Recent trends in Spanish income distribution : a robust picture of falling income inequality

0 Recent Trends in Spanish Income Distribution: A Robust Picture of Falling Income Inequality 1 Josep Oliver-Alonso † Xavier Ramos † José Luis Raymond-Bara ‡ † Departament d’Economia Aplicada ‡ Departament d’Economia i Història Econòmica Universitat Autònoma de Barcelona Abstract: Income distribution in Spain has experienced a substantial improvement towards equalisation during the second half of the seventies and the eighties; a period during which most OECD countries experienced the opposite trend. In spite of the many recent papers on the Spanish income distribution, the period covered by those stops in 1990. The aim of this paper is to extent the analysis to 1996 employing the same methodology and the same data set (ECPF). Our results not only corroborate the (decreasing inequality) trend found by others during the second half of the eighties, but also suggest that this trend extends over the first half of the nineties. We also show that our main conclusions are robust to changes in the equivalence scale, to changes in the definition of income and to potential data contamination. Finally, we analyse some of the causes which may be driving the overall picture of income inequality using two decomposition techniques. From this analyses three variables emerge as the major responsible factors for the observed improvement in the income distribution: education, household composition and socioeconomic situation of the household head. 1 We would like to thank Josep Pérez García for his computing support and to the participants at the Workshop Fighting Poverty and Inequality through Tax-Benefit Reforms: Empirical Approaches for their comments. We acknowledge financial support from the Fundación de las Cajas de Ahorro Confederadas (FUNCAS). 1 Table of Contents 1. Introduction 1 2. Data shortcomings and methodological choices 2 3. Trends in the Spanish income distribution 1985-1996 3 4. Income inequality by population subgroups 12 5. Robustness analyses to changes in the equivalence scale, in the income definition and to trimming 23 5.1. Equivalence scales and changes in inequality trends 23 5.2. Changes in the income definition 27 5.3. Robustness to trimming 29 6. Conclusion 31 References 32 Appendices Appendix 1. Components of the real net household income 34 Appendix 2. Income distribution, 1985-1996 35 Appendix 3. Definition of the variables used in the decomposition analyses of section 4 37 1 1. Introduction Income distribution in Spain has experienced a substantial improvement towards equalisation during the second half of the Seventies and the Eighties;2 a period during which most OECD countries experienced the opposite trend.3 In spite of the many recent papers on the Spanish income distribution, the period covered by those stops in 1990.4 The aim of this paper is to extent the analysis to 1996 employing the same methodology and the same data set for the whole period. That is, we analyse in detail the changes in the Spanish income distribution from 1985 to 1996 using the ECPF. Our results not only corroborate the (decreasing inequality) trend found by others during the second half of the Eighties, but also suggest that this trend extends over the first half of the Nineties. Several factors seem to be driving this trend towards equalisation: (i) a more unequal earnings distribution, which is the main income component;5 (ii) the redistributive rôle of the public sector;6 (iii) the substantial changes in (female) labour participation rates and unemployment rates, whose impact on the income distribution has been relatively small;7 and (iv) demographic changes leading to an ageing population which seems to have had a positive impact on the distribution of income.8 Even though, accounting for the observed level and change of income inequality is not the main purpose of this paper, we complement our descriptive analysis with a study of the distribution of income by population subgrups. From this analysis three variables emerge as the major responsible factors for the observed improvement in the income distribution: education, household composition and socioeconomic classification of the household head. The paper is organised as follows. Section 2 addresses some limitations of our analysis and the methodological choices we have taken to address them. Section 3 analyses the distribution of real net equivalent income —which is our reference definition of income— for the period 1985-1996 and its evolution. Section 4 presents the main results of the decomposition analyses of the income distribution by population subgroups. Section 5 checks the robustness of our main conclusions 2 This is the evidence obtained from the EPF (Encuesta de Presupuestos Familiares, Spanish Family Expenditure Survey) or from the ECPF (Encuesta Continua de Presupuestos Familiares, Spanish Family Expenditure Survey —interviews are held every three months) data. However, studies based on fiscal administrative records (i.e. el panel de declarantes del IRPF) find the opposite trend — see, Castañer, 1991; Lasheras et al., 1993; Melis and Díaz, 1933; Lambert and Ramos, 1997. See Garde et al. (1996) for likely explanations of such difference in trends. 3 See Atkinson et al. (1995), Smeeding and Gottschalk (1995), Gottschalk and Smeeding (1998), Zaidi and de Vos (1998). 4 See, however, Imedio et al. (1997) and Oliver and Raymond (1999) for a notable exception. 5 See, inter alia, Revenga (1991), Díaz and Melis (1993), and Ayala et al. (1996). 6 See Medel et al. (1988), Bandrés (1990, 1993) and Gimeno (1993). 7 See Revenga (1991) and Ayala et al. (1996) for the effect of unemployment changes on income inequality; and Alba and Collado (1999), and Gradín and Otero (1999) for the impact of increasing female labour force participation. 8 The population share and the relative income of households headed by individulas older than 65 years of age have increased over the period. 2 from section 3. In particular, we analyse the sensitivity of our results to changes in the equivalence scale, to changes in the definition of income and to potential data contamination, mostly at both ends of the distribution. Finally, Section 6 provides a summary and some concluding remarks. 2. Data shortcomings and methodological choices The data (ECPF) we use presents three major problems for distributional analyses. First, since income and related questions refer to previous term’s, information on extra pay (e.g. Christmas bonus) which accrue in the last quarter of the year, get reported in the first quarter interviews. Thus, in order to obtain annual earnings we estimate it from income information provided in the second quarter interviews. Second, like most surveys, income and expenditure (though to a lesser extent) data are clearly underreported when compared to National accounts figures. Differences amount to one third of the figures given by the Spanish National accounts, and decrease over the period of study. And third, the aggregate distribution by income source steaming from the ECPF differs substantially from that of the National accounts. Here, the main problem arises from the systematic misreporting of capital and self-employment incomes.9 In other words, once imputed rents are taken out, labour income and pensions are overreported in the ECPF —on average and over the period 1985-1996, labour income represents 57.4% of total income in the ECPF as opposed to 47.1% in the National Accounts, whereas pensions represent 21% in the ECPF and only 16% in the National Accounts; thus an overreporting of 10 and 5 percentage points, respectively— and capital and self-employment incomes are underreported by the same amount —capital incomes represent 1.4% of total income in the ECPF and 7.6% in the National Accounts, whereas selfemployment income represents 14.4% in the ECPF and as much as 23.9% in the National Accounts; that is, an underreporting of 5.5 and 9.5 percentage points, respectively.10 Given this bias towards labour income and pensions, the main results reported in section 3 should be read with caution. Notwithstanding all these data problems, in section 5 we carry out several robustness analyses to check how sensitive our results are to changes in the income definition, to changes in the equivalence scale and to problems of data contamination. These robustness analyses clearly 9 Notwithstanding that, Cowell, Litchfield and Mercader-Prats (1999) suggest that misreporting by the self-employed does not change the overall distributional picture substantially. Besides the afore mentioned problems with capital and self-employment incomes, imputed rents are also overreported in the ECPF as compared to national accounts. 