Are you unhappy? Then you are poor! Multi-dimensional poverty in Belgium
The Authors
Gijs J.M. Dekkers, Belgian Federal Planning Bureau, Brussels, Belgium
Acknowledgements
The author wishes to thank Conchita D’Ambrosio for her comments on an earlier version of this paper.
Abstract
Purpose – This paper aims to present a multi-dimensional measure of poverty. The proposed method has been applied to the Panel Set of Belgian Households dataset for Belgium for the years between 1994 and 2000.
Design/methodology/approach – First, a common model is decided upon by exploratory factor analysis, and applied by confirmatory factor analysis. Cluster analysis (CA) is then used to separate the multi-dimensional poor. Finally, the possible causes of multi-dimensional poverty are surfaced by estimating a discrete duration model.
Findings – The proposed method reveals three dimensions of poverty: “material deprivation”, “social deprivation” and “psychological health”. Between 9 and 11 per cent of the representative sample of Belgian individuals are poor. The paper also identifies causes of poverty, including not having a job, not having the Belgian nationality, having a poor health or a disability, being lower educated, experiencing financial poverty, being divorced or widowed, living in the Walloon or Brussels regions, and having a bad psychological health.
Research limitations/implications – Research implications include the use of polychoric and tetrachoric correlations as a starting point of factor analysis, as well as the combination of factor analysis and CA.
Originality/value – The paper proposes an alternative multi-dimensional measure of poverty. It argues that previous measures may suffer from categorisation errors and suggests a solution to this problem. The advantages of the proposed method are that all information is used to disentangle the poor from the non-poor and that dimensions of poverty are defined using the correlations between deprivations. Finally, the paper identifies “psychological health” as one of the dimensions of poverty.
Article Type:
Research paper
Keyword(s):
Belgium; Poverty; Social psychology; Personal health; Cluster analysis.
Journal:
International Journal of Sociology and Social Policy
Volume:
28
Number:
11/12
Year:
2008
pp:
502-515
Copyright ©
Emerald Group Publishing Limited
ISSN:
0144-333X
1. Introduction and problem definition
Many have an intuitive idea of what poverty is, who is poor and who is not, but the conceptualisation, measurement and causes of poverty give rise to lengthy debates. On the conceptual level, the most well-known definition states that individuals or households “can be said to be in poverty when they lack the resources to obtain the types of diet, participate in the activities and have the living conditions and amenities which are customary or at least widely encouraged or approved, in the societies to which they belong” (Townsend, 1979, p. 31). That poverty is indeed a multi-dimensional concept is now widely acknowledged. The question remains as to how it should be measured, what its causes are.
The aim of the present paper is to introduce a multi-dimensional measure of poverty and to identify its causes. It starts by discussing how multi-dimensional poverty is measured in the literature. Next, the alternative model will be presented and applied using the 1994-2000-waves of the Panel Set of Belgian Households (PSBH). The results of this alternative multi-dimensional measure will then be used to identify the causes of poverty. This paper differs from previous work (Dekkers, 2003, 2004) in several respects. Relative to the former, it uses a different dataset and replaces standard correlations by polychorical correlations in the methodology to identify the poor. Relative to the latter, it includes a more extended conceptual discussion underpinning the variables included in the multi-dimensional measurement. Furthermore, it revises the model to surface possible causes of poverty.
For a good understanding of what follows, a distinction must be made between “poverty”, a multi-dimensional notion, and “deprivation”, a specific non-financial arrear. Poverty occurs when an individual or household experience a number of cumulative deprivations. These deprivations need to occur in different dimensions of life. Finally, define “financial poverty” is defined as a situation where a lack of disposable income is experienced. In this study, an individual is financially poor when the equivalent income (using the modified OECD-scale) of the household is below 60 per cent median income.
2. Existing multi-dimensional approaches to measuring poverty
Measuring multi-dimensional poverty usually involves the construction of an index incorporating the information from the indicators. However, one still has to decide when a household or individual is said to be poor. One strategy is to assign each single indicator of deprivation its own threshold value (Chakravarty et al., 1998; Tsui, 2002; Guio, 2005). Then a minimum number of deprivations are decided upon, at which point one is considered poor.
