subsidies. The CSO s results are summarised in Section II, together with a discussion of some limitations of its approach, and the advantages (in some respects) of using GHS data. We present our main findings of distribution by income group in Section III. A more detailed discussion of the results for the separate areas of * Welfare State Programme, STICERD, London School of Economics; Department of Epidemiology and Public Health, University College London. Welfare State Programme. Welfare State Programme; School of Advanced Urban Studies, University of Bristol. The authors are grateful for comments from participants in the 1991 annual conference of the Social Policy Association, and from Tony Atkinson, Karen Gardiner, Stephen Jenkins, Judith Payne, Carol Propper, David Winter and two anonymous referees. They are also grateful for support and help from Jane Dickson and Michelle Eyles, and from the Central Statistical Office, the ESRC Data Archive and the Office of Population Censuses and Surveys. The paper forms part of the Welfare Research Programme at the London School of Economics, funded by the Economic and Social Research Council (under Programme Grant X206 32 2001). Institute for Fiscal Studies, 2000

Fiscal Studies education, the National Health Service and housing subsidies follows in Section IV. We summarise our results in Section V and make some suggestions for future work in this area. For a more detailed account of the findings, including an analysis of distribution by socio-economic group, see Evandrou et al. (1992). II. THE CSO ESTIMATES AND METHODOLOGY Each year the CSO publishes the results of an analysis of the distributional effects of taxes and benefits (in both cash and kind) in Economic Trends. (The most recent, for 1989, is in CSO (1992).) The results show the estimated value of various benefits and taxes for households in different quintile groups (fifths) of the income distribution, ranked by equivalised disposable income, that is, income after allowing for cash benefits and direct (but not indirect) taxes, with the incomes adjusted to allow for the greater needs of larger households. 1 The CSO s analysis assumes as does ours that the incidence of spending on education, the NHS and housing subsidies is on the households directly receiving each service. While this seems plausible, it is possible that the true incidence would show a different pattern for instance, if employers are able to pay lower wages than they would have had to in the absence of state provision. The valuation put on benefits in kind provided by the public sector is their cost. In reality, recipients might put a different value on them (and these differences may vary with income level). The results in CSO (1990) show that original income (from private sources) is very unequally distributed, with the top fifth of households (ranked by disposable income) receiving more than 20 times as much as the bottom fifth in 1987. Cash benefits are worth more to households at the bottom, so that for gross income (original plus cash benefits) the corresponding ratio between the two ends of the distribution is reduced to six to one. The CSO figures suggest that, taken together, all taxes (both direct and indirect) had, by 1987, become roughly proportional. Post-tax income is thus distributed in much the same way as gross income. The final part of the CSO s analysis allows for benefits in kind, the focus of this paper. It demonstrates the combined effect of spending on public sector education, the National Health Service, housing subsidies, 2 rail and bus subsidies, school meals and welfare milk. The CSO suggests that, taken together, these benefits are worth significantly more for those with lower than for those 1 The equivalence scale used by the CSO gives a value of 0.61 for a single person and 1.0 for a married couple, with additions for children depending on their ages for instance, 0.21 for a child aged 5 7 and 0.36 for one aged 16 18. Thus a single person with a disposable income of 6,100 would be placed at the same point in the income distribution as a married couple with an income of 10,000, or a married couple with children aged 7 and 16 and an income of 15,700. 2 General subsidies to council and housing association tenants mortgage interest tax relief is taken account of in the treatment of taxation, and housing benefit in cash benefits. 58

Benefits in Kind with higher incomes, for instance being worth almost twice as much for the bottom fifth as for the top fifth. The result is that final income (including these benefits in kind) is less unequal than post-tax income, with the ratio of top fifth to bottom fifth reduced to 3.7 to one. In what follows, we describe such a distribution with greater absolute values for those with the lowest incomes as being pro-poor, while we describe distributions with greater absolute value for those with the highest incomes as pro-rich. Note that this means that benefits can be pro-rich in this sense, but still progressive in that they represent a greater proportion of income for the poor than for the rich. If the CSO s estimates are correct, benefits in kind play a very important part in the living standards of households with low incomes. The CSO estimates that the bottom fifth of households receive only 6.9 per cent of post-tax income, but 9.9 per cent of final income. Conversely, while the top fifth receive 41 per cent of post-tax income, their share of final income is reduced to 36 per cent. The CSO estimates suggest that in-kind benefits are equivalent to over 70 per cent of post-tax income for the poorest fifth, but only 7 per cent for the richest fifth. The size of these differences makes investigation of the robustness of the estimates of considerable interest. Table 1 shows the CSO s estimates of the distribution of those benefits in kind. The first three columns, showing the services with which we are concerned here, are totalled in the fourth column. The final column shows the estimates including rail and bus subsidies, school meals and welfare milk, which we do not investigate below. The first point to note is that the overall pro-poor distribution (apart from a slight rise between second and third quintile groups) results from the combination of a more markedly pro-poor distribution within non-retired households (the results given are by successive fifths of non-retired households) and a roughly flat (and less valuable on average) pattern of benefits to retired households. The second is the contrast between the different services. Education is shown as benefiting the poorer non-retired households in particular, but, for obvious reasons, not retired households. As students away from home are excluded, the distribution mainly reflects the greater proportion of families with school-age children in the lower part of the non-retired income distribution, the presence of children being one of the factors which lowers their equivalent incomes. By contrast, health benefits are estimated to be more valuable to retired than to nonretired households. In both cases there is little gradient with income: the overall pro-poor distribution results from the greater value to retired households, which tend to be lower down the combined distribution. Housing subsidies are much smaller in scale and are shown as being generally pro-poor, and particularly so within the non-retired population. The overall pro-poor distribution thus results from a pro-poor distribution of each individual service. 59