10 Notwithstanding this, one has to bear in mind that, by definition, the sum of ECPF monetary and non monetary incomes does not correspond to the disposable gross income as defined in the National accounts; the two most important factors for such a difference being the fact that ECPF incomes are net of only tax withheld and that ECPF counts transfers amongst families as 3 suggest that the conclusions obtained in section 3 using real net equivalent income are robust to all these problems. In particular, regarding the problems pointed out above, we use a National accounts adjusted income definition and find that results are indeed robust to these data problems.11 Finally, real income is obtained by using a Retail Price Index for each decile and year, which we have estimated from the expenditure structure of each decile. Income is expressed in 1985 prices. 3. Trends in the Spanish income distribution: 1985-1996 In this section we analyse the changes in the income distribution for the period 1985-1996. We use real net equivalent income12 —the equivalence scale being that of the OECD for Spain— and a wide range of graphical as well as analytical tools.13 To visualise the income distribution, we first use Pen’s parade (1971). To construct the parade let the height of individuals (households) correspond to their income, order the individuals according to their income level (height) and let them march in front of you for one hour exactly.14 Then, the income distribution for 1996 in Figure 1, consists of a parade of dwarfs and few giants marching in the last couple of minutes. After the first five minutes, individuals marching in front of you are only 73 cm. tall. To see the first individual who is twice as tall one has to wait another 22 minutes, and after half an hour (median income) the height of the marcher is about a meter and a half (153 cm.). Seven minutes later we see the individual of average height, who is 175 cm. tall. Then, one has to wait another 20 minutes, that is, 3 to the end, to see the first individual whose height is twice the average. Finally, the two tallest people in the 1996 ECPF data set are 13.85 and 21.52 meters tall, respectively. Which are the main differences between this parade and that for 1985? At first sight, one clearly notices that the individuals marching in the 1996 parade are nearly all of them taller than those in the 1985 one.15 That is, from 1985 to 1996, income has increased in real terms for each and every individual. Average household income increased by 39.1% between 1985 and 1996 (from 512,512 to 712,934 pesetas), at an average annual rate of 3.1%. These results imply an unambiguous increase in overall social welfare, according to all individualistic, symmetric, additively separable part of their income —whereas in the National accounts they get cancelled out. 11 See Oliver (1997) for the methodology used in order to build such an income variable. 12 The exact definition of the real net equivalent income is provided in Appendix 1. 13 The structure of this section draws on Jenkins (1994, 1996). 14 Note that the income parade is the cumulative distribution function turned on its side. 15 Only the last 12 individuals are taller in the 1985 parade than in the 1996 one. In other words, the two lines cross at percentile 99.65. However, such a crossing may not be significant for it may be due to data contamination (Cowell y Victoria-Feser, 1998). 4 and increasing social welfare functions —rank dominance condition, Saposnik, 1981.16 We can also say that the proportion of households in poverty has fallen over time, regardless of which common real income level is used to define the poverty line for the two distributions (Atkinson, 1987). Figure 1. Changing income parades, 1985 and 1996 Real Net Equivalent Income Population share 0 .25 .5 .75 1 0 1.0e+06 2.0e+06 3.0e+06 Mean Income, 1996 Mean Income, 1985 1996 1985 Density functions provide another interesting picture of the income distribution, and as compared to the income parades, they allow for a closer look at both ends of the distribution. Which is the shape of the income distribution in 1985? As Figure 1 shows, the frequency at both ends of the estimated distribution is low: there are relatively few households with very low and very high incomes.17 Most households have incomes between 250,000 and one million pesetas. Notice that often relative poverty lines are defined as half the contemporary mean or median (256,256 and 213,380 pesetas, respectively). Hence, small changes in the low income cut-off might imply big changes in the estimated proportion of poor. The highest frequency (mode) is at about 300,000 pesetas. The distribution, however, does not have one peak only: at about 500,000 pesetas there is another 16 That is, for all social welfare functions ( ) ∑ =yUWn 1 1, U’>0 0≥∀y; where n is the number of households, y denotes income, and U(y) is a individualistic utility function of income. Note that Saposnik’s rank dominance condition or first order stochastic dominance does not require any assumption on the inequality aversion of the social welfare function. In other words, it does not impose any restriction on U’’. 17 All density functions in the paper are estimated using a Gaussian kernel and a window width of 10,000 pesetas, unless otherwise specified. Adaptative kernels would, no doubt, improve estimation at low frequency income levels. 5 great concentration of households, though with less frequency. According to this picture, then, the density does not look like being unimodal but rather bimodal.18 Notwithstanding this, such a bimodal frequency distribution might be due to sampling problems: a kernel density with a wider (optimal) window width shows a unimodal distribution function —see Figure 2, panel (b). Finally, the mountain side is steeper and smoother to the left than to the right, and indeed, there is a distinct plateau at around 900,000 pesetas. How has the shape of the income distribution changed after more than a decade?19 When comparing the densities in Figure 2 panel (a), two clear changes stand out: (i) the whole density has shift rightwards, and (ii) the peaks are not as high as in 1985, and now they have the same height. The shift of the distribution to the right corroborates the universal income increase for all households (percentiles). However, the shape has clearly changed. We still have a bimodal distribution —with modes at 475,000 and 600,000 pesetas— but the modes are lower. This fall in income concentration has a negative effect on inequality, which is counterbalanced by the frequency gain in the middle part of the distribution (between 500,000 and 1 million pesetas). As a result, income inequality falls between 1985 and 1996. 18 Such great concentration of incomes at various points of the distribution may suggest a certain degree of polarization. See Esteban and Ray (1994) or Esteban, Gradín and Ray (1997) for a theoretical study on polarization; and Gradín (1997) for an empirical analysis of polarization in Spain. 19 A graphical description of year to year changes are available from the authors on request. 