Other studies combine the individual indicators into one index variable and assign a threshold (Townsend, 1993, p. 57; Nolan and Whelan, 1996, p. 230; Tsakloglou and Papadopoulos, 2002). The advantage of this approach is that it is compensatory: a high score on a certain indicator may neutralize a low score on another. This strategy however involves taking two important decisions, which might cause debate. The first decision pertains to which variables to include in the index, or else how to weigh the variables in the index. Nolan and Whelan (1996) use factor analysis to select indicators. Halleröd (1995) does not exclude any indicators, but varies the weights. The second decision is what threshold value to adopt. Tsakloglou and Papadopoulos (2002) set the cut-off point equal to 70 or 80 per cent of the median of the index. Layte et al. (2001) set it so that the proportion of multi-dimensional poor equals the exogenous proportion of financial poor. Muffels (1993) combines the index with a question of the perceived welfare level of the household. By imposing a threshold set by convention, these approaches implicitly assume that the population can actually be divided into poor and non-poor households or individuals, as is the case with financial poverty. Given the multi-dimensional nature of the poverty measure, this assumption may not be true because deprivation scores in different fields may compensate each other.
3. An alternative multi-dimensional measurement of poverty
3.1 A proposed two-step measurement
If poverty is defined as a situation where deprivations in various fields cumulate, these fields may be seen as latent dimensions that are only approximated with the available manifest indicators. To the extent that the manifest indicators are correlated with one another, the more likely it is that they represent the same dimension of poverty. Factor Analysis (FA) and Principal Component Analysis (PCA) are data-reducing techniques, which use the patterns of correlation between the indicators to collapse the available information into one or more indices. Other studies have used FA or PCA to measure poverty. Nolan and Whelan (1996) use FA to select a “basic dimension” of indicators. Whelan et al. (2001, 2002), Layte et al. (2001) and Guio (2005) follow a comparable approach.
So, if the observed variables are linear combinations of the latent dimensions of poverty, the first question is: which dimensions? Here FA is applied to a multivariate dataset to bring the latent dimensions of poverty to the surface, yielding continuous factor scores representing the arrear of an individual or household on these dimensions of poverty. In contrast with aforementioned papers, not one dimension will be selected. This of course complicates things in that we do not end up with one but several optimally weighted indices.
Given one or more indices, the second question is how to categorize individuals or households on the basis on whether or not they are poor. An obvious way of doing this would be to use an exogenous threshold value. However, we have multiple dimensions, so the problem of how to combine these indices remains. We would want to combine the information of the available dimensions so that those households or individuals surface, where deprivations on the various dimensions cumulate. Cluster analysis (CA) is the general name for a number of techniques that surfaces a latent structure in the households or individuals. Every individual is described by n factor scores, one for each dimension of poverty. These factor scores can be seen as coordinates in an n-dimensional Euclidian space. The basic CA defines the groups such that the average between-group difference are the largest, while the average within-group differences are the smallest. If one latent group in the sample experiences cumulating one-dimensional deprivations, then CA should reveal this group.
An obvious critique to this two-step procedure would be that it contains one step too many. Why not follow Townsend (1993, appendix 3.2, p. 67) and apply CA directly to the observed variables? The reason is that using continuous and standardized factor scores equalizes the a priori weight of all input variables in the CA. Suppose that 6 and 2 indicators represent latent dimensions A and than B, respectively. Erroneously ignoring these dimensions by omitting FA would result in the weight of dimension A being three times greater than that of dimension B in the CA. Likewise, suppose that two variables each describe arrear, one being dichotomous and the other having a scale from 1 to 10. If applied directly in the CA, the second variable would have a larger weight than the first variable.
To finish this presentation, let us briefly recapitulate the advantages of the proposed method to measure multi-dimensional poverty. First, FA reflects the notion that the various deprivations should cumulate for poverty to exist. Secondly, the method proposed does not select some dimensions while keeping others out. Instead, all available information is used in identifying the poor and non-poor. Finally, using CA avoids having to impose a threshold value.