Fiscal Studies Quintile group (by equivalent disposable income) TABLE 1 CSO Estimates of Benefits in Kind Education National Health Service Housing subsidies Total Total (with other a ) All households Bottom 796 1,029 99 1,924 2,045 2 656 976 105 1,737 1,815 3 837 922 56 1,815 1,890 4 628 790 32 1,450 1,520 Top 360 633 15 1,008 1,101 All 656 870 61 1,587 1,624 Non-retired households b Bottom 1,510 880 130 2,520 2,680 2 1,040 860 70 1,970 2,040 3 910 820 40 1,770 1,850 4 620 720 30 1,370 1,450 Top 330 590 10 930 1,040 All 880 770 60 1,710 1,810 Retired households b Bottom 30 1,170 40 1,240 1,290 2 1,190 100 1,290 1,350 3 10 1,090 120 1,220 1,280 4 20 1,160 90 1,270 1,330 Top 20 1,090 20 1,130 1,200 All 20 1,140 80 1,240 1,300 a Other includes rail and bus subsidies, school meals and welfare milk. b The separate results for non-retired households are for quintile groups within each of the respective populations. Source: CSO, 1990, Tables 3, H and L. There are several reasons why alternative estimates might differ from the CSO s. The first is to do with the data source. The FES does not contain information on the use of the NHS. In order to allocate health spending to households, the CSO assumes that usage depends on the age and sex of household members, and allocates total spending accordingly. Yet usage may also depend on other factors, which may include or be correlated with income, which could mean that the distribution of health benefits in Table 1 was incorrect. The GHS data used below do contain direct information on the use of health services, as well as education and housing. The other reasons relate to methodology. First, in estimating benefits from public spending on education, the CSO excludes spending on students living away from home. This is not only an important part of education spending, but it 60

Benefits in Kind is also one which is unequally distributed across households. It might be argued that students are poor (as they have low current incomes) and so spending on them should be allocated to the lowest income groups. In our view this would be misleading. While students may have low current incomes, there is much evidence to suggest that they have relatively high lifetime incomes. Taking a longer-term perspective, whatever its indirect benefits for the economy and society as a whole, the direct benefit of spending on students more realistically goes to higher income groups. One solution to this problem (which we are examining in related research) is to look at all of these questions on a lifetime basis, allocating education to those directly receiving it, but looking at distribution in terms of lifetime incomes. Another, which is what we do in this article, is to examine the effect of allocating spending on the education of students living away from home to their parents. Their parents position in the income distribution may well be a better reflection of the student s own longterm position; also parents may be at least partial beneficiaries, in the sense that they would have paid for tertiary education for their children if the state had not. 3 A different methodological problem relates to the estimates of the value of housing subsidies. The CSO takes the total annual flow of recurrent subsidies to local authority housing and of (mainly capital) grants to housing associations, divided between Greater London, the whole of the rest of England, Wales, Scotland and Northern Ireland. It then divides the total between the council and association tenants in each of these areas in proportion to the gross (rateable) values of the properties they occupy. There are several reasons for being sceptical about such results. The most important is that the cash-flow subsidies into local authority Housing Revenue Accounts are a poor guide to the value of the difference between actual gross rents and the economic value of the accommodation occupied (see Hills (1991) for a detailed discussion of the problems). Additionally, there is scope for refinement in the allocation of total subsidy, using a narrower breakdown between different English regions. III. ESTIMATES BASED ON THE GENERAL HOUSEHOLD SURVEY Comparison of Sample Characteristics An immediate important difference between GHS- and FES-based results is that the latter are for the whole of the UK, whereas the former are for Great Britain. There are also some differences in the sampling methodology used and data 3 Note that although we have allocated benefits to non-resident students to their households of origin, we have not as we strictly should have done allowed for such non-resident members in calculating equivalent incomes. This means that our results may tend to position such households higher in the income distribution than they should be. 61