6 Figure 2. Density function of real net equivalent income in Spain, 1985 and 1996 Real Net Equivalent Income 0 500000 1.0e+06 1.5e+06 2.0e+06 2.5e+06 3.0e+06 0 5.0e-07 1.0e-06 1.5e-06 2.0e-06 Panel a: Gaussian kernel, window width of 10.000 grid=200 width=optimal Real Net Equivalent Income 0 500000 1.0e+06 1.5e+06 2.0e+06 2.5e+06 3.0e+06 0 5.0e-07 1.0e-06 1.5e-06 2.0e-06 Panel b: Gaussian kernel, optimal window width 1985 1996 1985 1996 7 Tables 1, 2 and 3 provide additional elements of analysis that corroborate the falling inequality conclusion. Firstly, income as percentage of the median increases for the poorer half of the distribution and decreases for the richer half of the distribution —see Table 1. As the last column shows, the first nine vingtiles (those below the median income) experienced a positive, though monotonically decreasing, change in their relative position, whereas the relative distance of the richest nine vingtiles became smaller over these twelve years. Table 1. Income distribution evolution in Spain. 1985-1996 (I) Median income = 100 and change 1985-1996 in percentage Percentile 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 Change 85/96 5 34 37 41 41 45 45 43 45 43 43 41 41 18,4 10 47 50 52 51 54 53 54 55 54 54 52 52 10,3 15 57 58 60 59 62 60 61 62 62 62 60 60 5,4 20 64 64 66 64 68 67 67 67 67 68 66 66 3,1 25 70 70 71 72 73 72 73 73 73 73 73 72 3,6 30 75 76 77 78 78 77 78 78 79 79 79 78 3,6 35 81 82 83 84 83 83 84 83 84 84 84 84 2,6 40 88 88 89 89 89 90 89 89 89 89 89 89 1,4 45 94 93 94 94 94 94 94 94 94 95 95 95 0,8 50 100 100 100 100 100 100 100 100 100 100 100 100 0,0 55 108 106 107 106 106 106 107 105 105 106 106 106 -1,2 60 116 113 113 113 113 112 113 111 113 113 113 114 -1,9 65 124 120 121 120 121 119 121 118 120 121 120 120 -3,2 70 134 129 131 127 129 127 129 126 129 129 129 129 -3,6 75 145 142 140 138 138 137 140 136 140 139 139 139 -4,4 80 161 155 152 151 150 150 153 149 154 152 151 152 -5,6 85 178 174 169 166 166 166 171 165 169 169 167 169 -5,3 90 208 203 194 189 190 189 197 190 195 193 194 190 -8,7 95 261 253 236 228 236 237 241 231 248 239 239 235 -9,9 Average 120 116 116 114 116 116 116 114 116 116 116 115 -4,6 Secondly, annualised income growth rates decrease monotonically with respect to the income vingtile, so that for the period 1985-1996 the income of the poorest vingtile grows twice as much as that of the penultimate vingtile, and only the last four vingtiles experience growth rates below the average income’s —see Table 2. Finally, Table 3 shows income shares and cumulative income shares for vingtiles of the population. The comparison between 1985 and 1996 figures clearly reveals the income distribution improvement over this period. The income share of the poorest fifth of the population grew nearly two percentage points, whereas the income share held by the richest fourth fell three percentage points. This improvement is best seen by means of the Lorenz curves in Figure 3. The 1985 distribution Lorenz dominates the 1996 distribution. This result corroborates the rank dominance shown by the income parades in Figure 1 and provides us with a powerful conclusion: according to all standard inequality indices,20 income inequality unambiguously decreased between 1985 and 1996. 20 That is, for all symmetric and scale invariant indices satisfying the Pigou-Dalton principle of transfers. See Lambert (1993), section 5.3 for a definition of those properties. 13 head. To account for income inequality and inequality trends we use three different additive decomposition techniques and the inequality measure I0.25 The first one is the well known decomposition by population subgroups put forward by Cowell (1980) and Shorrocks (1980). This decomposition will enable us to account for the effect on inequality levels of each characteristic separately —i.e. not taking into account the effect that other attributes also have on inequality. Let us consider a population of n income units yi (i.e. individuals or households) with mean income µ and variance σ2. Partition the population into k mutually exclusive and exhaustive set of subgroups —e.g. by household composition, age of the household head, educational attainment, etc.— comprising nk members each and mean income µk. Then, I0 can be decomposed into two contributions. The first one —second term in [5]— corresponds to the between groups component and measures the inequality which would obtain had every individual had the mean income of the subgroup she belongs to. The second contribution —first term in [5]— is the within group component and measures a weighted average of subgroup inequalities. [5] ( ) kkkk kkII λ νν 1 00 log ∑ ∑ += In [5], νk(≡nk/n) is the population share of group k and λk(≡µk/µ) denotes the mean income of group k relative to population (or rather sample) mean income. In order to account for inequality change we use Jenkins’ (1995) methodology. The change in I0 can be expressed as: [6] ( ) [ ] ( ) ≈∆−∆−∆+∆=−=∆ ∑ ∑ ∑ ∑ +k kk kkk kk k k kk tt IIIII λννλνν loglog 0 0,01,00 ( ) [ ] ( ) ( ) k kkk kkk k kk k k kkII µνθνλλνν loglog 0 0 ∑ ∑ ∑ ∑ ∆−+∆−+∆+∆≈ [T1] [T2] [T3] [T4] where θk (≡νk λk) is subgroup k’s share of total population income, subscript t denotes the period, ∆ is the difference operator and a bar over variables denotes the average of that variable over the two periods. As Jenkins (1995) argues “the approximation is more useful than the exact decomposition because it relates inequality changes to changes in sub-group inequalities, shares and means” (p. 38). For clarity’s sake, instead of using the decomposition in [6], we use proportionate changes (%∆I0 = ∆I0/I0,t), and reinterpret terms T1 to T4 accordingly. 25 Note that we could have used any of the generalized entropy family measures to do the decomposition analyses. We choose I0 for its good and convenient decomposition properties. In particular, I0 is the only measure of the generalised entropy class whose decomposition weights are a sole function of population shares. That is, weights do not depend on subgroup incomes—see Shorrocks, 1980. 14 Therefore, overall inequality changes can be decomposed into changes in sub-group inequalities [T1], changes resulting from changes in subgroup’s population weights [T2 and T3], and changes resulting form changes in subgroup’s relative incomes [T4]. As we pointed out above, when estimating the impact of a given attribute on inequality level or its change, these two decompositions only consider the effect of every attribute separately. In order to estimate the joint effect of all relevant characteristics available in the ECPF, we use a different decomposition suggested by Fields (1998). Fields (1998) considers a log-linear income generating function [7] ( ) εβ+==Ψ ∑ jjjXyln where Xj are individual or household characteristics, βj are the corresponding OLS coefficients, and ε is the regression residual. If [7] is a good representation of the income generating process, the contribution sj of each characteristic j can be expressed as: [8] [ ] 2 ,cov Ψ Ψ =σ βjj j X s where the numerator is the covariance between βjXj and Ψ, and the denominator is the variance of log-income, σ2Ψ. Note that this is an exact and additive decomposition. [9] ∑ = jjRs2 [10] [ ] 1 ,cov 1 12= Ψ ∑ + =Ψ J j jj X σ β. Hence, each factor’s contribution to the explanatory capacity of the model can be defined as, [11] 2 R sj j=η Fields (1998) shows that the decomposition in [10] holds for a wide class of inequality measures, including the Atkinson class, Ae, the generalised entropy class, Ic, and the Gini coefficient. In particular, [10] holds for any continuous and symmetric inequality measure I(Ψ1, …, ΨN), for which I(µ, …, µ)=0. In sum, as long as we agree on the income generating process in [7], Fields’ methodology allows 15 us to decompose overall income inequality by socioeconomic characteristics of the population without having to agree on a particular inequality measure. In other words, the effect of each characteristic on overall income inequality is independent on the inequality measure used in the empirical analysis. To estimate the contribution of characteristic j to the change in inequality, however, we do have to opt for a particular measure for the decomposition of inequality changes is not index independent. Using I0, the contribution to the change in income inequality of characteristic j, can be expressed as [12] [ ] tt ttjttj jII IsIs ,01,0 ,0,1,01, − − =Π + ++ where sj,t denotes the contribution to level inequality of characteristic