Professional Documents
Culture Documents
SSRN Id2276306
SSRN Id2276306
Ive Marx
DISCUSSION
Lina Salanauskaite
Gerlinde Verbist
May 2013
Forschungsinstitut
zur Zukunft der Arbeit
Institute for the Study
of Labor
The Paradox of Redistribution Revisited:
And That It May Rest in Peace?
Ive Marx
CSB, University of Antwerp
and IZA
Lina Salanauskaite
CSB, University of Antwerp
Gerlinde Verbist
CSB, University of Antwerp
IZA
Phone: +49-228-3894-0
Fax: +49-228-3894-180
E-mail: iza@iza.org
Any opinions expressed here are those of the author(s) and not those of IZA. Research published in
this series may include views on policy, but the institute itself takes no institutional policy positions.
The IZA research network is committed to the IZA Guiding Principles of Research Integrity.
The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
organization supported by Deutsche Post Foundation. The center is associated with the University of
Bonn and offers a stimulating research environment through its international network, workshops and
conferences, data service, project support, research visits and doctoral program. IZA engages in (i)
original and internationally competitive research in all fields of labor economics, (ii) development of
policy concepts, and (iii) dissemination of research results and concepts to the interested public.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available directly from the author.
IZA Discussion Paper No. 7414
May 2013
ABSTRACT
There is a long-standing controversy over the question of whether targeting social transfers
towards the bottom part of the income distribution actually enhances or weakens their
redistributive impact. Korpi and Palme have influentially claimed that “the more we target
benefits at the poor, the less likely we are to reduce poverty and inequality”. The basic
empirical underpinning of this claim is a strong inverse relationship at the country level
between social transfer targeting and redistributive impact. We show that this no longer holds
as a robust empirical generalisation. The relationship between the extent of targeting and
redistributive impact over a broad set of empirical specifications, country selections and data
sources has in fact become a very weak one. For what it matters, targeting tends to be
associated with higher levels of redistribution, especially when overall effort in terms of
spending is high. We try to make substantive sense of this breakdown of the originally
established relationship by focusing on two questions: first, what has changed in the
countries originally included in the study and, second, what is different about the countries
now additionally included in the analysis?
Corresponding author:
Ive Marx
Herman Deleeck Centre for Social Policy
University of Antwerp
Sint-Jacobstraat 2
2000 Antwerp
Belgium
E-mail: ive.marx@ua.ac.be
*
This study is part of the EU FP7 Funded GINI project (http://www.gini-research.org). We are very
grateful for the support of many individuals and institutions. In particular, our thanks go to Lane
Kenworthy for providing us with the code used in his own study and thus facilitating the starting point
of this analysis. Earlier versions of this paper have benefitted from comments during presentations at
the FISS Conference (June, 2012), the IARIW Conference (August, 2012), and several seminars. We
particularly thank Joakim Palme for comments.
1. Introduction
There is a long-standing controversy in the academic literature over the question of whether
targeting benefits towards the bottom part of the income distribution actually enhances or
weakens their redistributive impact. Korpi and Palme have influentially claimed that “the
more we target benefits at the poor, the less likely we are to reduce poverty and inequality”.
The basic empirical underpinning of this claim is a strong inverse relationship at the country
level between social transfer targeting and redistributive impact.
This paper shows that this key finding no longer holds as a robust empirical generalisation.
The relationship between the extent of targeting and redistributive impact over a broad set of
empirical specifications, country selections and data sources has in fact become a very weak
one, suggesting that the extent of targeting per se may not matter anymore as much as we
have assumed since Korpi and Palme.
Clearly, this issue is actually of more than academic importance. In its 2011 report on
inequality “Divided We Stand”, the OECD calls for “well-targeted income support policies”,
however without clearly specifying what form and strength this should take. By the same
token, the European Commission (2013) has launched a ‘Social Investment Package’ also
calling for better targeting and conditionality: “Support should be better targeted to those in
need at the times they need it”. Organisations like the IMF and the World Bank have long
advocated targeted benefits, specifically in the form of means-tested social safety nets. The
coming years will be marked by continued budget austerity in many countries. In the context
of rising demands on welfare systems, the issue of targeting is likely to become even more
relevant.
This paper proceeds as follows. We first summarize the story so far, devoting considerable
attention to the original Korpi and Palme thesis. We then discuss the more recent literature,
most notably recent re-examinations suggesting that targeting may not be so bad after all. We
contextualize the discussion about the particular link between targeting and redistribution in
the broader puzzle of egalitarianism. After discussing methodological and measurement issues
we move on to the empirical part, first presenting the main results and then robustness of
these results for variations in measurement and data. In the discussion part we try to make
sense of the results, focusing on two questions: a) what has changed in the countries originally
included in the study; b) what is different about the countries now additionally included in the
analysis? In a final section we conclude and set out some directions for further research.
Diametrically opposed views still exist on this issue. On the one side there are those who
belief that a welfare state can only fight poverty effectively and efficiently (i.e. cost-
3
effectively) when benefits are mainly targeted to those most in need, i.e. when benefits are
selective. Goodin and Le Grand (1987: 215): “the beneficial involvement of the non-poor in
the welfare state is not merely wasteful – it is actually counterproductive. The more the non-
poor benefit, the less redistributive the impact of the welfare state will be.”
Economists have moreover claimed that the budgetary return to targeting also entails an
economic advantage. Lower public expenditures, after all, imply lower taxes, which in turn
are supposed to be conducive to economic growth. Economic growth, the argument proceeds,
benefits the poor directly (although not necessarily proportionally so) and increases at the
same time the fiscal base for redistributive policies.
This view of selectivity has never been commonly shared. Two sorts of arguments underpin
this more critical stance.
First, there are technical considerations. Van Oorschot (2002) sums up the most important
dysfunctions of means-testing. First, these include higher administrative costs. Establishing
need or other relevant criteria require monitoring, whereas universal benefits allow for less
complex eligibility procedures. Furthermore, means tested benefits are subject to higher non-
take up, partly because of stigmatization issues. Finally, targeted benefits can give rise to so-
called dependency traps, where benefit recipients have little incentive to take up work or
increase working hours because this would entail loss of benefits.
The juxtaposition outlined above forms the starting point for Korpi and Palme. In their 1998
article, they employ a somewhat more complex typology, but their classification of welfare
states is essentially based on the dimension universalism versus selectivity. Based on data
relating to institutional characteristics of welfare states on the one hand, and data relating to
observed income distributions and financial poverty on the other, they conclude that more
selective systems, paradoxically, have a smaller redistributive impact than universal systems
offering both minimum income protection as well as income security and cost compensations
(for children) in a broader sense. Korpi and Palme find that this relationship is mediated by
the relative size of available means for redistribution. In essence, selective systems are
generally smaller systems, and for that reason less redistributive, despite their design to that
effect. To be fair, Korpi and Palme do not go as far as saying that the more universal system
are the more redistributive they will be. But they do say that strong targeting implies weak
redistributive outcomes.