Fiscal Studies collected by the two surveys. As a result of these, there is a slight difference between the income definitions used to rank households in the income distribution. The CSO uses equivalent disposable income, that is, income after including cash benefits and deducting direct taxes including rates. Our ranking is based on equivalent net income, which differs in that rate payments are not deducted from it (nor rate rebates included). The CSO s measure of disposable income averages 10,000 per household (3,800 for the bottom quintile group). Our measure of net income averages 9,000 per household (2,800 for the bottom quintile group). 4 We use the same equivalence scale as the CSO to adjust the ranking for household size (for this and further details see Evandrou et al. (1992, Appendix 1)). We also follow the CSO in using households as the unit of analysis (an approach which has the drawback that large households are implicitly given the same weight as small ones). TABLE2 Composition of Quintile Groups, 1987 Percentage of group in each category QUINTILE GROUP OF HOUSEHOLDS (by equivalent disposable or net income) ALL HOUSEHOLDS Bottom 2 3 4 Top CSO analysis (UK, FES-based) 26 Retired 48 46 18 11 8 Non-retired: One adult 7 8 9 12 18 11 Two adults 8 9 18 25 40 20 One adult with child(ren) 9 6 2 1 4 Two adults with child(ren) 19 21 33 29 20 24 Three or more adults 8 10 20 22 14 15 Our analysis (GB, GHS-based) Retired 52 5 1 23 1 2 10 29 Non-retired: One adult 11 6 8 11 19 11 Two adults 7 10 16 27 38 20 One adult with child(ren) 10 4 2 1 1 4 Two adults with child(ren) 16 18 33 28 18 23 Three or more adults 4 11 18 21 16 14 Sources: CSO, 1990, Table B; own analysis of 1987 GHS. 4 16 The differences in incomes in the two samples are significant. The treatment of rates should raise net income above disposable income. Part of the difference results from the greater preponderance of pensioners in the GHS sample. Part may reflect closer attention to the detail of income in the FES. However, it is the ranking of households which matters here, not absolute income levels, so for these purposes the problem is not so great. 62

Benefits in Kind Despite these differences, Tables 2 and 3 show a high degree of consistency between the composition of the different quintile groups of all households in the CSO s analysis and in ours. Looking at Table 2, we have more retired households 5 than the CSO. In the poorest quintile group we find rather more oneadult non-retired households and rather fewer households with children. As Table 3 shows, the main effect of this is that our bottom quintile group contains somewhat fewer adults than the CSO s bottom group. Overall, however, the make-up of our quintile groups is very close to the CSO s. As a result, there is no reason to expect that there would be significant differences in results stemming from demographic differences between the samples or the ways in which they have been divided into income quintile groups. TABLE3 Summary Demographic Characteristics of Quintile Groups, 1987 Average numbers per household QUINTILE GROUP OF HOUSEHOLDS (by equivalent disposable or net income) ALL HOUSEHOLDS Bottom 2 3 4 Top CSO analysis (UK, FES-based) Children 0.7 0.6 0.8 0.6 0.4 0.6 Adults 1.7 1.7 2.1 2.1 2.0 1.9 Persons 2.4 2.3 2.9 2.7 2.3 2.5 People in full-time education 0.6 0.5 0.6 0.5 0.3 0.5 Economically active people 0.3 0.7 1.4 1.7 1.7 1.2 Retired people 0.7 0.7 0.4 0.2 0.2 0.4 Our analysis (GB, GHS-based) Children 0.6 0.6 0.8 0.6 0.4 0.6 Adults 1.5 1.8 2.0 2.1 2.0 1.9 Persons 2.1 2.3 2.8 2.7 2.3 2.4 People in full-time education 0.4 0.4 0.6 0.4 0.3 0.4 Economically active people 0.4 0.7 1.3 1.7 1.7 1.2 Retired people 0.7 0.8 0.5 0.2 0.2 0.5 Sources: CSO, 1990,Table A; own analysis of 1987 GHS. 2. Demographic Differences between Income Groups The differences in demographic composition between the different income groups are part of the reason for the different patterns of receipt of benefits in kind which we describe below. As can be seen from Table 3, in both sets of figures the households in quintile groups 3 and 4 contain more people than the others. They also contain a larger number of children, particularly those in the 5 Defined in both cases as those where the combined usual gross weekly income of retired members amounts to half or more of that of the household as a whole. 63