j in period t, I0,t is the inequality in period t, and [13] 1=Π ∑ jj. Therefore, as in [6], the decomposition in [12] allow us to estimate the contribution of characteristic j to the change in income inequality. Unlike [6], though, contribution Πj is net of the effect of the other characteristic included in the model on the change in income inequality. Obviously, changes in income inequality are due either to changes in the characteristics or to changes in the part of income that the income generating process in [7] cannot explain —the regression residuals. Notwithstanding this, it would be of interest to know whether changes in the characteristics result from changes in the coefficients, from changes in the distribution of the variable, from changes in the covariance or correlation between the characteristics and income, or from mere changes in income inequality. Now, note that contribution sj can also be written as [8’] [ ] Ψ Ψ⋅⋅ =σ σβ, jXj j Xcorr sj where σXj is the standard deviation of characteristic j, corr[Xj, Ψ] is the correlation between Xj and Ψ, and σΨ is the standard deviation of Ψ. For infinitesimal changes, an exact decomposition of the difference in any given sj can be obtained by logarithmically differentiating [8’] [14] [ ] Ψ −Ψ++= σσβ, jXjj Xcorrsj 16 where a bar under a variable denotes logarithmic differences.26 Why using two different decompositions to analyse the inequality change? Our strategy is to take the best of the two decompositions. On the one hand, Fields’ decomposition allows us to estimate the effect of a given factor net of the influences of other variables, but the different components of such effects are sometimes difficult to interpret. On the other hand, Jenkins’ decomposition does not net out the effect of other factors on the contribution of a given factor, but overall inequality change decomposes into very intuitive components. Thence, we use Fields’ decomposition to account for inequality change and Jenkins’ decomposition to unravel what is going on within each subgroup, that is, what explains those contributions. Our empirical results reveal the relative importance of the effects of all socioeconomic characteristics contained in the ECPFs on inequality. Generally speaking, these effects are small. Our income generating model is [15] Ψ = β0 + β1(Age) + β2(Age)2 + β3(Education Level) + β4(Sex) + β5(Civil Status) + β6(Partner with income) + β7(Size of municipality) + β8(Socioeconomic classification) + β9(Household size) + ε where individual characteristics refer to the household head. The explanatory power of all characteristics in [15] is less than 50% of income inequality (45,3% in 1985 and 38,5% in 1996, first two columns of Table 8). The individual characteristic which has a greater impact on income inequality is the educational attainment of the household head. Nonetheless, the socioeconomic classification of the household head and household composition also play an important rôle. Each one of the remaining characteristics explains less than 5% of overall inequality. Finally, note that age and sex have a small disequalizing effect. According to the empirical evidence that derives from decomposition [12], the characteristics included in [15] explain about two thirds (68.6%) of the change in income inequality between 1985 and 1996 —see Πj in the fifth column of Table 8. The variables that explain most of this changes are exactly the same as those which have the greatest contribution to inequality at a point in time: 26 Note that whenever any of the four contributions in [8’] has different signs for the two periods, the decomposition in [14] will no be applicable for that variable. For instance, if coefficient βj has different signs in the regressions for t and t+1, the logarithmic difference [log(βj,t+1/βj,t)] does no exist. Moreover, as Fields himself acknowledges, changes that occur in the real world are non-infinitesimal. The decomposition in [14] could be approximated by using percentage changes. However, in practice the latter can result in a non-exact decomposition, i.e. where all components are far from accounting for the entire change in sj —this has, indeed, been our case, and that is why results are not shown. Another objection to the exact decomposition is that both the coefficients and the correlations are functions of the covariance between the characteristics and income, so that the former cannot change without changes in the latter. A way out is to assume that characteristics are orthogonal, but again, the resulting decomposition may no be exact. 17 the socioeconomic classification, education level and household composition. Notwithstanding this, in general, the overall inequality reduction witnessed between 1985 and 1996 is accounted for by reductions in within group inequalities (term T1 in decomposition [6], shown in Table 7). To some extent, this finding challenges the traditional explanations of inequality changes founded on changes in the age or education distribution, in household composition or in the socioeconomic classification of the household head. This is not to say that those factors do not account at all for the overall income inequality reduction experienced in Spain between 1985 and 1996 —as we shall show below, the ageing population, changes in the socioeconomic classification or in the distribution by education level do have a certain impact on inequality trends—, but separately, those contributions never explain more than a third of the overall change —that is, the sum of terms T2, T3 and T4 is always smaller than a third. 18 Table 5. Income inequality decomposition by population subgroups, 1985 and 1996. Decomposition [5] Component % I0 Partition Year I0Within Between Within Between Socioeconomic classification of the household head 1985 0.202 0.030 0.171 15.1 84.9 1996 0.148 0.016 0.132 10.5 89.5 Education level of the household head 1985 0.202 0.041 0.161 20.4 79.6 1996 0.148 0.026 0.122 17.4 82.6 Age of the household head 1985 0.202 0.003 0.199 1.6 98.4 1996 0.148 0.001 0.146 1.0 99.0 Cohort of the household head 1985 0.202 0.003 0.199 1.6 98.4 1996 0.146 0.001 0.146 0.5 99.5 Sex of the household head 1985 0.202 0.000 0.202 0.1 99.9 1996 0.148 0.000 0.147 0.2 99.8 Household size 1985 0.202 0.017 0.185 8.4 91.6 1996 0.148 0.008 0.139 5.5 94.5 Household composition 1985 0.202 0.021 0.181 10.4 89.6 1996 0.148 0.010 0.138 6.7 93.3 See [5] for the definition of within and between components. Partitions are defined in Appendix 3 19 Table 6. Population shares, relative mean income and within group inequality, 1985 and 1996. νk(≡nk/n)λk(≡µk/µ)I0µk 1985 1996 1985 1996 1985 1996 1985 1996 Socioeconomic classification of the household head Unemployed 0.07 0.06 0.51 0.64 0.25 0.33 258,950 457,058 Entrepreneurs w/ dep. empl. 0.03 0.02 1.31 1.15 0.18 0.14 673,635 818,203 Agr. Entrepr. wo/ dep. empl. 0.05 0.03 0.68 0.76 0.28 0.16 349,672 542,215 Agricultural workers 0.03 0.02 0.64 0.77 0.15 0.13 326,902 548,846 Entrepreneurs wo/ dep. empl. 0.07 0.08 0.93 0.83 0.16 0.13 474,393 594,947 Other employees 0.33 0.33 1.26 1.21 0.150.13 646,009 865,287 Retired 0.30 0.39 0.93 0.95 0.14 0.10 474,551 677,367 Blue collar employees 0.09 0.07 0.97 0.89 0.25 0.13 494,703 633,356 Other non-active 0.03 0.01 0.89 0.77 0.24 0.16 456,247 547,421 Education level of the household head Illiterate 0.04 0.03 0.67 0.74 0.16 0.10 342,189 524,448 Without studies 0.25 0.19 0.77 0.83 0.15 0.10 395,888 593,420 Primary education (1st cycle) 0.48 0.41 0.90 0.90 0.17 0.11 461,807 638,242 Primary education (2nd cycle) 0.08 0.15 1.29 0.96 0.17 0.13 660,773 681,806 Secondary education 0.06 0.14 1.41 1.23 0.12 0.17 722,904 876,807 Short BA & BSc. 0.05 0.04 1.62 1.57 0.12 0.12 828,729 1,122,794 BA, BSc. & Doctors 0.04 0.04 2.18 1.91 0.15 0.14 1,118,301 1,358,608 Age of the household head <25 0.02 0.02 0.91 0.80 0.26 0.18 466,699 567,408 26-35 0.15 0.13 1.17 1.11 0.21 0.21 601,281 791,376 36-45 0.21 0.20 0.99 0.99 0.26 0.16 508,427 703,496 46-55 0.23 0.20 0.94 1.03 0.20 0.17 479,928 734,905 56-65 0.20 0.20 1.03 0.96 0.21 0.13 525,561 685,342 >65 0.19 0.26 0.930.97 0.10 0.10 475,559 693,821 Cohort