4
Korpi and Palme’s main conclusion went relatively uncontested for a while although some
scholars expressed reservations because of the rather rudimentary character of the research
methods (Bergh 2005). The degree of redistribution, for example, is measured by comparing
the actually observed income inequality or at-risk-of-poverty rate with a rather
unsophisticated ‘counterfactual’ distribution. In theory this counterfactual ought to accurately
reflect the income distribution that would prevail in the absence of social transfers. However,
the construction of this counterfactual is hampered by theoretical and practical problems. In
most cases, including in Korpi and Palme’s paper, pre-transfer income is simply calculated by
deducting observed social transfers and re-adding observed taxes. Full abstraction is thus
made of any behavioural effects which a change in tranfer/tax regime would entail. While
patently less than perfect, the reality is that no satisfactory method exists to adequately model
such behavioural effects, except for very specific measures and marginal deviations.
Another critique was formulated by Moene and Wallerstein (2001; 2002) who argued that
analyses of redistribution need to be done at a more disaggregated level than ‘the welfare
system’ because the determining redistributive principles may differ substantially for, say,
unemployment, health care or pensions. Some schemes may rest heavily on the insurance
principle, while others may put more weight to the need-principle. Thus universality and
selectivity can coexist within one system. Yet Moene and Wallerstein (2001; 2002) also
conclude that universal provisions provoke the largest political support because of the higher
chance of middle class citizens to become a beneficiary. Some opinion based studies also
confirm that universal welfare schemes enjoy broader support (Forma, 1997; Kangas, 1995)
and that universal schemes are more politically robust in times of austerity (Nelson, 2007;
Ferrarini, Nelson and Höög, 2013). It is plausible, however, that public opinion is influenced
by the institutional set-up of a welfare state and so the causality cannot be seen as running one
way (see Larsen, 2008; McCarthy and Pontusson, 2009). Also, one should not overestimate
the effect of public opinion on social policy (e.g. Brooks and Manza, 2006), as public policy
is also influenced by resource mobilization, path dependence, political framing etc. There is
other evidence in support of universalism, for example on the basis of studies that look at
particular programmes. Corak, Lietz and Sutherland (2005) - in a study based on incomes at
the end of 1990’s or beginning of 2000’s - for example find that universal child related
benefits – not those targeted at the poorest – provide better protection against poverty. Their
conclusion that ‘targeting within universalism’, in Skocpol’s (1991) words, yields the best
outcomes is echoed by Van Mechelen and Bradshaw (2013), Van Lancker et al. (2012), Figari
et al., 2011. Brady et al. (2012), however, conclude in favour of universalism if it comes to
anti-poverty policies for single mothers.
Returning to focus of this paper, some recent studies have claimed that the cross-country
relationship between overall targeting and obserserved redistributive impact has weakened, or
even reversed over time. Kenworthy (2011) reproduces and updates Korpi and Palme’s
analyses, which related to the situation in 11 countries as of 1985. Kenworthy’s findings
confirm that countries with more universal benefits achieve more redistribution (measured in
the size of redistributive policies in the budget) for the period 1980 to 1990. By 1995, the
image becomes less clear. Data for 2000 and 2005 seem to indicate that there is no longer any
association (either positive or negative) between the two variables. Evidently, the findings are
based on a small number of cases (10 countries), which make them particularly sensitive to
outliers. A trend towards more targeting in Denmark, in conjunction with an evolution
towards more universal benefits in the US, is largely responsible for the shift in conclusions.
Moreover, the new findings may be driven to some extent by the growing share of pensions in
social spending. However, analyses on an alternative dataset, controlling for pensions and
5
featuring a larger number of countries, suggest that as of the mid 2000s, universalism is
negatively associated with redistribution.
Kenworthy refers here to an earlier study by Whiteford which shows that ranking method
matters, a more than technical measurement issue to which we return in this paper. In the
studies by Korpi and Palme and, as it is said to be a replication, Kenworthy, the calculations
to establish the degree of targeting are based on households’ position in the income
distribution before taxes and after transfers (i.e. gross income). Whiteford uses post
tax/transfer income, i.e. disposable income, as the ranking measure. We return to the issue of
the ranking income concept in the section on “Measuring targeting and redistribution”, and
will test the sensitivity of the results for using different ranking income concepts. Whiteford
finds that universalism correlates negatively with redistribution. Kenworthy writes about this:
“This by no means settles the question, but it does suggest additional reason to rethink the
notion that targeting is an impediment to effective redistribution” (Kenworty, 2011:58). This
paper takes the quest further from there, addressing measurement issues in more depth,
expanding the number of countries included in the analysis and checking for robustness
against selected methodological choices, such as data source or income concepts.
Before we move to the empirical part it may be useful to recall that the issue of universality
versus targeting makes up only one element in a wider ‘puzzle of egalitarianism’ (Alvarez
2001). After all, at the country level we are confronted with several empirical relationships
that beg for a more thorough understanding. Lindert (2004) evokes the ‘Robin Hood Paradox’
“in which redistribution from poor to rich is least present when and where it seems most
needed”.
The connection between universality, the level of social expenditure and redistributive impact
is part of a wider puzzle and it is important to be aware of this. For one, we know there also to
be a strong relationship at the country level between wage (or market income) inequality and
social expenditure. In other words, it is countries with egalitarian wage structures that tend to
have universal welfare systems, generous benefits and, as a consequence, high social
expenditures. This connection is again contra-intuitive, because at first sight, one would
expect the opposite relation, namely that a greater wage disparity would require more
redistribution, and, therefore, higher social expenditures. Hence, the causal chain may well
start with institutions and policies shaping income distributions before taxes and transfers. For
example, as argued by Bradley et al. (2003), the causal relationship between market inequality
and redistribution occurs indirectly and is mainly due to the strong influences of labour unions
and left parties. But it may also run in other ways.
Let us briefly discuss these alternative causal narratives. First, the direction of causality may
go from an extensive welfare state to a condensed waged distribution. This is the line
followed by Beramendi Alvarez (2001), who has argued that second-order effects of social
expenditure are a large part of the explanation of the ‘puzzle of egalitarianism’. First order
effects of redistribution on inequality consist of direct income transfers from high-income to
low-income households, through taxes, social security or social assistance. But second order
effects are equally important: the higher taxes and transfers of large welfare states influence
labour supply in such a way that a more condensed wage distribution results. High-wage
6
earners substitute monetary income for leisure in response to taxes, while generous benefits
reduce labour supply among those commanding low wages (through higher reservation
wages). These second order effects may be highly contingent upon national institutions,
particularly with regard to funding of the welfare state, the level of wage bargaining and fine
details of institutional design.
Second, the causal mechanism between redistribution and inequality may run in the opposite
direction. A highly unequal distribution of market incomes may make it politically and
technically more difficult to redistribute income. McCarty and Pontusson (2009) review a
number of political economy theories with regard to voter behaviour under different
conditions of economic inequality. The so-called median voter models assume that changes in
the income distribution lead to a shift in the preference of the median voter, or the ‘political
middle’. Moene and Wallerstein (2001, 2003) argue that under conditions of rising income
inequality, the median voter has a preference for reduced expenditure on insurance and social
spending. Earlier Meltzer and Richard (1981) formulated an opposing hypothesis, predicting
that rising income inequality leads to a shift in preferences of the median voter towards more
redistribution.