Fiscal Studies third quintile group. In our results the bottom group has the lowest average number of people per household while the top group has the lowest number of children per household. This trend changes somewhat if one distinguishes between retired and nonretired households. The largest households with the greatest number of children are to be found in the middle of the overall income distribution. However, as Table 2 shows, more than half of the households in the bottom two groups are retired. If one looks at non-retired households only, the number of children declines as one moves up the income distribution. These differences in the number of children per household affect the numbers in full-time education per household (shown in Table 3) and the distribution of education spending discussed below. 3. Summary of Results Based on the GHS Against this background, our main results are summarised in Table 4. The patterns in each service area, including the distinction between low and high estimates for the value of housing subsidies, are discussed further in section IV below. Our estimated totals for all households suggest no clear trend in absolute value of benefits in kind with income level. Benefits from the three services combined are most valuable for the middle quintile group, and least valuable for the top quintile group. By comparison with the CSO estimates in Table 1, the average level of benefits allocated is much the same (especially if the high estimate of housing subsidies is used), but the distribution is less markedly pro-poor. The reason for the difference can be seen to lie in the combination of benefits to non-retired households, which are pro-poor (but not to as great a degree as in Table 1) and benefits to retired households, which show a flatter pattern and which have a much lower average value. Looking at the individual service areas, the greatest contrast is in the distribution of education benefits. In contrast to the CSO estimates, which suggest that the value of education (excluding students away from home) for the top quintile group is not much more than half the average, our estimates suggest that the top quintile group receives more than the average, and more than the bottom two quintile groups. Both sets of results allocate the greatest receipts to the middle quintile group. The key difference comes from the greater value of benefits we allocate to the top two quintile groups of non-retired households, and the lower (but still above average) receipts which we estimate for the bottom non-retired group. The shape of the distribution of health benefits for all households is very similar to that found by the CSO, but on a smaller scale. Our estimates only account for about 75 per cent of NHS spending, and allocate much less to retired households than do those of the CSO. As far as non-retired households are 64

Benefits in Kind concerned, we find an even more clearly pro-poor distribution than the CSO. However, this is more than offset by the way in which our estimates suggest that health benefits for retired households half of all the households in the bottom two groups of the overall distribution are less than 50 per cent of the size estimated by the CSO. The reasons for this are discussed below. TABLE 4 Estimated Distribution of Benefits in Kind per year; GB, 1987; GHS-based Quintile group (by equivalised net income) Education National Housing Subsidies Total Health Service Low High Low High All households Bottom 550 780 130 350 1,460 1,670 2 580 780 80 220 1,430 1,570 3 820 810 70 170 1,700 1,800 4 780 680 30 100 1,500 1,560 Top 720 480 20 40 1,230 1,240 All 690 700 70 180 1,460 1,570 Non-retired households Bottom 1,120 1,070 110 310 2,300 2,510 2 1,100 900 60 170 2,050 2,170 3 970 840 50 130 1,870 1,940 4 870 650 40 80 1,550 1,600 Top 780 450 20 40 1,250 1,270 All 970 780 60 150 1,800 1,900 Retired households Bottom 20 470 130 330 610 820 2 20 510 130 310 660 840 3 30 520 100 250 640 790 4 30 630 90 210 750 870 Top 30 440 40 100 510 570 All 30 510 100 240 640 780 Source: Own analysis of 1987 GHS. Our low estimates of housing subsidies average much the same as the CSO s, but our high estimates are much greater. (Note that our estimates are only for council tenants, not those of housing associations.) In either case, our estimates are markedly more pro-poor than those of the CSO, resulting in particular from a greater and more pro-poor estimate of subsidies for non-retired households. Because housing subsidies are so clearly pro-poor, the overall shape 65

Fiscal Studies of the distribution of in-kind benefits shown in the final column of Table 4 is affected by the choice of low or high estimates of them. With the high estimate, the bulge in the middle of the distribution is less marked and the relative position of the top group is worse. IV. BENEFITS IN KIND BY SERVICE In this section we examine the estimates outlined above in more detail, disaggregating the totals for each service, and looking in more detail at the differences between retired and non-retired households. 1. Education As noted above, the main difference between our estimates and the CSO s comes from the distribution of education benefits. Across all households, we have shown that the greatest education benefits go to the middle of the income distribution and the benefits for the top of the distribution are higher than those for the bottom. Table 5 shows that this is the result of a combination of factors. The bottom two quintile groups of all households receive very little in the way of education benefits beyond compulsory schooling. Meanwhile the top group receives, on average, conspicuously less from state schooling than the other groups, but substantially more from further and higher education. The top group of all households receives nearly five times as much from tertiary education as the bottom group. This picture is slightly muddied by the fact that the vast majority of education benefits are received by non-retired households. Thus the lower average receipts of the bottom two quintiles are in part due to the fact that they contain a large number of retired households who are receiving no benefits in kind from education. Focusing on non-retired households only, it can be seen from Table 5 that total education benefits are pro-poor, with successive quintile groups receiving less than the group below. However, benefits from schooling and those from tertiary education operate in opposite directions. The main reason for the gradient in benefits received from schooling is the average number of children per household in each group. Households in the bottom quintile of non-retired households contain an average of 1.2 children compared with only 0.4 in the top one. This demographic effect is compounded by differential use of private schools, overwhelmingly concentrated at the top of the distribution. 6 Of households in the bottom two income groups, 44 per cent report some receipt of benefit from primary and/or secondary schools, in contrast to only 18 per cent of households in the top quintile group. 6 Differential use of private schools by quintile group could not be allowed for directly from GHS data, so we imputed use of private schools on the basis of FES data for different income groups (see Evandrou et al. (1992, Appendix 1)). 66