of the household head >=1960 0.02 0.16 0.91 1.08 0.26 0.20 466,699 774,315 1950-59 0.15 0.20 1.17 0.99 0.21 0.17 601,281 711,584 1940-49 0.21 0.20 0.99 1.01 0.26 0.17 508,427 725,763 1930-39 0.23 0.21 0.94 0.98 0.200.13 479,928 698,213 1920-29 0.20 0.17 1.03 0.96 0.21 0.09 525,561 687,308 <1920 0.19 0.08 0.93 0.96 0.10 0.09 475,559 684,949 Sex of the household head Male 0.86 0.79 0.99 0.99 0.20 0.15 509,354 704,556 Female 0.14 0.21 1.04 1.04 0.20 0.14 531,262 744,568 Household size 1 Member 0.08 0.11 1.29 1.17 0.24 0.15 662,331 835,604 2 Members 0.21 0.24 1.12 1.08 0.19 0.13 571,937 772,001 3 Members 0.19 0.22 1.07 1.04 0.14 0.14 549,189 744,195 4 Members 0.24 0.26 1.04 0.95 0.20 0.12 534,014 675,673 5 Members 0.15 0.11 0.83 0.85 0.16 0.18 424,902 608,225 6+ Members 0.13 0.06 0.66 0.68 0.22 0.14 340,717 485,347 Household composition 1 adult 0.04 0.04 1.65 1.40 0.32 0.23 845,988 996,779 2 adults 0.11 0.12 1.26 1.19 0.21 0.15 647,801 847,366 3+ adults 0.22 0.27 1.01 1.01 0.17 0.14 516,515 717,649 1 adult + child 0.01 0.01 0.96 0.95 0.63 0.29 492,961 674,687 2 adults + 1 child 0.07 0.07 1.19 1.05 0.15 0.18 607,458 750,600 2 adults + 2 children 0.11 0.08 1.08 0.93 0.19 0.15 553,134 666,513 2 adults + 3+ children 0.06 0.02 0.83 0.63 0.26 0.13 426,568 446,544 3+ adults + children 0.20 0.13 0.73 0.80 0.19 0.14 372,327 569,642 1 adult, HH>65 years 0.04 0.07 0.94 1.02 0.08 0.07 483,746 727,639 2+ adults, HH>65 years 0.15 0.19 0.92 0.96 0.11 0.11 473,334 682,001 20 Table 7. Subgroup decomposition [6] of the changes in overall income inequality, 1985 and 1996. Percentages Partition T1 T2 T3 T4 Socioeconomic classification 58.9 13.7 7.2 20.8 Education 75.3 -3.8 -5.6 30.8 Age 84.6 12.3 0.4 2.9 Cohort 127.2 -31.6 -1.0 3.2 Sex 100.1 0.2 -0.1 -0.2 Household size 81.2 2.4 8.3 7.8 Household composition 70.4 9.3 6.5 15.3 Terms T1, T2, T3 and T4 are defined in [6] and in the main text. T1 is the contribution resulting from changes in inequality within subgroups. T2 and T3 correspond to the contribution resulting from changes in subgroup population shares. T4 is the contribution resulting from changes in subgroup mean incomes. Partitions are defined in Appendix 3. Next we use the three decompositions defined above to analyse the contribution of every characteristic to inequality level at a point in time and its change between 1985 and 1996. a) SOCIOECONOMIC CLASIFICATION OF THE HOUSEHOLD HEAD As we have pointed out above, the social classification of the household head is the factor that most contributes to the change in income inequality —term T1 in table 7 is about 60% and Πj, in Table 8, is 25%. In the decomposition [6] the remaining 40% (T2+T3+T4) is accounted for by changes in subgroup’s population shares (T2+T3) and changes in subgroup’s mean incomes (T4), in the same proportion. These changes imply a change in the correlation between the socioeconomic classification and income that explains the change in the contribution of this variable to the change in income inequality —see Table 8. The most notorious change in the subgroup’s population shares is closely related to the ageing of the population: the retired increase their relative share by 9 points —see Table 6. This change in the retired’s share reinforces the equalising impact that the substantial fall in this subgroup’s income inequality has on overall inequality change. As the between component in Table 5 shows inequality among relative income means also fell. Notwithstanding this, not all groups having mean income below the sample mean, increase their relative mean. The relative mean incomes of the only two subgroups having mean incomes above the sample mean (‘entrepreneurs with dependent employees’ and ‘other employees’) fall. Indeed, several groups experience a substantial fall in their relative mean incomes. The ‘entrepreneurs without dependent employees’ see their mean income reduced 10 points, whereas that of the ‘blue collar workers’ and of the ‘other workers’ gets reduced 8 points. It is also worth noting the 13 points increase in the relative mean income of the ‘unemployed’. The ‘retired’, however, benefit from a slight increase in their relative income mean. Such a small increase is, nonetheless, insufficient to push their mean income above the sample mean. 21 Table 8. Contribution of the socioeconomic characteristics to overall income inequality and its change, 1985 and 1996. (Decompositions [8], [11], [12], [14]). sj% R2ΠjDecomposition of the change in sj (%) Variable 1985 1996 1985 1996 A B C D Age -0.5 -0.2 -1.1 -0.4 -1.5 2.6 -3.7 108.5 7.4 Education 15.7 13.2 34.6 34.3 24.1 116.5 30.3 9.2 56.0 Sex -0.4 -0.4 -0.9 -1.0 -0.4 1365.1 -412.3 -539.6 313.2 Civil status 2.2 1.4 4.9 3.7 5.0 83.6 -22.5 65.2 26.4 Spouse with income 4.4 4.4 9.8 11.4 4.5 2382.0 -2730.9 2308.2 1859.2 Size of Municipality 2.5 2.9 5.6 7.4 1.3 256.4 4.6 -434.0 -273.0 Socioeconomic classification 12.9 9.428.5 24.4 25.0 -3.1 4.8 106.3 7.9 Household composition 8.4 7.7 18.5 20.1 10.5 15.9 81.4 13.3 10.5 R245.3 38.5 100.0 100.0 68.6 No. Observations 3034 3067 sj is the contribution of variable j to income inequality at a point in time (see [8]). % R2 is the contribution of variable j to the change in income inequality as a percentage of the explanatory power of the model: 100[sj /R2]. Πj is the contribution of variable j to the change in income inequality (see [12]). Decomposition of the change in sj (expressed as % sj): A: Relative change of the coefficient. B: Relative change of the standard deviation of the variable. C: Relative change of the correlation between the variable and the logarithm of income. D: Relative change of the standard deviation of the logarithm of income. Variables are defined in Appendix 3. Inequality fell for all socioeconomic groups, with the notable exception of the unemployed. The reduction, though, is not the same for all groups. In general, groups with lower mean income experience a greater reduction in within inequality —note the substantial fall of the ‘agricultural entrepreneurs with dependent employees’, the ‘blue collar workers’ and the ‘other non active’. The most notable exception is the ‘unemployed’ group, for which inequality increases drastically. What is driving such and increase? Although the answer to this question is clearly beyond the scope of this paper, some tentative explanations could include a change in the benefits policy, a change in the composition of the unemployed subgroup, or a change in the income distribution of the unemployed in the employment spell previous to the interview. b) EDUCATION LEVEL OF THE HOUSEHOLD HEAD The change in population shares by education level has a disequalizing effect of about 10% of the overall inequality change —see T2+T3 in Table 7. Such an equalizing effect is mainly due to the decreasing proportion of households whose head has primary education (1st cycle) or less. Changes in subgroup’s relative mean income, however are equalizing —term T4 in Table 7. Overall, relative mean incomes increased for households whose head attained lower education levels (having also smaller relative mean incomes) and fell for those headed by individuals with higher educational attainments. In terms of the decomposition in [14] this implied a reduction in the education coefficient.27 27 It is worth noting that those coefficients are not returns to education estimates. Actually, according to Barceinas et al. (2000), returns to education increased for more educated individuals. Thence, changes in the distribution of income sources other than 22 The fall in within group inequality, which accounts for three fourths of overall inequality reduction, is mainly experienced by households headed by individuals with low educational attainments (less than secondary education). In fact, within group inequality only rises for households headed by individuals with secondary education —this is also the group with the highest subgroup population share increase. Changes in within group inequality follow a clear pattern. In 1985 within group inequality was lower the more educated groups than for less educated ones. Twelve years later, the situation is the