The evidence is quite mixed. Kenworthy and Pontusson (2005) find empirical support for the
Meltzer and Richard thesis. “In contrast to widespread rhetoric about the decline of the
welfare state, redistribution tended to increase in response to the rise in household market
inequality. And it did so in proportion to the degree of increase in inequality, producing a
strong positive association between changes in market inequality and changes in
redistribution.” Milanovic (2000) finds a consistent association between gross household
income inequality and more tax/transfer redistribution in a set of 24 democracies in the period
of the mid seventies-mid nineties. More recently Olivera (2012), performing an analysis on a
pool of 33 European countries, finds that inequality increases the demand for redistribution
and that increases in income inequality stimulate the demand for redistribution. Yet the
empirical evidence varies and some studies arrive at opposite conclusions (Iversen and
Soskice: 2006, 2009; Finseraas 2009; McCarty and Pontusson 2009; Lupu and Pontusson
2011; Toth, Horn and Medgyesi, 2013).
As McCarty and Pontusson (2009) note, one clear complication is that the majoritarian
assumption underlying the median voter models is not universally applicable. In many
contexts, a vast number of parties (including trade unions, employers’ organisations) compete
for political influence. The partisan politics theory assumes that, rather than moving all parties
either to the left or the right, income inequality can cause polarization of the electorate. As the
political spectrum widens, the outcome depends largely on the extent to which low-income
groups are mobilized, in terms of election turnout and union density. In turn, social security
arrangements can strengthen trade unions, particularly in so-called Ghent countries where
they are involved in the provision of unemployment benefits (Van Rie et al., 2011)
Furthermore, the recent attention in the literature to insider-outsider conflicts and divergent
interests within ‘Labour’ or ‘the Left’ are of key importance.
Third, causality between equality and redistribution may run in both directions, in a process of
mutual reinforcement. This argument has been developed by Barth and Moene (2009) in a
recent NBER paper. They argue that a more equal wage distribution leads to welfare
generosity through a process of political competition. In turn, more income redistribution
produces more equality. The authors hypothesize that this multiplier operates mainly through
the bottom of the income distribution: the amplification occurs where wages at the bottom of
7
the distribution are compressed, not where higher incomes are compressed. The hypothesis
finds empirical support in their analyses on 18 OECD countries over the years 1976 to 2002.
Finally, an extensive welfare state, as well as a limited degree of wage inequality may both be
the result of variables that determine both. As Atkinson (2000) suggests, countries may be
characterised by notions of equity that are widely shared within any society, but that differ
across societies. A society in which the value of solidarity is widely shared may
simultaneously support pay norms, collective agreements and adequate minimum wages, as
well as quasi-universal and generous benefits.
Thus, variables like wage dispersion, primary income inequality, generosity, universality,
level of expenditures and disposable income inequality make up a complex web of
interrelations of which the causalities can run in different directions. In this paper, the primary
focus lies on the relationship between the factors of universality, generosity and income
inequality.
Targeting, redistribution and generosity are key concepts in this paper. The aim of this section
is to address the conceptual clarity of these terms, as misunderstandings related to their
interpretations may easily arise (e.g. van Oorschot, 2002, p. 173). For instance, targeting is
often equated with means-testing although it does not necessarily imply a means test, as other
eligibility criteria (e.g. family composition) can be established to channel benefits to specific
groups (e.g. lone parents). By contrast, ‘universal’ benefits are aimed at broad segments of the
(national) population. Still, it should be noted that universal benefits are rarely truly universal,
as they often apply for instance a residency criterion, which can be more or less strict.
Moreover, whether benefits are flat-rate or earnings-related, is a question that is closely linked
but distinct from universality or targeting.
Korpi and Palme employ three aspects to classify transfer systems: a) the basis of entitlement;
b) the benefit level principle and c) the form of governance, particularly the extent of
employer-employee corporation. In their classification targeted models have two
distinguishing features: the basis of entitlement is proven need and benefits are there to
provide a minimum income.
However, the actual operationalization they use in their 1998 article is much more
straightforward and simple. It is also for that reason contestable on a number of grounds. In
their 1998 ASR article they do not build on institutional indicators to gauge the level of
targeting within tax/transfer systems. The extent of targeting is measured on the basis of a
single outcome indicator, the targeting coefficient, which reflects effective redistributive
outcomes rather than redistributive intentions, as we will argue below. This means that
‘targeting’ here is interpreted as social transfers being more beneficial for lower incomes,
irrespective whether this comes about because of system characteristics like means testing
(“income selectivity”) or providing benefits for specific categories (“categorical selectivity”).
Basically, this means that we look at the position of the beneficiaries in relation to the median
(voter).
8
In our analysis, targeting is measured in two ways: on the one hand we use the concentration
coefficient of transfers in the same way as Korpi and Palme (1998) do. On the other hand, we
also express it as a share of transfers going to the bottom quintile.
1
Though of course this does not mean that all concentration coefficients of zero correspond to equal absolute
amounts over the income distribution.
9
Concentration coefficient
-1 -0,8 -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 1
1
B: Weakly pro-poor
G
A: Strongly pro-poor
i
0,5
n
i
C: Pro-rich
0
Source: own presentation.
As social transfers are composed of various transfer types (i.e. old-age pensions, family
benefits, etc.), we try to identify the distributional properties and contribution of these income
sources to the overall concentration coefficient. Following Kakwani (1977), we apply a factor
decomposition analysis of the concentration coefficient: the concentration coefficient of total
transfers (CT) can be decomposed as the sum of the concentration coefficients (Ci ) of the
different transfer categories i weighted by their share si in total transfers s:
n
∑sC
si
CT = Ti
i =1
Quintile distributions are based on five equal-sized population groups which are divided
according to their income. Income units are ranked from low to high income. A comparison of
the share of social transfers going to the bottom quintile with the corresponding concentration
coefficient indicates how the concentration coefficient comes about: is it driven by targeting
towards the bottom quintile (the poor), or are rather by patterns higher up the income ladder?
Note that we use the term ‘targeting’, which suggests that outcomes are due to the
characteristics of the system, but this need not be the case. Moreover, the outcomes of a
system are highly dependent on the characteristics of the underlying population, in terms of
socio-demographic characteristics, income inequality, composition of income, etc. If, for
instance, a benefit is designed in such a way that all children are eligible, but all children are
situated in the bottom quintile, then this policy measure may appear as very targeted in its
outcomes, even though its design may not include any means-testing or needs-based
characteristic. This means that strictly speaking we cannot derive from the concentration
coefficient or from the quintile distribution how pro-poorness of a transfer comes about.
10
• The ranking of beneficiaries in the income distribution. This means that the same
distribution of transfers will only result in different concentration coefficients if
recipients have a different ranking in the underlying income distribution (e.g. market
versus disposable income, see further).
The concept of redistribution refers to the impact of taxes and transfers on income inequality.
The impact on inequality is driven by the size of transfers and taxes, as well as by their pro-
poorness, i.e. whether these transfers are going relatively more to lower or higher incomes.
Redistribution is measured by the difference between the Gini coefficients with and without
tax-transfers relative to pre-transfer income; this corresponds in our analysis to the difference
of the Gini coefficients of market and disposable income relative to that of market income.