Benefits in Kind TABLE 5 Education Benefits in Kind, by Quintile Group per year; GB, 1987; GHS-based Quintile group Primary and Tertiary Total education (by equivalised net income) secondary schools education All households Bottom 460 90 550 2 440 140 580 3 640 180 820 4 470 310 780 Top 290 430 720 All 460 230 690 Non-retired households Bottom 930 190 1,120 2 860 240 1,100 3 720 250 970 4 440 440 870 Top 300 480 780 All 650 320 970 Source: Own analysis of 1987 GHS. The amount of benefit per recipient also varies across income groups: households in the bottom group that do benefit from state schooling gain on average 2,100 compared with 1,650 for beneficiary households in the top fifth (see Evandrou et al. (1992, Table A2)). Benefits in kind from schooling are thus pro-poor even when one looks only at recipient households. For those non-retired households benefiting from tertiary education, there appears to be little difference in the average value (4,150 for the bottom group and 4,020 for the top), but benefits from tertiary education are disproportionately received by households in the top two quintile groups. The main reason for this is the inclusion in our analysis of benefits in kind from higher education, in particular university education, accruing to non-resident students (further discussed below). Only 5 per cent of the bottom quintile group report receipt of any benefit in kind from further education, compared with 12 per cent of the top two groups. Given the importance in our estimates of benefits from tertiary education in moderating the pro-poor influence of educational benefits, it is useful to compare the effect of our methodology with that used by the CSO. The impact of the different techniques is illustrated in Figure 1. The figure shows our estimates, separating out the average values of tertiary education excluding non-resident students (that is, using the CSO methodology, but not presenting the actual CSO results). 67

Fiscal Studies The procedure employed for calculating receipt of benefits from state schools by the CSO was virtually identical to ours, except that the CSO had direct access to private sector utilisation rates from the FES whereas we imported weights from that data set into our analysis. However, the impact of this on average benefits in kind received from schooling by quintile group should be minimal. The CSO then allocates benefits from further and higher education to adults in the households which reported use of them. Again this procedure was replicated in our analysis, as was the exclusion of student-only households. Beyond this, our approach deviated from the CSO s. We additionally assigned benefits accrued by non-resident students to their household of origin. The rationale behind this is explained in Section II above. The effect of this on the total distribution is shown by the upper shaded area in Figure 1. The figure shows that the addition of benefits to non-resident students has little impact on the gradient up to the third quintile group. However, for those households higher up the income distribution, the effect of including benefits to non-resident students is substantial. Only 0.5 per cent of the bottom two quintile groups were allocated benefits for non-resident students compared with 3.3 per cent of the top quintile group. In the top group of all households, the average benefit due to non-resident students is actually greater than benefits from tertiary education accruing to residents within the household. This is not surprising, given what is known about differential access to higher education. Furthermore, residents tend to be studying on part-time courses or courses at colleges of 68

Benefits in Kind further education whilst non-resident students are primarily studying at (what were) polytechnics, universities or other higher education establishments, the costs of which are much higher. This is reflected in the average value of benefits received per recipient an average of 2,430 for households in the top quintile group with resident students compared with 6,800 for those with non-resident students. Putting both of these factors together, the average value of higher education benefits for non-resident students was 10 times as great for those in the top quintile group as for those in the bottom quintile group. Table 6 compares our results including and excluding non-resident students with the CSO s actual estimates, concentrating on non-retired households only. It can be seen that the difference in estimates of education benefits in kind is not only due to differences in methodology. Although the average benefit across all non-retired households if one excludes non-resident students is much the same in both sets of results (i.e. using the CSO s methodology), the distribution across quintile groups is very different. Our results are significantly less pro-poor than the CSO s, even before taking benefits accruing to non-resident students into account. The reasons for this difference are not clear as the demographic composition of both sets of quintile groups is very similar (see Tables 2 and 3). Education benefits are primarily affected by the number of children in each group, which are similar in both analyses. It could be that their age composition is different, but it is more likely that the difference results from the application of different cost data. TABLE 6 Education Benefits: Effects of Methodological Differences Non-retired households; per weeks; 1987 Quintile group (by equivalised net or disposable income) GHS-based estimates (GB) Including non-resident students Excluding nonresident students CSO estimates (UK) (excluding on-resident students) Bottom 1,120 1,090 1,510 2 1,100 1,070 1,040 3 970 900 910 4 870 760 620 Top 780 490 330 All 970 860 880 Sources: Own analysis of 1987 GHS; CSO (1190, Table 3). 69