3.2 Two methodological issues
The suggested two-step measure to measure multi-dimensional poverty has two methodological problems. The first is that most observed non-income variables in the PSBH are ordinal or dichotomous. This poses a problem since standard FA derives the underlying latent structure on the basis of the Pearson correlation-matrix of the observed variables. This assumes that the variables are continuous and have an interval level of measurement. Ignoring this assumption results in substantial categorisation errors if the response options are low (Muthén and Christoffersson, 1981, p. 407; Mislevy, 1986, p. 9; Coenders and Saris, 1995, p. 126). Most papers to date that use standard FA or PCA to measure poverty (Nolan and Whelan, 1996; Whelan et al., 2001, 2002; Layte et al., 2001) ignore this problem, and their empirical results therefore suffer from categorisation errors. Only Guio (2005, p. 11) follows Dekkers (2003) and uses tetrachoric and polychoric correlations as the basis of FA[1]. This paper uses the same strategy. The estimators – and therefore the factor scores – are thus consistent, although the standard errors as well as the chi-square tests of the models as a whole are not (Muthén and Christoffersson, 1981).
A second problem stems from the fact that the model is to be applied to multiple points in time. The fit of the model should be maximal at every point. However, the results should also be comparable between points in time. A compromise between fit and comparability is to develop a “base model” using exploratory factor analysis for every year, and use this in a confirmatory factor analysis.
4. The data
The two-step procedure described in the previous section was applied to the 1994-2000 waves of the PSBH. The description of the variables is included in Table I.
The three columns contain the variable name, a description and the place of the variable in the base model. There are 41 variables in total, of which 21 and 20 variables, respectively, describe the household and the individual. Note that the household and individual attributes have a different character. In fact, differences between households are based on material attributes, whereas possible differences between individuals in the same household are contentment with education, psychological health and social and cultural participation. Many variables in Table I reflect a possible difference between a current and wanted state of being, and these were selected for two reasons. First, such a difference means that a person lacks the “capability” to achieve the alternative situation (Brandolini and D’Alessio, 2000, p. 4). A second reason is that these variables are not directly dependent to what happens in the rest of the population, with the exception of the comparison of living conditions with friends and relatives. As a result, the multi-dimensional poverty measure that uses these variables as the point of departure will be conceptually absolute.
Finally, contrary to other studies (e.g. Brandolini and D’Alessio, 2000), variables that may cause poverty (such as labour market status, gender, level of education) are excluded from Table I. Including these variables would confuse cause with outcome (Tsakloglou and Papadopoulos, 2002, p. 213), and would also mean that they could no longer be used as covariates in models to be presented later in this paper.
5. The results: multi-dimensional poverty in Belgium[2]
5.1 The first step exploratory and confirmatory factor analysis
The measure of poverty described is developed by first applying FA to identify latent dimensions of poverty. A first step in confirmatory FA involves the decision how many latent dimensions or factors the model should retain. Exploratory factor analysis was therefore applied to select a base model. Following Knol and Berger (1991), we used unweighted least squares. Furthermore, we applied oblique rotation to the factor axes, for the concept of poverty postulates that one-dimensional deprivations should occur simultaneously and therefore might correlate. Table II contains for an increasing number of factors and for each year separately the eigenvalues (ev) and their proportional contributions (pr). This denotes the fraction of the total variance that is explained by the factor.
There are three broad criteria for selecting the number of factors. A first criterion is to keep those factors with an eigenvalue exceeding one. This criterion suggests keeping seven factors as an upper boundary. The second criterion, the Scree test, involves plotting the eigenvalues and checking for breaks between those with large and small eigenvalues. This test suggests keeping three factors. A third criterion is to keep those factors that account for a minimal proportion of variance, usually set at 0.10. This criterion suggests keeping two or three factors. The proportional value of the third factor lies in all cases close to (or above) 10 per cent, so the decision is to keep three factors for all years. Furthermore, based on the “factor loadings”, the covariances between the variables and the latent factors, a common model was chosen where the first 21 variables (from “unpaid bills” to “debts”) have a high loading on a first factor, which would then reflect the “material deprivation” of the household. The nine dichotomous variables that describe the social and cultural participation of each individual have a high loading on the second factor denoted “social deprivation”. Finally, the third factor denoted “individual psychological health” includes the nine psychological-health variables. Note that this classification resembles that of Guio (2005).
Whether or not one is a member of a social or cultural organisation, as well as whether or not one is satisfied with one's training or education did not have a clear relation with any of the three factors in any year. This of course does not mean that these variables are not related to poverty, but only that they are not attributable to just one dimension. These variables were thus excluded from the confirmatory model.
The base model was then applied to the same PSBH data using confirmatory unweighted least squares FA. Table III shows that the fit of the base model seems reasonable, but not marvellous. One should however keep in mind that these results are inconsistent with the use of polychoric correlations.