Alternative measures of redistribution can also be used. For example, Kenworthy and
Pontusson (2005) rely on the absolute difference between disposable and market income Gini
coefficients, as it allows capturing redistribution that is not affected by (e.g. time variant)
level changes of market income inequality. The redistributive effect of taxes, resp. transfer
alone can be the focus too and is measured as the difference of the Gini coefficients of gross
and disposable income relative to that of gross income, resp. the difference of the Gini
coefficients of market and gross income relative to that of market income. In addition, Jesuit
and Mahler (2010) suggest a measure of redistribution that captures the second-order effects
of public pensions. This measure addresses the issue of artificially highly unequal market
incomes, as elderly households relying on public pensions appear to be poor, though in the
counterfactual scenario of no state pension provision they would have relied on other sources
of income. Finally, the redistribution measures can be based on other than Gini indicators of
income inequality, e.g. percentile ratios as in Mahler (2008). We test the impact of some of
these alternative redistribution measures.
For indicating the impact of the size of transfers, we use the concept of generosity: how much
is spent on social transfers? Generosity is measured here by expressing average social
transfers as a share of average income (either market income, gross income or disposable
income). An alternative measure is also used for the sensitivity analysis: cash social spending
as percentage of GDP (for a further discussion on the pros and cons related to diverse
measures of welfare state generosity see e.g. Allan and Scruggs, 2004).
Overall, the building blocks of our analysis consist of some commonly used income concepts,
namely market income, social transfers, gross income and disposable income 2. These income
concepts are constructed in such a way that they cover exactly the same components as in
Korpi and Palme, unless specified otherwise 3.
2
Note that in the distributional analyses negative incomes are set to zero.
3
As we show in Appendix 1, seemingly the same income concepts, if taken from different data sources,
could have a different composition regarding individual components.
11
In brief, market income includes income from labour and capital, as well as mandatory
individual and occupational pensions (or further referred to as “occupational pensions”) 4.
Gross income is defined as market income plus social transfers plus net transfers between
households. Social transfers consist of the total of work-related insurance transfers, universal
benefits and social assistance benefits 5. Finally, disposable income is arrived at when
deducting taxes from gross income. These taxes refer to personal income taxes and social
contributions (excluding employer contributions).
For the distributional analyses, all relevant revenues are pooled at the household level and the
individual is the unit of observation. An equivalence scale is used to correct for household
size. We apply the same equivalence scale as in Korpi and Palme (1998), namely the square
root of household size.
We use different income concepts to rank equivalised incomes when calculating quintile
distributions and concentration coefficients. In a first instance we use the same income
concept to rank income units as the one used by Korpi and Palme (1998), as well as by
Kenworthy (2011), namely gross income, or ‘Who gets what after social transfers but before
taxes’. Of course, there are arguments to use other income concepts for ranking individuals.
The most obvious candidates are market income and disposable income, which are also the
income concepts used to calculate the redistributive effect in Korpi and Palme (1998) and in
this paper. With market income (or ‘Who gets what before taxes and transfers’), income units
are ranked according to the position they take in the hypothetical situation that there would be
no social redistribution. This effectively puts households that rely solely on transfers at the
bottom of the distribution. By definition the impact of the transfer is very large. This makes
more sense for people who fail to gain access to the labour market whereas they are supposed
to be economically self-reliant. They would probably have no other means of existence. But
this is a stronger assumption for the elderly who rely on public pensions. In the counterfactual
scenario they would probably have saved. Thus in calculations based on pre transfer rankings,
retired households count as extremely poor and they populate the very bottom of the
distribution. This probably overstates the impact of social transfers on the income distribution.
For this reason Whiteford (2010) and OECD (2008) use disposable income (i.e. ‘Who gets
what after taxes and transfers’) as the ranking measure. The drawback here is that the impact
of the welfare state may be underestimated. Some pensioners, for example, might have
occupied an entirely different position in the income distribution in the absence of a public
pension system. In our empirical application, we test the sensitivity of the results for using
either market or disposable income as ranking measure.
If taxes and transfers do not alter the ranking of income units, then concentration coefficients
will be the same for all three income concepts. In practice, income units do change ranks.
Especially the inclusion of social transfers (i.e. moving from market to gross income) causes
income units to change rank, implying that concentration coefficients can substantially differ
4
LIS has recently reclassified some income concepts, notably occupational pensions, which are now part of
social (security) transfers. We explore the implications of this income classification (change) in the
sensitivity part of our analysis, when we compare the “old” and the “new” LIS terminology.
5
Essentially, social transfers are equivalent to “social security redistribution” transfers based on the most
recent LIS terminology, except of the above mentioned mandatory individual and occupational pensions.
This implies, that the “old” LIS classification refers to higher market income, but lower social transfers in
comparison to the new terminology. The concept of gross income is not affected by this methodological
change.
12
when market income compared to gross or disposable income is used. Reranking due to taxes
(i.e. moving from gross to disposable income) is in general much smaller, implying that
concentration coefficients based on the ranking of either gross or disposable income will
probably not be very different.
As this article seeks to replicate the findings by Korpi and Palme and subsequent studies, we
aim for maximum comparability of concepts in the main analytical part. Here, we reproduce
the “original” relationship between targeting and redistribution as displayed in Korpi and
Palme (1998) and later on in Kenworthy’s (2011) study. Our selection of the countries covers
the “original” set, but also goes beyond it, as the LIS database has expanded its country
coverage substantially. Furthermore, for a better understanding of the driving forces of the
targeting and redistribution relation, we decompose the concentration coefficient over
separate transfer types, and we analyse how targeting and redistribution relate to the
generosity of social transfers.
We challenge most of the “original” methodological choices in the sensitivity analysis part:
country selection, data source, choice of the ranking income type and classification of social
transfers. A number of other sensitivity checks have also been made, but are not reported in
the main text of this study. For example, the impact of using alternative measures (as on
targeting) is noted in the footnotes with the corresponding graphs in Appendix 2. A few other
sensitivity tests, such as checking the influence of using a different equivalence scale, are not
reported due to their smaller influence on the main findings in comparison to the
aforementioned methodological choices.
4.4. Data
In the main analytical part, indicators are calculated on the basis of the micro survey data of
the Luxembourg Income Study (LIS), in line with the approach of Korpi and Palme (1998)
and Kenworthy (2011). We include the following countries, referring to 2004 (unless
specified otherwise between brackets after the country’s name): Australia (2003), Austria,
Belgium (2000) Canada, the Czech Republic, Denmark, Estonia, Finland, France (2005),
Germany, Greece, Hungary (2005), Ireland, Israel (2005), Italy, Luxembourg, the
Netherlands, Norway, Poland, Slovenia, Spain, Sweden (2005), Switzerland, United
Kingdom, United States. LIS tries to provide datasets that are as comparable as possible 6.
However, some issues remain.
For example, one of the major issues relevant for our study is the fact that not all LIS national
datasets include gross income amounts, implying that gross income in fact corresponds to
disposable income and that market income does not include taxes. This is the case for
Belgium, Greece, Hungary, Slovenia and Spain. Taxes are also only partially captured in the
French and Italian data. This implies that (part of) social transfers are net of taxes in these
6
More details on methodological aspects concerning LIS data can be found in Appendix 1.