Fiscal Studies 2. National Health Service Overall the NHS results by income group in Table 4 show a distribution that is pro-poor, the lowest quintile group receiving 62 per cent more than the highest. Interestingly, though, the pattern is not smooth, with the lowest two groups receiving about the same, but the third (middle) group receiving more than any other. Indeed, were it not for the relatively low figure for the top group, the distribution would be better described as equal or unpatterned. A distributional pattern that favours the poor overall is not surprising since the poor tend to report more ill health than the better off. One would also expect the higher groups (and perhaps particularly the highest) to make more use of private care. 7 Moreover, the pattern is consistent with other work using previous years of the GHS (Hurst, 1985; O Donnell and Propper, 1991). 8 However, our pattern is not quite consistent with that suggested by the CSO. As noted above, the CSO does not allocate NHS expenditures by actual usage. Instead it is estimated according to the estimated average use made of various types of health service by people of the same age and sex and according to the total cost of providing those services. 9 As with our estimates, the CSO s show a pro-poor distribution, with the lowest quintile receiving 63 per cent more than the top (see Table 1), but its gradient is smoother and the scale larger. Table 7 gives a breakdown of the distribution by income group for different sectors within the NHS. It is apparent that much of the pattern of the aggregate distribution derives from the distributions of the two most expensive services GP consultations with a prescription and in-patient stays, especially the latter. For all households these both show a similar pattern, with the bottom three quintile groups receiving about the same, and the top group receiving the least. The distributions of expenditure on out-patients and on GP consultations without a prescription show little clear pattern. The predominance of in-patient stays, both in terms of overall expenditure and in determining the aggregate distribution, is not surprising, but is none the less a pity; for it is here that GHS data are weakest. The data refer only to number of spells as an in-patient, not to length of stay (which may well differ by income group). Also, for obvious reasons, the GHS does not interview people in hospital and hence it misses a proportion of users (which again may differ by income group). 7 It should be noted that the tax expenditures associated with private health care are not included in this part of the analysis. 8 There is an important (and controversial) question as to whether the pro-poor distribution of NHS spending matches the distribution of ill health. This is not the concern of this paper; for a review of the debate, see Le Grand (1991). 9 CSO, 1990, p. 110. The benefits from maternity services are assigned separately to those households containing children under the age of 12 months. 70

Benefits in Kind Further insights may be gained from disaggregating households into retired and non-retired. The distribution for retired households shown in Table 7 is much more equal than that for all households, with the highest quintile group receiving nearly as much as the lowest (both receive less than the middle three groups). This suggests that the possible explanation for the relatively low consumption of the top group overall in terms of private medical care may have some validity, since the chronic conditions that affect elderly people are often ineligible for private insurance coverage. It will be interesting to see whether there is any change in this pattern in later years following the extension of tax relief to private health care insurance payments by elderly people. Quintile group (by equivalised net income) TABLE 7 Health Care Benefits in Kind, by Quintile Group GP consultation with prescription GP consultation without prescription Out-patient visit per year; GB, 1987; GHS-based In-patient stay Total NHS All households Bottom 110 20 70 570 780 2 110 20 110 530 780 3 130 30 90 560 810 4 110 30 90 450 680 Top 70 20 70 310 480 All 110 20 90 480 700 Non-retired households Bottom 150 30 100 790 1,070 2 130 30 100 630 900 3 130 30 100 580 840 4 100 30 90 430 650 Top 70 20 70 290 450 All 120 30 90 550 780 Retired households Bottom 80 10 60 320 470 2 80 20 60 350 510 3 90 10 70 340 520 4 90 10 110 420 630 Top 80 10 90 270 440 All 80 10 80 340 510 Source: Own analysis of 1987 GHS. 71

Fiscal Studies Not surprisingly, given the pattern for retired households, that for non-retired households is much more pro-poor. Interestingly, the decline with income is also smooth. Again, much of the work is being done by in-patient stays and by GP consultations with prescription. However, this disaggregation between retired and non-retired households reveals a problem in comparing our results with those of the CSO. While our estimate of average expenditure for non-retired households is broadly the same as that of the CSO (780 compared to 770), that for retired households is about half (510 compared with 1,140). Moreover, the fact that, within our estimates, the expenditure per retired household is less than that per non-retired household is itself a little surprising, given what we know of the relative morbidity patterns of the two groups of households. However, retired households contain 1.45 people on average, compared with 2.85 for non-retired households. This means that our estimates of NHS spending per household member average 350 for retired households, compared with 270 for non-retired households. This differential in favour of members of retired households is still smaller than one might expect. The reason for our low estimate for the retired is not clear. According to the GHS, the proportion of all retired households making any use of each of the NHS sectors investigated is less than the proportion for the non-retired. In addition, fewer retired users of NHS services reported multiple use. 10 The explanation probably lies in the fact that the GHS does not interview people in institutions, including old people s homes, nursing homes and longstay hospitals. Since these institutions are likely to contain a substantial proportion of retired households, and, moreover, these households are likely to have a higher-than-average use of the NHS, their omission must result in an underestimate of NHS use by the retired. The CSO estimates do not suffer from this problem, since the age utilisation rates they employ in their imputation procedures include utilisation by people in institutions. However, the CSO s procedure of allocating institutional spending to the retired households in the FES sample (which also excludes those in institutions) will overstate the receipt of health benefits by the non-institutional elderly population. Our procedure has the advantage, important in this context, of allowing for variation in use by income group within the non-institutional population. The main cause of the pro-poor distribution which we find within the nonretired population is the much lower use of health services by those in the top quintile group. Most importantly, only 14 per cent of households in the top quintile group reported an in-patient stay in the previous year, compared with 29 10 See Evandrou et al. (1992, Tables 10 and 11 and Appendix 1). Given the way in which health care benefits are valued, the variation in costs per recipient reflects the proportions of recipients who report more than one occasion on which each service is used, not any assessment of differential costs of each individual usage. 72