Table III also shows the covariances between the three factors underlying poverty. The three factors have positive correlations, indicating that the arrears effectively cumulate.
5.2 Cluster analysis
Next, CA was applied to the factor scores generated in the first step. The goal of CA is to surface latent groups in the sample where the average within-groups distance is the smallest, while the average between-group distance is the largest. The procedure starts by considering each observation as a separate group and then iteratively merges the groups closest together, until the whole sample ends up in one group. Each grouping comes with a loss of information and one should choose that number of clusters that minimizes information loss, while maximizing the difference between the clusters. The information loss is described by the pseudo-R 2 and pseudo-t, and the number of clusters to be retained is one higher than that number where these variables are maximal. The pseudo-F equals the between-cluster variance divided by the within-cluster variance, and the number of clusters should be set such as to maximize this pseudo-F. If these three indicator variables suggest retaining a different number of clusters, then the lowest number of clusters is chosen. Among these, all clusters containing <1 per cent of the sample are treated as outliers, and thus ignored. Table IV describes the four variables.
Table IV shows that the optimal number of clusters equals four for most years. Only for the years 1996 and 1998 it is three and for 1997 it is five. Those clusters that are not outliers are interpreted on the basis of the mean factor scores. A significant positive mean factor score on all dimensions of poverty means that the cluster is considered poor. The t-tests of the three factor scores are denoted by t(f1) to t(f3) in Table V and reflect the hypothesis that the average factor scores are equal to zero.
The results for the year 1994 show four clusters, two of which (3 and 4) contain outliers. Remain two clusters of 11 and almost 88 per cent of the sample. For the smallest cluster the average scores are significant, suggesting that the individuals in this cluster have arrears on all three dimensions of poverty.
Of the four clusters that describe the situation in 1995, two contain outliers. Remaining two clusters that are unequal in size. The significant average factor scores of the smallest cluster of 9.6 per cent of the sample show that individuals in this cluster are considered poor. The results for 1996 show the same pattern, with a cluster of 11 per cent of the sample containing poor individuals.
The optimal number of clusters in the 1997 results is five, of which one contains outliers. The first of the remaining clusters of about 86 per cent of the sample has significantly negative average factor scores. The individuals in this cluster therefore are not poor. The individuals in the second cluster have an arrear in their material deprivation (f1) and social deprivation (f2), but not in their psychological health (f3). We thus cannot conclude that these people are poor. The individuals in the third and fourth cluster, together 11.37 per cent of the sample, show a consistent arrear on all three factors and thus are poor.
The situation in 1998 again is straightforward. One of the three clusters contains outliers, and the average positive factor scores of the smallest cluster of the remaining two (11 per cent of the sample) suggest that the individuals in this cluster are poor.
The results of 1999 again show four clusters, of which the fourth contains outliers. The first cluster of the remaining is the largest and contains non-poor individuals. The average factor scores in the remaining two clusters (together 11.44 per cent of the sample) are positive, so these individuals are poor.
The results of 2000 show four clusters, of which one contains outliers. The largest of the remaining clusters contains the non-poor individuals. The results of the second cluster are inconsistent, and the third cluster (of about 9 per cent of the sample) contains the poor individuals.
The conclusion is therefore that multi-dimensional poverty can be surfaced from the Belgian PSBH dataset. In the years between 1994 and 2000, poverty lies around 11 per cent of the sample, with two negative outliers of 9 per cent in 1995 and 2000.
6. What causes financial and multi-dimensional poverty?
Table VI compares the incidence of multi-dimensional and financial poverty.
That the financial poverty risk is always, somewhat, higher than multi-dimensional poverty risk is not surprising, because the former reflects relative poverty while the latter is based on absolute deprivations. A more detailed comparison between the two measures of poverty would be like comparing the proverbial apples and oranges. Instead, we attempt to surface the causes of multi-dimensional poverty, including financial poverty. This requires a longitudinal analysis, for causes should occur before consequences. The starting point of the analysis is a non-poor individual at time t. What is the probability that this individual will “survive” until t + z, and become poor only then? Survival analysis is a class of methods that study the occurrence and timing of events, in this case the probability that an individual falls into poverty at a certain time, given that he or she did not do so before. As events can occur continuously but are observed on regular intervals, we use a complementary log-log specification. An attractive characteristic of this model is that a one-unit increase in covariate x i results in a 100(exp(β i )–1) percentage change of the probability of falling into poverty. Table VII shows the estimation results.