13
countries, whereas in the other countries gross transfers (i.e. before deduction of taxes) are
used. This is particularly relevant for countries where transfers are subject to substantial
taxation, as is the case in the Nordic countries. It also means that the value (and distribution)
of available gross income is lower (different) for the countries having not full tax information
in the dataset. We tackle this “net/gross” issue in the sensitivity part of our analysis.
In the sensitivity analys, we also compare the main outcomes calculated on LIS data with the
outcomes computed on EU-SILC 2005 data, which covers 25 EU countries at the time, plus
Iceland and Norway (incomes refer to 2004). For a number of European countries (i.e.
Austria, Estonia, Greece, Spain, Ireland, etc.) the LIS dataset is actually derived from the
national EU-SILC surveys – seemingly, the most common type of LIS information source for
the European countries as of mid 2000’s.
When using EU-SILC data, we apply income definitions that are as close as possible to the
ones used by Korpi and Palme with the LIS dataset. Such alignment is needed as original
(EUROSTAT) income definitions of, for example, gross income are somewhat different from
those used in the LIS dataset. A more detailed explanation of differences and similarities
between the LIS and EU-SILC data is provided in Appendix 1.
5. Results
As our first aim is to replicate the earlier findings (see Appendix 2) for more recent data, we
start our analysis by using the same concepts, operationalisations and data (LIS) as Korpi and
Palme (1998) and Kenworthy (2011). We relate the concentration coefficient of social
transfers (with income units ranked according to gross income) to the redistributive effect of
taxes and transfers, as measured by the difference between the pre and post-tax/transfer Gini
coefficient (see Figure 2). We extend the country coverage in comparison to the previous
studies.
Let us first consider the two key variables in Figure 2 separately, starting with our measure of
targeting 7. All concentration coefficients are smaller than the Gini coefficient of gross
income, which means that all systems are progressive (i.e. pro-poor). There is however wide
variation in the spectrum of targeting-universality. Remember, that the smaller/more negative
the concentration coefficient is, the more targeted the transfers are whereas the closer the
concentration coefficient is to the Gini, the more universal transfers are. Australia, the United
Kingdom and, maybe somewhat surprisingly, Denmark have the most negative concentration
coefficients and can be characterized as strongly pro-poor. Negative concentration
coefficients are found in the majority of the countries, pointing to a substantial degree of
targeting. In nine countries though, such as e.g. Southern Europe, Austria, France, Poland or
7
Note that, as shown in appendix Figure A.5 of Appendix 2, concentration coefficients match quite closely to
an alternative measure of targeting: the share of transfers going to the bottom quintile.
14
Hungary, we find positive concentration coefficients, and thus weak pro-poorness of transfer
systems.
Figure 2. Concentration index (ranking by gross income) and redistributive impact, mid 2000s
50
FIN
SWE
DNK HUN
BEL
CZE NORDEU NLD POL
Redistribution index
LUX AUT
40
IRL SVN
FRA
AUS EST ITA
GBR
30
CAN ISR ESP GRC
CHE
USA
20
-.4 -.2 0 .2
Targeting (Concentration index, ranking on gross income)
Note: 1) Due to data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France. 2) The countries
included in Korpi and Palme (1998) are in grey font.
Source: own calculations using LIS data
Let us now consider the relationship between the degree of targeting and redistributive impact
– from the perspective of the x and y axes.
Especially noteworthy at the very left hand side of the x axis are Australia, the United
Kingdom and Denmark: they are all characterized by the strongest pro-poor benefit systems
of all countries 8. Yet the redistributive impact in Denmark appears much stronger. Similarly,
looking at the countries with still strong pro-poor spending (concentration indices between
about -0.3 and 0), the corresponding redistributive impact differs a great deal. Some of the
countries with the strongest redistributive tax/transfer systems are to be found here (Sweden
and Finland), together with some countries with the weakest redistribution (the USA, Canada,
Israel and Switzerland). In summary, no clear relationship is found between targeting and
redistribution among the countries with a negative concentration coefficient. A more clear and
8
Here and further on, the discussed country groups or specific value ranges of the
concentration/redistribution indices are estimated using a cluster analysis.
15
negative relationship, however, appears when focusing on the right hand side of the graph.
This concerns the countries with positive targeting coefficients and thus countries with weak
pro-poor spending: France, Italy, Spain, Austria, Poland and Greece.
Looking at this graph from the perspective of the y axis (the extent of redistribution achieved
by the tax/transfer system) essentially yields a similar picture of wide cross-country variation.
We also observe an extremely scattered range of values for the targeting measure in the
countries with moderately high levels of redistribution (from about 30 to 40 per cent reduction
of the post-tax/transfer Gini as compared to the pre-tax/transfer level). It ranges from the
strongest pro-poorness of almost -0.4 observed in Australia to weak pro-poorness in countries
like Italy or France.
In summary, our major finding – against the background of the earlier discussed literature - is
that actually there is no clear relationship between targeting and redistributive impact for
more recent years. This may partly be due to the fact that more countries are included in our
analysis. But also other factors may play, which will be put to the test in the sensitivity
analysis part of this paper.
Korpi and Palme (1998) have used a broad index of redistribution, as they look at the
difference between the Gini coefficients of market and disposable income, i.e. the combined
effect of taxes and transfers. However, for their measure of targeting, they only look at the
way transfers are distributed. Hence, we think that it would actually be more coherent to
measure redistribution as the difference between the Gini coefficients of market and gross
income (i.e. capturing the effect of transfers only), as is presented in Figure 3. By doing this
the initial relationship found by Korpi and Palme seems to re-emerge, though this association
is still a very weak one. Hence, we assert that the main conclusion from Figure 2 still holds:
the relationship between targeting and redistribution has become very weak. Throughout this
paper, we will continue to use the redistribution indicator that Korpi and Palme have used, as
they are our point of reference), and look at the sensitivity of outcomes of changing one
parameter at the time. But in subsequent studies on this topic we recommend to apply
coherence between the targeting and the redistribution concepts. Moreover, a comparison of
Figures 2 and 3 also indicates that the role of taxes in the redistribution process merits further
investigation, not only because the redistributive effect of taxes differs across countries, but
also because transfers are taxed in very different ways, with typically high burdens in the
Nordic countries and low or no taxes in other countries.
16
Figure 3. Measuring redistribution as the effect of transfers only
50
HUN
GINI:(market-gross)/market
BEL
POL
Redistribution index,
40
SWE FIN SVN
DNK CZE NOR AUT
DEU NLD LUX
FRA
30
IRL ESP GRC
GBR
CHE EST
AUS ITA
20
CAN ISR
USA
10
-.4 -.2 0 .2
Targeting (Concentration index, ranking on gross income)
Notes: due to data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France.
Source: own calculations using LIS data
As noted earlier, the high degree of targeting should not necessarily imply an extensive usage
of means-tested benefits (i.e. minimum income protection schemes) but rather reflect the
socio-demographic profile of beneficiaries. For example, a universal old-age pension system
may be characterised as highly targeted if the elderly disproportionately cluster at the lower
end of the income distribution. To better understand the main drivers of the concentration
coefficients’ values, we employ a factor decomposition analysis. Unfortunately, the
decomposition cannot be done for all countries, as often LIS data information is missing to
distinguish individual transfer categories.