Benefits in Kind per cent for the bottom group. This was reinforced by somewhat fewer reported multiple in-patient stays (Evandrou et al., 1992, Tables 10 and 11). 3. Housing Subsidies In estimating general housing subsidies, we measured subsidies to local authority tenants as the difference between actual gross rents and those which would be charged if local authority housing departments earned an economic real return on their assets. While this is preferable in principle to the CSO s approach of allocating cash-flow subsidies, it does not lend itself to very precise estimates. This is because it is not entirely clear what number should be taken to represent an economic return, nor how much of this return should be expected to accrue by way of real capital gains on the property owned. Appendix 1 of Evandrou et al. (1992) describes the methodology we used to establish a range between low and high estimates of economic subsidies. Averaged over all households, these give a range of 70 to 180 per year. Table 8 shows that the clear decline in average subsidy as one rises through the income distribution results from two very different trends. On the one hand, there is a steep decline in the proportions of households in each income group which are local authority tenants: from 56 per cent of the bottom quintile group to only 4 per cent of the top one (the downward trend is not quite so steep amongst retired households, but is still dramatic). By contrast, if one looks at recipients (i.e. council tenants) only, average subsidy rises for those in the higher income groups those in the top quintile group receive between 50 per cent and 100 per cent more than those in the bottom group, depending on which estimate is used. 11 In other words, general housing subsidies are worth most on average at the bottom of the income distribution, but this is because more of those at the bottom of the distribution are council tenants. Within the population of tenants, those with higher incomes tend to receive more valuable subsidies. The distribution within the tenant population mainly reflects the small differentials in council rents in 1987 88 between larger and smaller properties and between those in high- and low-cost parts of the country. This latter point is illustrated by Table 9. This shows that the regional averages of our capital value estimates (at unadjusted first quarter of 1988 prices) vary considerably, values in London being 3.5 times those in Scotland. Net rents (after deducting management and maintenance but not depreciation) varied much less, and were actually lower in 1987 88 in Greater London than in any other region (this is because gross rents were only a little higher, but spending was much higher, than in other regions). The result is the regional pattern of average subsidies shown in the third and fourth columns. Our estimates suggest average subsidies in London between 2.6 and 3.5 times the national average (allowing for the actual stock 11 This confirms the pattern found in Hills (1991, Ch. 14), which contains estimates for 1989 90 based on FES data. However, the average size of subsidies found here is rather smaller than the earlier estimates. 73

Fiscal Studies distribution across the country), but between a negative value actual rents exceeding estimated economic rents and 30 per cent of the national average in the North of England. This differential is much greater than would be implied by the CSO s procedure of allocating subsidies in proportion to rateable values. 12 TABLE8 Housing Subsidies: Proportions in Receipt and Average Receipts GB, 1987; GHS-based Quintile group (by equivalised net income) Proportion of group in receipt (%) Subsidy per beneficiary ( per year) Average subsidy for group ( per year) Low High Low High All households Bottom 56 230 610 130 350 2 38 200 570 80 220 3 23 290 730 70 170 4 12 260 830 30 100 Top 4 470 990 20 40 All 27 250 650 70 180 Non-retired households Bottom 54 200 580 110 310 2 28 220 610 60 170 3 17 310 760 50 130 4 9 400 890 40 80 Top 4 490 1,030 20 40 All 23 250 660 60 150 Retired households Bottom 57 220 570 130 330 2 47 270 650 130 310 3 40 240 610 100 250 4 28 330 760 90 210 Top 12 370 820 40 100 All 37 260 650 100 240 Source: Own analysis of 1987 GHS. 12 However, since 1990, rent differentials have grown substantially across the country, so that the current position may well look very different. 74