opposite. In sum, during these twelve years, more educated group’s income fell in relative terms, but its dispersion rose, whereas the less educated group’s relative income and dispersion behaved the opposite way. c) HOUSEHOLD COMPOSITION Changes in household composition account for 10.5% of overall income inequality change, and this percentage rises to 70.4 if the effect of the other variables are not taken into account. On the one hand, these changes in household composition are due to a greater equality of subgroup’s mean incomes: relative mean income falls (increases) for those groups having mean incomes above (below) the sample mean. On the other hand, those changes point at certain changes in the population shares of the more numerous households and those with and individual aged 65 or more, which reduces the dispersion of this variable. Within group inequality falls for all groups, except for those with two adults and a child. Finally, we would like to comment on the lone parent households group. Even though its population share is still very small, and has hardly changed over the period, and its mean income is only 5 points below the sample mean, its inequality fell drastically. Notwithstanding all this, in 1996, lone parent households still showed the highest income inequality. d) AGE OF THE HOUSEHOLD HEAD The sign of the effect of age on the overall inequality change differs according to the decomposition used. When the joint effect of all variables is taken into account —decomposition [14]— age has a small disequalizing impact. However, when considering the sole effect of age — decomposition in [6]—, the impact is equalizing. The small disequalizing effect is accounted for by the falling (negative) correlation between age and income. Note also that the increasing dispersion of the age variable has prevented the negative effect from being greater. Indeed, the main change in population shares is an increase of the ‘older than 65’ group’s share. It is also worth noting the substantial fall in relative mean income of the youngest group, which has a small effect on overall inequality change because its population share is very small (2%) and it does not change over the period. labour income should explain the decreasing education coefficients observed in decomposition [14]. 29 Figure 10. I0 for 6 definitions of income (equivalent, OCDE). 1985-1996. 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 NI GI BTI Expenditure Adjusted Exp. Income Adjusted to NA 5.3. Robustness to trimming One of the problems that arise when working with survey micro data concerns the quality of the data: these may be misreported, miscoded, or just missing. Obviously, poor quality data may lead to misleading conclusions if the observed income distribution differs substantially form the true and unobserved one. Even, when dealing with relatively big samples, one or two anomalous observations may hamper or drive the estimation of inequality measures and other distributional tools such as the Lorenz curve.36 The aim of this section is to check the robustness of our main conclusions to dirty data. Following Cowell and Victoria-Feser’s suggestions we strategically trim the income distribution and implement dominance criteria to the modified distributions. Data imperfections could be modelled using data contamination models (Cowell, 1998).37 These models assume that the observed distribution is a mixture of two distributions: the true and an ‘alien’ contamination distribution. Unfortunately, we do not have sufficient information to model such data imperfections. Instead we compare dominance results from strategically trimmed 36 See Cowell and Victoria-Feser (1996, 1998) for analyses of the robustness of several distributional tools usually employed to assess income distributions. 37 To model certain types of data contamination, such as the systematic income misreporting of well-identified groups (e.g. selfemployed), error measurement models are thought to be the most appropriate ones. 30 distributions. Given that dominance outcomes are more likely to be affected if random ‘spots’ of contamination occur in either tail of the distribution (Cowell et al., 1999), we compare dominance outcomes under 1% and 5% ‘balanced’ and ‘unbalanced’ trims. A ‘balanced’ trim removes a small percentage (q) of observations from each tail of the distribution (that is, removes a 2q percentage of the sample), whereas ‘unbalanced’ trims remove a small percentage (q) from only one tail. Table 12 shows the results of the trimming in terms of Lorenz and generalized Lorenz dominance for three equivalent scales: OECD’s (i.e. a = 0.69 and b = 0.82 in CK), household income (i.e. b = 0), and per capita income (i.e. a = 1 and b = 1). In general, the Lorenz dominance conclusion of the 1996 income distribution on the 1985’s is robust to all trims and for the three equivalence scales. Moreover, both Lorenz and generalized Lorenz dominance outcomes are also quite robust when dividing the whole period in 4 sub-periods that replicate the phases of the Spanish economic cycle. Table 12. Effect of trimming on Lorenz and generalized Lorenz dominance, 1985-1996. Equivalent Income (OCDE) household Income Per Capita Income Period NT BT UU UL NT BT UU UL NT BT UU UL Lorenz dominance 1% Trimming 1985-1996 D= = = D= = = D=== 1985-1989 D= = = D= = = D=== 1989-1991 Ca1 = = = Ca2 Ca5 Ca4 =Ca1 dCa1 = 1991-1994 Cb1 Ca2 Ca2 =Cb1 =D=Cd4 =Ca5 = 1994-1996 Cb1 = = = Ca1 d d =Ca1 === 5% Trimming 1985-1996 D= = = D= = = D=== 1985-1989 D= = = DCa1 Ca2 =D=== 1989-1991 Ca1 d d =Ca2 Cb3 Ca1 =Ca1 dd= 1991-1994 Cb1 D D =Cb1 D D =Cd4 Ca5 Ca3 Ca3 1994-1996 Ca1 = = = Ca1 = = DCa1 === Generalized Lorenz dominance 1% Trimming 1985-1996 D= = = D= = = D=== 1985-1989 D= = = D= = = D=== 1989-1991 D= = = D= = = D=== 1991-1994 D= = = Cb1 D D =D=== 1994-1996 D= = = D=Ca4 =D=== 5% Trimming 1985-1996 D= = = D= = = D=== 1985-1989 D= = = D= = = D=== 1989-1991 D= = = D= = = D=== 1991-1994 D= = = Cb1 D D =D=== 1994-1996 D= = = DCb1 Cb1 Ca2 D=== NT: No trimming; BT: Balanced trim; UU: Upper tail Unbalanced trim; UL: Lower tail Unbalanced trim. D: Dominance of the final year of the period’s curve over that of the first one. d: Dominance of the first year of the period’s curve over that of the final one. Cb#: the final of year period’s curve crosses from below that of the first one # times. Ca#: the final of year period’s curve crosses from above that of the first one # times. =: same result as that for the whole untrimmed sample. 31 6.Conclusion In this paper we analyse the Spanish income distribution for the period 1985-1996 using data from the ECPF and a wide range of visual and analytical techniques. Our results corroborate the existing evidence on falling income inequality for the second half of the Eighties and show that this trend extends over the first half of the Nineties. Moreover, we also find that this trend is robust to changes in the equivalence scale, to changes in the income distribution and to data contamination. Taking advantage of the socioeconomic information contained in our data set, we also study some of the causes which may be explaining the overall picture of income inequality and its change over this period by means of three decomposition analyses. Three main responsible factors clearly stand out: the socioeconomic situation and education attainment of the household head, and household composition. Those trends and the factors accounting for them raise important policy issues to be tackled in further investigations. In particular our analysis by population subgroups identifies several key subgroups such as the lower educated, the retired or the lone parents, and which denote the importance of the demographic change our economy is experiencing. The link between the distribution of labour income and the distribution of total income is also specially important if we want to understand the impact of certain changes such as the increasing number of temporary contracts, the growing female participation rates, technical change, the growing terciarization of the economy or the increasing proportion of public sector contracts. 