First, we evaluate the individual input of social assistance benefits as opposed to other
transfers. This already gives a flavour of how the design of systems operates, as these social
assistance benefits are by default designed to target the most income vulnerable people. The
social assistance income category pools diverse assistance benefits, such as general social
assistance, but also assistance benefits in case of old-age, disability and various other
circumstances (e.g. family, education).
Second, we evaluate the individual input of three major categories of social transfers by need
function: old-age pensions, family benefits and a residual category of other active age
benefits. Old-age benefits consist of the following LIS variables: old-age insurance public
17
pensions, old-age universal pensions and old-age assistance pensions 9. Family benefits cover
maternity/parental wage replacement, family/child universal benefits and
family/maternity/child assistance income variables. The other active age benefits refer to a
range of diverse social security benefits, such as sickness wage replacement, disability or
survivor pensions, unemployment or education benefits, etc.
AUS BEL CHE CZE DEU DNK FIN FRA GBR HUN IRL ISR ITA NOR POL SVN SWE USA
Notes: due to data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France; Source: own
calculations using LIS data
The targeting degree of social assistance benefits is, as expected, much higher than that of
other social transfers: for all countries the concentration indices are lower (and always
negative) than of those of other social transfers (see Figure 4). The variation of targeting is
highly diverse, ranging from about -0.7 in Sweden to about -0.2 in Italy. This very strong
targeting in comparison to other social transfers is an important driver of overall targeting,
though much depends on the size of these transfers, as is shown in Figure 5. Even though
social assistance benefits are relatively small in all countries (from about 2% in Belgium and
Slovenia to about 38% in the UK), their contribution to the overall concentration index is not
negligible. In Ireland, for instance, social assistance benefits make up around 33% of total
social transfers, but they contribute for almost 80% to the concentration coefficient of total
transfers. Social assistance also contributes to more than half of the total concentration
coefficient in the United Kingdom and Slovenia. Overall, this signals that despite the
relatively small size (as for instance in the case of Slovenia), social assistance benefits play a
significant role in determining the level of the overall concentration index, and thus the
relationship between targeting and redistribution as portrayed in Figure 2.
Furthermore, we find a clear association between the targeting degree of social assistance
benefits and redistribution: the higher targeting of these benefits, the more redistribution is
achieved (Figure 6). Conversely, no association is found between targeting of other social
transfers and redistribution.
9
Occupational pensions are excluded here.
18
Figure 5. Social assistance: relative size and contribution to the concentration index
.8
.6
.4
.2
0
AUS BEL CHE CZE DEU DNK FIN FRA GBR HUN IRL ISR ITA NOR POL SVN SWE USA
Relative size of social assistance Relative contribution of social assistance
Notes: due to data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France; the ratios of
contribution are of minus sign in France, Hungary, Italy, Poland, Slovenia, and have been rescaled to represent
the share of total concentration coefficient without the directional effect10.
Source: own calculations using LIS data
Figure 6. Social assistance and the rest of social transfers: targeting and redistribution
50
FIN
SWE SWE FIN
DNK HUN DNK HUN
DEU BEL BEL
NOR CZE NOR DEU
CZE POL POL
Redistribution index
Redistribution index
SVN
40
40
IRL SVN
IRL FRA
FRA
AUS AUS ITA
GBR ITA GBR
30
30
ISR ISR
CHE CHE
USA USA
20
20
Notes: due to data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France;
Source: own calculations using LIS data
10
As social assistance benefits and other transfers have concentration coefficients with opposite signs in these
countries, the negative contribution of e.g. social assistance (i.e. -1.0639 in Hungary) is counteracted by the
positive contribution of other transfers (i.e. 2.0639 in Hungary) so that the total sum still adds up to 1. In
rescaling to the share without the directional effect, the contribution of social assistance in e.g. Hungary is
transformed to 0.34 relative contribution to the value of total concentration index.
19
Figure 7. Concentration indices of social transfers – decomposition by type
.4
.2
0
-.2
-.4
-.6
AUS BEL CHE CZE DNK FIN FRA GBR HUN IRL ISR ITA POL USA
All social transfers Old-age Family benefits Other active age benefits
A decomposition of social transfers by their need function is given in Figure 7 for those
countries where we can distinguish between old-age, family and other active age transfers.
The overall concentration index of social transfers is usually of the same sign as the
concentration index of old-age pensions, no matter of how targeted the other two types of
social transfers appear. The exception is the United States, where the concentration index of
social transfers is slightly negative, but the concentration index of old-age pensions is
positive, pointing to the stronger influence of the other active age transfer types. The degree
of targeting of old-age pensions is highly varied. For example, in Australia, Belgium,
Switzerland, and particularly Denmark, the old-age pensions are (much) more pro-poor in
comparison to the other categories. In the Czech Republic, Finland, the Great Britain and
Israel, at least one type of social transfers is more targeted at the poor than old-age pensions.
Finally, in the countries with positive targeting coefficients, namely France, Hungary, Italy
and Poland, old-age pensions are driving the weak overall pro-poorness of social transfers,
against the strong pro-poorness of family and other active age benefits. In other countries
weak pro-poorness of the transfers is rare, with exceptions in the United States for old-age
benefits and in Belgium for family benefits.
In Figure 8 we report the contribution of each transfer category to the total concentration
index of social transfers. This confirms that old-age pensions are the primary source for the
level and sign of the total concentration index in France, Hungary, Italy and Poland, all
countries with positive concentration index values. In the other countries, family and other
benefits together have a larger influence than old-age pensions, though with high variation.
For example, all three social transfer types are of equal importance in Austria, whereas the
contribution of the old-age pensions is by far the smallest one among the three types of social
transfers in Ireland and the United States.
20
Figure 8. Factor decomposition of CI of social transfers - contributions
3
2
1
0
-1
-2
AUS BEL CHE CZE DNK FIN FRA GBR HUN IRL ISR ITA POL USA
Figure 8 also highlights that often the distributional properties of family benefits are rather
different from old-age pensions, with Belgium being the prime example. Besides, family
benefits also have the widest spectrum of the targeting degrees, as shown in Figure 7. But
despite this wide range of targeting degrees of family benefits, we find a positive relationship
between targeting of family benefits and redistribution (see Figure 9, Panel A). No such
relationship could be traced for either old-age pensions or other active age benefits. This
implies that the relationship (or rather absence of it) in our Figure 2 is actually driven by the
distributional features of these two types of transfers (i.e. unemployment, disability, social
assistance, etc.) rather than benefits aimed at families with children.
Figure 9. Concentration index of social transfers (ranking on gross incomes) and redistribution index
50
FIN
(II) BEL DNK HUN DNK FIN HUN
NLD DEU BEL BEL
POL NOR CZE POL CZE POL Redistribution index
40
(I) IRL
IRL
ITA FRA FRA
AUS AUS ITA
GBR EST GBR AUS ITA
GBR
30
ISR ISR
CHE CHE ISR
USA USA CHE
USA
20
-.6 -.4 -.2 0 .2 .4 -.6 -.4 -.2 0 .2 .4 -.4 -.3 -.2 -.1 0
Targeting of family benefits Targeting of old-age benefits Targeting of other active age benefits
Note: (I) – refers to the group of countries: FRA, SLN, IRL; (II) – DNK, FIN, SWE, HUN.