The model I separates the effects of the consecutive years. The first column to the right of the variable “year” therefore only contains an indicator for the level of significance of the duration variable. The intercept in model I is the estimator of α7, the log-odd that an individual is observed to have become poor after six years, assuming all other variables equal to zero. The hypothesis that the estimators αx−α7 (reflecting the effect of the consecutive years in difference to the seventh year) differ from zero should be rejected, so the other models assume a linear or quadratic effect of time. Model II shows that the probability that an individual who previously was not poor is observed in poverty, decreases by 100(exp(0.17)–1) = 15 per cent per year. The other models show that this decrease slows down in later years.
The models I and II show that there is no significant difference between males and females in their probability of falling into poverty. Additional analyses showed that the overall probability of being poor differs, but that this difference is captured by the other explanatory variables. Models III-VI thus exclude gender as an explanatory variable.
Models II-IV support the hypothesis that financial poverty increases the probability to fall into multi-dimensional poverty. This is not the case in model V, which will be explained later.
The various models show that there are several factors that increase the probability of falling into poverty. These are (1) not having a job, (2) not having the Belgian nationality, (3) having a poor health or a disability, (4) being lower educated or (5) living with a poor individual in the same household. The effect of marital status and region on the probability of being observed as poor will be discussed in detail later. Finally, the models I-III show that the probability of falling into poverty decreases with age. Models IV and V show that this negative effect diminishes with increasing age.
Which of the three dimensions underlying multi-dimensional poverty determine most the probability that an individual effectively falls into poverty? To answer this question, the individual scores of the three dimensions were included in the regression. The estimator for financial poverty now becomes insignificant, which is due to the inclusion of the factor “material deprivation” in the model. Note the important effect of psychological health on poverty. Indeed, an increase of the score of the first factor (material deprivation) and second factor (psychological health) with one unit increases the hazard of falling into poverty by, respectively, 14.6 and 69.3 per cent. Even though the title of this paper seems exaggerated, the conclusion is that the poverty risk of an individual seems primarily determined by his or her psychological health.
Finally, we take a closer look at the categorical variables that reflect marital status and region of the individual. The models I-V assume a linear effect of these variables, which is obviously not very informative. The model VI is the same as model IV, but the variables “marital status” and “region” are dummy-coded. The dummy codes of the two three-level variables cannot be estimated in one model, but as the dummies are mutually excluding, the two models only differ in their intercept and the estimators of the dummies. So model VI is a compound model, and the value of the intercept is therefore not presented. It shows that those living in couples have the lowest probability, and those being divorced or widowed have the highest probability of falling into poverty. Furthermore, those living in Brussels or the Walloon region have a higher probability of falling into poverty than those living in Flanders.
7. Conclusions
This paper presents a multi-dimensional measure of poverty. By the choice of the underlying variables, it reflects an absolute notion of poverty. First, a common model is applied by confirmatory factor analysis. CA is then applied to separate the multi-dimensional poor. The advantages of this method are that all information is used to disentangle the poor from the non-poor, that categorisation errors are avoided by using tetrachoric and polychoric correlations and that dimensions of poverty are defined using the correlations between deprivations.
The proposed method has been applied to the PSBH dataset for Belgium for the years between 1994 and 2000 and reveal three dimensions of poverty: “material deprivation”, “social deprivation” and “psychological health”. Between 9 and 11 per cent of the representative sample of individuals show an arrear on all three dimensions and are therefore multi-dimensional poor.
Finally, the possible causes of multi-dimensional poverty are surfaced by estimating a discrete duration model. This includes not having a job, not having the Belgian nationality, having a poor health or a disability, being lower educated, experiencing financial poverty, being divorced or widowed or living in the Walloon or Brussels regions.
Notes
- The SAS-macro POLYCHOR (version 1.3), was used and adjusted to account for cross-sectional weighting of the indicators.
- Presenting and discussing all results is impossible here. However, all results and outputs are available from the author upon request.
Table IThe indicators and the base-model
Table IIEigenvalues of the exploratory factor analysis
Table IIIFactor loadings
Table IVParameters for the choice of the number of clusters
Table VSize and content of the clusters
Table VIMulti-dimensional and financial poverty
Table VIIDuration model (complementary log-log probit) of the probability that a non-poor individual after one to seven years becomes poor, given that it did not happen before
References
Brandolini, A., D’Alessio, G. (2000), "Measuring the well-being in the functioning space", mimeo, .