32 References Alba-Ramírez, A. and Collado, L. (1999), “Do wives’ earnings contribute to reduce income inequality?: evidence from Spain”, Working Paper, AD99-11, IVIE. Alcaide Inchausti, J. (1989): “La distribución de la renta española en la década de los ochenta”, Documentos de Trabajo de la Fundación FIES, 38/1989 (publicado también en Cuadernos de Información Económica, abril 1989). Alcaide Inchausti, J. (1991): “Política de distribución de la renta”, Revista de Economía, 11. Ayala, L., Martínez, R. and Ruiz-Huerta, J. (1996), “La distribución de la renta en España desde una perspectiva internacional: tendencias y factores de cambio”, Actas del II Simposio sobre desigualdad y distribución de la renta y la riqueza, Vol. 6, La desigualdad de recursos, Fundación Argentaria, Madrid. Atkinson, A.B. (1970): “On the measurement of inequality”, Journal of Economic Theory, 2: 244-63. Atkinson, A.B. (1987): “On the measurement of poverty”, Econometrica, 55, 749-64. Atkinson, A.B., Rainwater, L. and Smeeding, T.M. (1995), Income Distribution in OECD countries, Social Policy Studies, 18, OECD. Bandrés, E. (1990), Los efectos de los gastos sociales sobre la distribución de la renta en España, Madrid: Instituto de Estudios Fiscales, Monografía, 92. Bandrés, E. (1993), “La eficacia redistributiva de los gastos sociales. Una aplicación al caso español (198090), in I Simposio sobre igualdad y distribución de la renta y la riqueza. Volum 7, Fundación Argentaria: 123-171, Madrid. Barceinas, F., J. Oliver-Alonso, J.L. Raymond and J.L. Roig (2000): “Private rates of return to human capital in Spain: New evidence”, Documentos de Trabajo No. 162/2000, FUNCAS, Madrid. Buhmann, B., et al. (1988): “Equivalence scales, well-being, inequality, and poverty: sensitivity estimatesacross ten countries using the Luxembourg Income Study (LIS) Database”. Review of Income & Wealth, 34(2):115-42. Castañer, J.M. (1991), “El efecto redistributivo del IRPF 1982-1988”, Hacienda Pública Española, monográfico, 2/1991. Cowell, F. (1980): “On the structure of additive inequality measures”, Review of Economic Studies, 47:52131 Cowell, F. (1995): Measuring Inequality, 2ª edición, Hemel Hempstead: Harvester Wheatsheaf, Londres. Cowell, F. (1998): “Measurement of Inequality” in Handbook of Income Distribution, (eds. A.B. Atkinson y F. Bourguignon), North Holland, Amsterdam, chapter 2. Cowell, F. and M.P. Victoria-Feser (1996): “Robustness properties of inequality measures”, Econometrica, 64(1): 77-101. Cowell, F. and M.P. Victoria-Feser (1998): “Distributional dominance with dirty data”, Working Paper, STICERD, London School of Economics, Londres. Cowell, F., J.A. Litchfield and M. Mercader-Prats (1999): “Income inequality comparisons with dirty data: the UK and Spain during the 1980s”, DARP Working Paper 45, STICERD, London School of Economics, Londres. Cutler D. and L. Katz (1992): “Rising Inequality? Changes in the Distribution of Income and Consumption in the 1980s. AER Papers and Proceedings. Díaz, C. and F. Melis (1993), “La distribución sectorial y personal de la renta de las empresas personales”, en I Simposio sobre igualdad y distribución de la renta y la riqueza. Volum II La distribución de la renta, Fundación Argentaria: 171-198, Madrid. Duclos, J.-Y. and Mercader-Prats, M. (1999): “Household needs and poverty: with application to Spain and the UK”, Review of Income and Wealth, 45(1): 77-98. Esteban J.M. and D. Ray (1994): “On the measurement of polarization”, Econometrica, 62, 819-51. Esteban J.M., C. M. Gradín and D. Ray (1997): “A new measure of polarization”, mimeo. Fields, G. (1998): “Accounting for income inequality and its change”, Cornell University, mimeo. Garde, J.A., Martínez, R. and J. Ruiz-Huerta (1996), “Los estudios sobre distribucion de la renta en España: fuentes, resultados, perspectivas de futuro”, in II Simposio sobre igualdad y distribución de la renta y la riqueza. Volum 6 La desigualdad de recursos, Fundación Argentaria: 257-314, Madrid. Gimeno, J.A. (1993), “Incidencia del gasto público por niveles de renta (España 1990 vs. 1980), Actas del I Simposio sobre Igualdad y Distribución de la Renta y la Riqueza, Vol. VII: 63-121, Madrid. Goërlich, F.J. and Mas, M. (1999), “Medición de la desigualdad: contribución a una base de datos regional”, Documento de Trabajo, IVIE, septiembre. Gottschalk, P. and Smeeding, T.M. (1998), “Empirical Evidence on Income Inequality in Industrialised Countries”, in Atkinson, A.B. and Bourguignon, F. (eds) Handbook of Income Distribution, North Holland, Amsterdam, chapter, 3. 33 Gradín, C. (1997): “The measurement of bipolarization: Decomposition by subpopulation. The case of Spain, 1973-1991”, Universidad de Vigo, mimeo. Gradín, C. and M.S. Otero (1999): “Incorporación laboral de la mujer en España: efecto sobre la desigualdad en la renta familiar”, Documento de Trabajo 9905, Universidad de Vigo. Imedio, L.J, E.M. Parrado and M.D. Sarrión (1997): “Evolución de la desigualdad y la pobreza en la distribución de la renta familiar en España en el periodo 1985-1995”, Cuadernos, 32: 93-109. Jenkins, S.P. (1991): “The measurement of economic inequality”, in Readings on Economic Inequality, L. Osberg (ed.). M.E. Sharpe, Armonk, N.Y. Jenkins, S.P. (1994): “Dwarfs and giants in the 1980s: Trends in the UK income distribution”, Fiscal Studies, 15 (1): 99-118. Jenkins, S.P. (1995): “Accounting for inequality trends: decomposition analyses for the UK, 1971-86”, Economica, 62, 29-63. Jenkins, S.P. (1996): “Recent trends in the UK income distribution: What happened and why?”, Oxford Review of Economic Policy, 12 (1): 29-46. Lambert, P. (1993): The Distribution and Redistribution of Income, Manchester University Press, Manchester. Lambert and Ramos (1997): “Horizontal Inequity and Vertical Redistribution”, International Tax and Public Finance, 4: 25-37. Lasheras, M.A., Rabadán, I. and Salas, R. (1993), “Política redistributiva en el IRPF entre 1982 y 1990 y efectos redistributivos del IRPF entre Comunidades Autónomas 1982-1989”, Actas del I Simposio sobre Igualdad y Distribución de la Renta y la Riqueza, Vol. VII Desequilibrios interregionales:. 7-24, Madrid. Medel, B., Molina, A., and Sánchez, J. (1988): Los efectos distributivos del gasto público en España”, FIES Documento de Trabajo, nº 28/1988. Melis, F. and C. Díaz (1993), “La distribución personal de salarios y pensiones en las fuentes tributarias”, en I Simposio sobre igualdad y distribución de la renta y la riqueza. Volumen II La distribución de la renta, Fundación Argentaria: 151-170, Madrid. Oliver, J. (1997): “Ingreso, consumo y ahorro de las familias: propuesta de una metodología para la explotación de la Encuesta de Presupuestos Familiares”, Documentos de Trabajo FIES Nº 132, Madrid. Oliver, J. and J.L. Raymond (1999): “La distribución de la renta en España en el periodo 1985-1996: Resultados derivados de la Encuesta Continua de Presupuestos Familiares”, Cuadernos de Información Económica, 150: 1-4. Pen, J. (1971): Income Distribution, Harmondsworth, Middx, Penguin. Revenga, A. (1991): “La liberalización económica y la distribución de la renta: la experiencia española”, Moneda y Crédito, nº 193: 179-223. Río, C. del and Ruiz-Castillo, J. (1996): “La desigualdad de recursos”, in II Simposio sobre igualdad y distribución de recusos de la renta y de la riqueza. Volum 6, La desigualdad de recursos, Fundación Argentaria, Madrid. Saposnik, R. (1981): “Rank dominance in income distribution”, Public Choice, 36, 147-51. Shorrocks, A. (1980): “The class of additively decomposable inequality measures”, Econometrica, 48(3): 613-625. Shorrocks, A. (1982): “Inequality decomposition by factor components”, Econometrica, 50, 193-212. Smeeding, T.M. and Gottschalk, P. (1995): “The International Evidence on Income Distribution in Modern Economies: Where Do We Stand?”, Working Paper 137, Luxembourg Income Study, CEP/INSTEAD. Theil, H. (1967): Economics and Information Theory. North Holland, Amsterdam. Zaidi, M.A. and Vos, K. de (1998): “Trends in Consumption-based Poverty and Inequality in the Member States of the European Union during the 1980s”, Discussion Paper, Economics Institue, Tilburg. 