Source: own calculations using LIS data
Recall that there are two causal steps in Korpi and Palme’s thesis. First, universal systems
tend to be larger systems, spending more on the poor and non-poor alike. Second, larger
systems tend to be more redistributive.
21
Figure 10 shows the first part of this causal chain, setting the concentration coefficient against
the measure of generosity we use here (share of transfers in gross income). The graph shows
that weaker targeting is associated with higher generosity11. Although in line with Korpi and
Palme, the relationship is a relatively weak one with again a lot of cross-country variation. A
strongly pro-poor system like Denmark’s, for example, is not smaller than weakly pro-poor
systems like Spain, Italy or Greece.
Figure 11 shows the second step in the causal chain and links generosity to redistributive
impact. Here the relationship is positive and it is also relatively consistent. There is not a
single system achieving a strong redistributive impact with a low level of spending and,
conversely, higher levels of spending tend to be associated with stronger redistributive
impacts. This is consistent with the findings from a host of studies (Nolan and Marx, 2009;
OECD, 2008 and 2011). Note here that the strongest redistributive impact is achieved by
countries that combine moderate (Sweden and Finland) to strong targeting (Denmark) with
comparatively high levels of spending. This marks the difference from a country like
Hungary, which achieves very high redistribution with an extremely high level of spending
but weak pro-poor targeting.
Figure 10. Concentration index (ranking by gross income) and generosity, mid 2000s
SVN AUT
BEL FRA
25
SWE
FIN LUX
CZE NORDEU ESP GRC
DNK NLD
EST ITA
IRL
15
GBR CHE
AUS ISR
CAN USA
5
-.4 -.2 0 .2
Targeting (Concentration index, ranking on gross income)
Note: due to data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France.
Source: Luxembourg Income Study
11
Similar pattern emerges when cash social spending as a share of GDP is used as an alternative indicator for
generosity, see Figure A.4 in Appendix 3.
22
Figure 11. Reduction in inequality due to taxes and transfers compared with size of social transfers
(expressed as a share of gross income), mid 2000s
50
SWE
DNK FIN
BEL HUN
NLD CZE
DEU POL
Redistribution index
NOR LUX AUT
40
IRL SVN
FRA
AUS ITA
EST
GBR
30
GRC
ISR CAN ESP
USA CHE
20
10 15 20 25 30 35
Transfers as % of gross income
Note: due to data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France.
Source: Luxembourg Income Study
5.2. Sensitivity
How robust are these results? They depend on a number of choices in our approach and
changing one of the parameters may considerably affect the outcomes and hence conclusions.
We test the sensitivity of the results for changes in a) the choice of the ranking income
concept, b) the role of taxes, c) the role of mandatory individual and occupational pensions
and d) the choice of the data source.
In this section, we look at the effect of changing the variable used to rank incomes when
calculating the concentration coefficients. The ranking determines the weight of the income
unit in its contribution to the concentration coefficient. Changing ranks thus affects the
weights, and consequently the value of the concentration coefficient.
So far results have been calculated using gross income as the pivotal income concept (notably
for ranking income units and determining generosity). When this is shifted towards disposable
income (as is done in e.g. Whiteford, 2010 and OECD, 2008), roughly the same pattern as in
our main analysis emerges (see Figure 12): no relationship between targeting and
redistribution.
23
Figure 12. Concentration index (ranking on disposable income) and redistributive impact
50
SWE FIN
DNK
BEL DEU HUN
CZE NOR NLD
Redistribution index
LUX POLAUT
40
SVN
IRL FRA
AUS ITA
GBR EST
30
ESP GRC
CAN ISR
CHE
USA
20
-.4 -.2 0 .2
Targeting (Concentration index, ranking on disposable income)
Note: for Belgium, France, Greece, Hungary, Slovenia and Spain calculations are based on disposable incomes
instead of gross incomes due to data availability.
Source: Luxembourg Income Study
The picture is similar to the one based on gross incomes mainly due to two reasons. On the
one hand, for some countries there is no information on taxes implying that gross income
corresponds to disposable income in this analysis. On the other hand, taxes in general cause
only little reranking of income units (in line with other results in literature), meaning that the
relative contribution of an income unit’s transfers to the concentration coefficient is hardly
affected.
When ranking incomes on the basis of market income (Figure 13), all concentration
coefficients turn to be negative, indicating that in all countries transfers are strongly pro-poor.
Strongest pro-poorness is found in Australia, the Netherlands, the United Kingdom,
Switzerland, and Denmark. According to this measure Estonia turns out to have the most
universal system.
As already mentioned, these differences between ranking based on market income and gross
income point to different degrees of reranking in countries. Reranking is very important in
countries where many social transfer-recipients have a market income of zero. Take the case
of the Netherlands. In the market income based approach this country has just about the most
targeted transfer system; in the gross income based calculation, social transfers appear to be
far more distributionally neutral. An important factor here is the relatively generous pension
system in the Netherlands. In the market income based calculation it is effectively assumed
that pensioners would have zero income in absence of the transfer. While this is clearly an
unrealistic counterfactual, the market income based approach does say something about the
extent to which transfers go to households solely reliant on them.
24
Figure 13. Concentration index (ranking by market income) and redistributive impact
50
DNK FIN SWE
BEL HUN
NLD CZE DEU
POL AUT NOR
Redistribution index
40
SVN LUX IRL
FRA
AUS ITA
GBR EST
30
GRC ESP ISR
CHE CAN
USA
20
-.5 -.45 -.4 -.35 -.3 -.25
Targeting (Concentration index, ranking on market income)
Note: due to data availability incomes are reported net of taxes for Belgium, Greece, Hungary, Slovenia and
Spain; part of “gross” incomes are reported net of taxes for Italy and France.
Source: Luxembourg Income Study
Not all LIS datasets include gross income amounts, implying that gross income in fact
corresponds to disposable income and that market income does not include taxes. This is the
case for seven countries in our analysis, namely Belgium, France, Greece, Hungary, Italy,
Slovenia and Spain. This also implies that social transfers are net of taxes in these countries,
whereas in the other countries gross transfers (i.e. before deduction of taxes) are used. The
gap from the “net” countries is thus particularly big for the countries where transfers are
subject to substantial taxation, as is the case in the Nordic countries.
25
Figure 14. Concentration index (ranking by gross income) and redistributive impact, restricting the
analysis to countries with full information on transfers and taxes
50
SWE FIN
DNK
CZE NORDEU NLD POL
Redistribution index
LUX AUT
40
IRL
AUS
GBR EST
30
CAN ISR
CHE
USA
20
-.4 -.2 0 .2
Targeting (Concentration index, ranking on gross income)
Therefore we restrict the analysis to countries with a similar gross income concept. Figure 14
shows that the relationship between the extent of targeting and redistributive impact becomes
a positive one, with countries on both ends of the targeting-universalism dimension
essentially achieving similar levels of redistribution. This also highlights that the omitted
countries (with the exception of Belgium), have positive concentration index values; focusing
on these countries yields a negative relationship between targeting and redistribution. The
question, though, remains if the nature of this negative relationship is related to certain system
characteristics of the countries with positive concentration index values or is it due to the
incomplete information on taxes. Overall, these results suggest that the difference between net
and gross transfers may play an important role in the outcomes and merit further
investigation.