Chakravarty, S., Mukherjee, D., Ranade, R. (1998), "On the family of subgroup and factor decomposable measures of multidimensional poverty”, in", in Slottje, D. (Eds),Research on Economic Inequality, Vol. 8 pp.13-30.
Coenders, G., Saris, W. (1995), "Categorization and measurement quality: the choice between Pearson and polychoric correlations”, in", in Saris, W., Münnich, A. (Eds),The Multitrait Multimethod Approach to Evaluate Measurement Instruments, Eötvös University Press, Budapest, pp.125-44.
Dekkers, G. (2003), Financial and Multidimensional Poverty in European Countries: Can the Former be Used as a Proxy for the Latter?, Centre d’Etudes de Populations, de Pauvreté et de Politiques Socio-Economiques (CEPS), Differdange, IRISS Working Paper Series 2003-13, .
Dekkers, G. (2004), "Entre Pauvre et se Sentir Pauvre. Mésure de la Pauvreté Multidimensionelle d’après les données du PSBH”, in", in Doutrelepont, R., Mortelmans, D., Casman, M.-Th. (Eds),Onze Ans de Vie en Belgique: Analyses socio-économiques à partir du Panel Démographie Familiale, série Sience et Société, Politique Scientifique Féderale/Academia Press, Brussels, pp.143-66.
Guio, A.-C. (2005), "Material deprivation in the EU", Statistics in Focus, European Communities, Brussels, No. 21, .
Halleröd, B. (1995), "The truly poor: direct and indirect consensual measurement of poverty in Sweden", European Journal of Social Policy, Vol. 5 No.2, pp.111-29.
Knol, D., Berger, M. (1991), "Empirical comparison between factor analysis and multidimensional item response models", Multivariate Behavioral Research, Vol. 26 No.3, pp.457-77.
Layte, R., Maître, B., Nolan, B., Whelan, C. (2001), "Persistent and consistent poverty in the 1994 and 1995 waves of the European Community Household Panel Survey", Review of Income and Wealth, Vol. 47 No.4, pp.427-49.
Mislevy, M. (1986), "Recent developments in the factor analysis of categorical variables", Journal of Educational Statistics, Vol. 11 No.1, pp.3-31.
Muffels, R. (1993), "Deprivation standards and style of living indices”, in", in Berghman, J., Cantillon, B. (Eds),The European Face of Social Security, Avebury, Aldershot, .
Muthén, B., Christoffersson, A. (1981), "Simultaneous factor analysis of dichotomous variables in several groups", Psychometrica, Vol. 46 No.1, pp.407-19.
Nolan, B., Whelan, C. (1996), "Measuring poverty using income and deprivation indicators: alternative approaches", Journal of European Social Policy, Vol. 6 No.3, pp.225-40.
Tsakloglou, P., Papadopoulos, F. (2002), "Aggregate level and determining factors of social exclusion in twelve European countries", Journal of European Social Policy, Vol. 12 No.3, pp.211-25.
Townsend, P. (1979), Poverty in the United Kingdom, Penguin, Harmondsworth, .
Townsend, P. (1993), The International Analysis of Poverty, Harvester Wheatsheaf, New York, NY and London, .
Tsui, K.-Y. (2002), "Multidimensional poverty indices", Social Choice and Welfare, Vol. 19 No.1, pp.69-93.
Whelan, C., Layte, R., Maître, B. (2002), "Multiple deprivation and the persistence of poverty in the European Union", Journal of European Social Policy, Vol. 12 No.2, pp.91-105.
Whelan, C., Layte, R., Maître, B., Nolan, B. (2001), "Income, deprivation and economic strain", European Sociological Review, Vol. 17 No.4, pp.357-72.
About the author
Gijs J.M. Dekkers graduated cum laude in economics at Maastricht University, and holds a doctorate in social sciences from Tilburg University. He currently works at the Belgian Federal Planning Bureau in Brussels. He is also a research associate to the Centre for Sociological Research, Katholieke Universiteit Leuven and he is external professor at the IUP Gestion du Patrimoine, Université Dauphine, Paris. His research interests include microsimulation, income redistribution and its decomposition, poverty, and pensions. Gijs J.M. Dekkers can be contacted at: gd@plan.be