Benefits in Kind TABLE 9 Housing Subsidies, by Region, 1987 Estimated capital value a ( ) Net rent b ( per year) Estimated subsidy ( per year) Low High North 18,600 501-58 211 Yorkshire and Humberside 19,400 396 68 349 North-West 19,600 410 59 343 East Midlands 22,500 429 108 433 West Midlands 21,900 423 100 415 East Anglia 32,200 349 420 886 Greater London 56,600 320 1,000 1,809 Rest of South-East 46,700 401 714 1,388 South-West 33,600 373 430 916 Wales 19,270 454 6 285 England and Wales 30,000 406 305 736 Scotland 16,400 407-16 221 Great Britain 27,700 406 252 652 Great Britain (reweighted) c 29,100 402 288 707 a Based on hedonic price index derived from BSM survey. At 1988 Ql prices unadjusted either to trend or for difference between owner-occupied and local authority values. b Gross rent minus management and maintenance costs. c Weighted by actual stock distribution at end 1988 (from Department of the Environment (I 990, Table 9.4).). V. SUMMARY AND CONCLUSIONS (1) The Central Statistical Office s estimates, based on FES data, suggest that benefits in kind from education, the National Health Service and housing subsidies are strongly pro-poor, with a total worth 91 per cent more for the bottom than for the top quintile group. As taxation as a whole represents much the same proportion of gross income for all income groups, the net effect of this spending is substantially redistributive (in the sense that if public spending on these services was eliminated and the savings used to reduce all taxes by equal proportions, the poor would be worse off and the rich better off). (2) Our estimates, based on GHS data, suggest a much less clearly pro-poor distribution of benefits in kind, with the total for these services worth most in the middle of the distribution (and education found to be of least value for the bottom two quintile groups). None the less, the total is between 19 and 35 per cent higher for the bottom quintile group than for the top, so that the pattern is still redistributive (in the sense described above). (3) Like the CSO, we find a flat distribution for retired households and a propoor distribution for non-retired households. However, this finding for non- 75

Fiscal Studies retired households results from a more strongly pro-poor distribution of benefits from the NHS than found by the CSO, but less strongly pro-poor benefits from education. (4) For education, the most important reason for the difference between the estimates is our allocation of tertiary education for non-resident students to their household of origin (whereas the CSO omits them). This item appears to be so strongly pro-rich (worth 10 times as much for the top fifth of the income distribution as the bottom fifth) that it cannot be regarded as redistributive (in that its elimination combined with equi-proportional tax cuts would result in lower income groups being better off and higher groups being worse off). (5) For health, our methodology suggests lower use of services at the top of the distribution with, in particular, half as many of the top quintile group of non-retired households reporting an in-patient stay in the last year as of the bottom group. (6) Our estimates of general housing subsidies are larger in scale than those of the CSO (even on our low estimate), and more strongly pro-poor. This pattern results from the much greater proportion of council tenants at the bottom of the income distribution than at the top, partly offset by a pattern of subsidy per tenant which favours tenants with higher incomes. In conclusion, this analysis using GHS data is interesting in its own right and has also suggested some lessons as to how future exercises of this kind (both official and non-official) could be improved: (a) Omission of higher education for students living away from home is a problem, arguably making the distribution appear to be more pro-poor than it really is. (b) Income may be an important factor in receipt of services from the NHS for non-retired households, a factor which is not captured by the CSO s current methodology. (c) The relativity between the benefits from the NHS received by retired and non-retired households has a significant effect on the overall distribution; future studies could probably improve on both our methodology and the CSO s. (d) There are problems with the cash-flow measures of housing subsidies used by the CSO (especially for housing associations). Although using estimates of economic subsidy would be a departure, the CSO already does something similar in estimating the value of owner-occupiers imputed rents. Our estimates also suggest that it is important to distinguish between regions within England, not just between London and the rest of the country. This might also be worthwhile for the other services (if suitable cost data could be found). 76

Benefits in Kind (e) We imported some information from the FES to improve our GHS-based estimates. The process could be used in reverse for FES-based studies, for instance importing data on health service utilisation from the GHS. REFERENCES CSO (1990), The effects of taxes and benefits on household income, 1987, Economic Trends, May, pp. 84 118. (1992), The effects of taxes and benefits on household income, 1989, Economic Trends, January, pp. 127 65. Department of the Environment (1990), Housing and Construction Statistics 1979 1989, London: HMSO. Evandrou, M., Falkingham, J., Hills, J. and Le Grand, J. (1992), The distribution of welfare benefits in kind, London School of Economics, Welfare State Programme, Discussion Paper no. WSP/68. Hills, J. (1991), Unravelling Housing Finance: Subsidies, Benefits and Taxation, Oxford: Clarendon Press. Hurst, J. W. (1985), Financing Health Services in the United States, Canada and Britain, Nuffield/Leverhulme Fellowship Report, London: King Edward s Hospital Fund for London. Le Grand, J. (1991), The distribution of public expenditure on health care revisited, Journal of Health Economics, vol. 10, pp. 239 45. O Donnell, O. and Propper, C. (1991), Equity and the distribution of UK National Health Service resources, Journal of Health Economics, vol. 10, pp. 1 21. 77