34 Appendix 1. Components of the real net household income Our definition of net income is: 1. Net: + Monetary net income labour net income self-employment net income capital net income net pensions net unemployment benefits other net benefits other monetary net income + Non-monetary net income labour non-monetary income self-employment non-monetary income imputed rents for house ownership38 other non-monetary income - SS Contributions Social Security contributions paid by the employees Social Security contributions paid by the employer Social security contributions paid by the self-employed Contributions for unemployment paid by the employees to Social Security - Direct taxes Tax withheld on salary to employees Tax withheld on salary to self-employed Tax withheld on capital Balance of tax payable (refundable) 2. Real. As pointed out above, we deflated nominal income by using a retail price index for each income decile. These income decile RPI was estimated using the household expenditure structure given in the ECPFs. Income is expressed in 1985 prices. 3. Equivalent: in accordance with the inequality and welfare literature we use the OECD equivalent scale. This way income becomes a better indicator of living standard. 4. Household: the unit of analysis is the household. Therefore, and consistent to our previous considerations, each household is assigned its real net equivalent income. It would be interesting to analyse to what extent our results would change if we assigned the household equivalent income to each individual.39 38 We have imputed the rents of households living in their own house. See Oliver (1997) for further details on the imputation procedure. 39 For an example of this practice, see Jenkins (1995). 35 Appendix 2. Income distribution, 1985-1996 Table 13. Real net equivalent income by vingtiles, 1985-1996. In 1985 prices percentile 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 5146,554 161,992 186,504 203,246 232,946 246,476 242,543 272,888 259,798 256,955 253,152 253,081 10 201,158 218,052 240,201 254,084 280,266 292,627 305,068 333,640 327,775 321,481 320,160 323,643 15 241,375 252,299 276,742 290,816 318,576 331,721 346,492 371,116 372,189 370,944 367,201 370,954 20 274,325 281,162 301,161 319,768 350,082 367,102 381,623 404,610 407,801 406,538 405,892 412,741 25 298,259 308,179 327,985 355,179 376,985 398,651 415,230 437,876 442,158 436,776 448,389 450,748 30 320,312 331,728 355,224 385,196 405,393 425,961 444,428 471,265 478,289 471,183 483,161 483,943 35 347,501 357,628 380,942 415,146 430,971 460,011 477,371 501,051 508,386 502,315 513,883 519,887 40 373,471 385,213 407,322 439,206 460,021 494,702 504,124 535,166 538,033 535,332 544,611 552,406 45 401,073 409,529 433,190 466,603 488,546 519,711 536,252 564,197 568,951 569,418 580,467 589,610 50 426,761 438,177 459,082 496,176 517,107 551,248 568,065 602,763 604,957 599,983 612,725 622,508 55 458,819 463,830 489,411 528,377 549,869 585,463 608,561 634,626 636,175 633,933 649,124 661,161 60 494,893 493,910 518,390 558,730 584,445 618,339 642,276 666,867 681,168 676,543 691,362 708,116 65 527,935 525,242 554,604 597,146 625,415 654,804 689,386 710,588 723,904 724,665 737,826 745,176 70 569,801 566,880 601,111 632,444 665,835 698,936 731,458 760,951 781,215 776,244 790,120 800,896 75 618,726 620,392 643,909 682,336 712,844 754,005 794,163 820,663 844,399 835,392 853,986 862,890 80 685,454 679,815 699,575 747,410 777,698 825,195 868,710 895,895 932,343 913,194 924,283 944,129 85 761,317 763,403 776,745 822,931 857,755 915,266 970,990 991,863 1,021,353 1,015,144 1,024,853 1,051,374 90 889,607 887,312 892,214 938,574 980,832 1,039,265 1,118,375 1,147,413 1,182,496 1,155,487 1,186,191 1,184,574 95 1,111,840 1,107,687 1,085,300 1,128,940 1,222,284 1,308,198 1,370,938 1,391,527 1,502,338 1,433,845 1,466,947 1,461,742 Mean 512,512 508,327 531,715 565,340 602,146 637,448 661,551 688,494 703,228 698,005 710,191 712,934 Table 14. Cumulative real net equivalent income, 1985-1996. (Generalized Lorenz curves co-ordinates) In percentage Percentile 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 54,041 6,505 5,978 7,128 6,020 6,112 7,838 9,056 9,615 7,944 8,104 9,485 10 12,607 15,984 16,056 17,912 16,894 17,928 20,548 22,309 22,584 21,185 21,384 23,486 15 23,112 27,321 28,354 31,255 30,680 32,596 35,800 38,655 38,460 37,295 37,592 40,197 20 35,396 39,774 42,360 46,609 46,797 49,301 53,170 57,209 56,795 55,611 56,275 59,618 25 49,134 53,899 57,732 63,425 64,855 68,451 72,235 77,751 77,366 76,134 77,209 80,846 30 64,488 69,326 74,983 81,886 84,662 89,231 93,418 100,060 99,950 98,643 100,170 103,880 35 81,297 86,425 93,430 101,809 106,215 111,844 116,296 124,799 124,546 123,138 125,156 128,721 40 99,561 105,011 113,414 123,011 129,128 136,290 141,292 151,084 150,931 149,620 152,168 155,594 45 119,279 124,898 134,765 145,859 153,983 162,367 167,983 179,578 179,329 177,867 180,981 184,500 50 140,614 146,458 157,482 170,352 180,197 190,481 196,792 209,839 209,739 208,320 211,819 215,213 55 163,564 169,504 181,565 196,489 208,353220,428 227,507 242,309 242,608 240,760 244,909 247,957 60 188,131 194,224 207,699 223,906 238,062 252,209 260,129 276,767 277,713 275,186 279,799 282,961 65 214,476 220,802 235,369 253,334 269,907 285,823 294,870 313,434 315,055 312,260 317,166 319,996 70 242,922 248,867 264,918 284,407 303,889 321,678 332,152 352,088 355,080 351,761 357,234 359,516 75 273,792 279,162 296,859 318,038 340,394 359,978 371,977 393,615 397,564 393,690 399,778 401,971 80 307,249 312,059 331,190 354,046 379,618 401,130 415,190 438,233 443,402 438,929 445,923 447,588 85 344,261 348,116 368,597 393,161 422,531 446,356 461,793 486,607 493,042 488,361 495,670 497,270 90 385,960 388,448 410,614 437,028 470,105 496,472 513,480 540,281 547,824 543,089 550,595 552,597 95 434,124435,284 458,781 488,573 524,669 554,533 573,005 601,908 610,880 605,320 615,199 616,732 100 512,512 508,327 531,715 565,340 602,146 637,448 661,551 688,494 703,228 698,005 710,191 712,934 36 Table 15. Lorenz and generalized Loren dominance for three definitions of income, (equivalent, OECD) Period NI GI Expenditure Period NI GI Expenditure Lorenz dominance Generalized Lorenz dominance 1985-1996 D= = 1985-1996 D= = 1985-1989 D= = 1985-1989 D= = 1989-1991 Ca1 =Ca2 1989-1991 D= = 1991-1994 Cb1 =D1991-1994 D= = 1994-1996 Ca1 =D1994-1996 D=Ca1 NI: Net income; GI: Gross income; D: Dominance of the final year of the period’s curve over that of the first one. Cb#: the final of year period’s curve crosses from below that of the first one # times. Ca#: the final of year period’s curve crosses from above that of the first one # times. =: same result as that for the whole untrimmed sample. Figure 11. Joint density of net equivalent income and equivalent expenditure, 1996 37 Appendix 3. Definition of the variables used in the decomposition analyses of section 4 Variable Description Socioeconomic classification Unemployed (of the household head) Entrepreneurs with dependent employees Agricultural entrepreneurs without dependent employees Agricultural workers Entrepreneurs without dependent employees Other employees Retired and pensioners Blue collar employees Other non-active Education Level Illiterate (of the household head) Without studies Primary education (1st cycle) Primary education (2 nd cycle) Secondary education Short BA & BSc. BA, BSc. & Doctors Age groups <25 (of the household head) 26-35 36-45 46-55 56-65 >65 Cohort >=1960 (of the household head) 1950-59 1940-49 1930-39 1920-29 <1920 Sex Male (of the household head) Female Household size 1 Member 2 Members 3 Members 4 Members 5 Members 6+ Members Household composition 1 adult 2 adults 3+ adults 1 adult + children 2 adults + 1 child 2 adults + 2 children 2 adults + 3+ children 3+ adults + children 1 adult, HH head>65 years 2+ adults, HH head>65 years Civil status Married (of the household head) Not married Partner with salary Partner with salary Partner without salary Size of Municipality < 10,000 inhabitants Between 10,000 - 50,000 inhab. (except capitals of provinces) > 50,000 inhab., and capitals of provinces Madrid and Barcelona