In this section we trace the influence of mandatory individual and occupational pensions in
more detail. The LIS data offers such information for most of our countries, except of Austria,
the Czech Republic, Estonia, Spain, Greece, Hungary, Italy, Luxembourg, the Netherlands,
Poland, and Slovenia.
Why care about the impact of occupational pensions? Two primary reasons prevail. First, the
LIS database has recently underwent a major restructuring of its income variables. In the
“old” LIS terminology, which is used in Korpi and Palme, as well as Kenworthy studies,
social transfers do not include occupational pensions, as the latter income category is
considered a part of market (“factor”) income. This has changed in the new terminology,
where occupational pensions are considered as long-term insurance transfers, and thus
included in the list of social security transfers. Second, occupational pensions become an
26
increasingly important income source, especially given that progressively more people reach
the retirement age enabling the claim of the benefits. Hence, if less relevant in the past (when
such schemes were introduced), it becomes a substantial source of the old-age income
nowadays. For example, based on our calculations with LIS data, occupational pensions
constitute on average from 2% to 6% of gross incomes (or up to one third of household
received social transfers) in such non-EU countries, as Australia, Canada, Switzerland, the
United States or Israel, and such EU-countries, as Denmark, France, Norway, Sweden or the
United Kingdom. Occupational pension schemes are also ever more offered in countries,
which previously did not use them. Based on Eichhorst et al. (2011), many EU pension
system reforms are actually related to the introduction of the occupational pension schemes.
This is seen as a potential response to counteract the projected decreases in the average gross
public pension benefit and the average gross wage (or the benefit ratio). As such,
methodological consequences related to classification of occupational pensions are more than
a matter of pure taxonomy.
In the Kenworthy graph, the “old” typology is used. We replicate the graph in panel A of
Figure 15 (see Appendix 2 for the original figure). Application of the “new” terminology
implies a directional change, as shown in panel B of Figure 11. The main cause of this change
is related to two factors. First, the distribution of the occupational pensions is either weakly
pro-poor or even pro-rich. This contrasts with the strong or weak pro-poorness usually
observed for the rest of the social transfers and hence has important implications on the
overall distributional properties of social transfers. Second, the reclassification of
occupational pensions affects the value and the distributional properties of market income too.
For all countries of our analysis, the Gini coefficients of market income (which is used for
calculation of the redistribution index) have increased following this re-classification –
pointing to the equalizing nature of occupational pensions if considered as market income.
Consequently, the value of the redistribution index has increased. As such, the shift from the
upward to the downward sloping relationship is influenced both by effects on the
concentration and redistribution indices.
Figure 15. Social transfers inclusive of occupational pensions: concentration index (ranking by gross
income) and redistributive impact, mid 2000s – “Kenworthy” selection of countries
SWE
50
DNK
SWE FIN FIN
DNK NOR
DEU NLD DEU
NOR NLD
Redistribution index
Redistribution index
40
40
GBR
AUS
AUS CAN
GBR
30
30
CAN
USA
USA
20
20
Targeting (Concentration index, ranking on gross income) Targeting (Concentration index, ranking on gross income)
Note: countries with no information available on occupational pensions are in grey font.
Source: own calculations using LIS data
27
Figure 16. Social transfers inclusive of occupational pensions: concentration index (ranking by gross
income) and redistributive impact, mid 2000s – all countries
50
SWE
50
FIN DNK
SWE FIN
DNK HUN NOR HUN
BEL DEU BEL DEU
CZE NOR NLD POL CZE NLD
AUTPOL
Redistribution index
AUT
Redistribution index
LUX LUX FRA
40
40
IRL SVN IRL
SVN
FRA GBR
AUS CHE CAN
AUS EST ITA ITA
GBR EST
ISR
30
30
ISR ESP GRC ESP GRC
CAN
CHE
USA
USA
20
20
-.4 -.2 0 .2 -.3 -.2 -.1 0 .1 .2
Targeting (Concentration index, ranking on gross income) Targeting (Concentration index, ranking on gross income)
Note: due to data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France; countries with no
information available on occupational pensions are in grey font.
Source: own calculations using LIS data
Figure 16 repeats the message of Figure 15, but for all countries: if occupational pensions are
treated as part of social security transfers, the relationship between targeting and redistribution
becomes a negative one. A note of cautiousness is needed, though, here: this finding cannot
be verified for all our countries due to the lack of information in the LIS data. The question,
however, shows the importance of classification of social transfers and the role of the
occupational pensions per se.
Another important robustness check involves the one for data source. Figure 17 replicates the
main analysis on the basis of EU-SILC 2005 data (the value of indices using LIS and EU
SILC data can be found in Table A.1 of Appendix 3). As EU-SILC covers a different range of
countries than the LIS data, we focus only on the countries that are available in both datasets.
This implies that non EU-countries are excluded. All EU-SILC income types (e.g. gross
income, social transfers) are aligned as close as possible to the methodological choices of the
main part of our study on the basis of LIS data, though some differences are unavoidable due
to the dataset specificities 12. We discuss the latter issue below.
12
To ensure the highest degree of comparability between the two datasets, we derive own EU-SILC income
lists in accordance to methodological choices of our main analysis. First, this implies, that our EU-SILC
based gross incomes are somewhat different from the “original” (EUROSTAT derived) EU-SILC gross
incomes. For example, we include private individual pensions, while this is not the case in the EUROSTAT
list of gross incomes. Second, some income components are not traceable in EU-SILC and therefore cannot
be correctly allocated to the LIS analogous income lists. For example, we can neither exclude occupational
pensions from the old-age pensions’ category in EU-SILC nor we are sure this is the actual income category
of their allocation. Some other income types are not possible to identify in EU-SILC and thus to allocate
correctly too.
28
Figure 17. Concentration index (ranking by gross income) and redistributive impact in countries with
information available both in LIS and EU SILC data
55
55
FIN DNK FIN DEU HUN
Redistribution index
Redistribution index
SWE NOR SWE
DNK SVN
BEL DEU NLD HUN BEL
45
45
CZE NOR POL CZE NLD FRA
LUX AUT
LUX AUT
IRL SVN GBR
FRA IRL
35
35
ITA POL
GBR EST EST ITA
GRC
ESP GRC
ESP
25
25
-.4 -.3 -.2 -.1 0 .1 .2 -.2 -.1 0 .1 .2
Targeting (Concentration index, ranking on gross income) Targeting (Concentration index, ranking on gross income)
Note: due to LIS data availability disposable instead of gross incomes are used for Belgium, Greece, Hungary,
Slovenia and Spain; part of “gross” incomes are reported net of taxes for Italy and France; in EU-SILC data,
disposable instead of gross incomes are reported for Greece, Italy and Spain.
Source: own calculations using LIS and EU-SILC data
As shown by Figure 17, the relationship between targeting and redistribution using LIS data
would remain rather unclear, though slightly downward sloping. The picture based on the EU-
SILC data is somewhat different: a clearer pattern appears indicating that the more targeted
systems tend to be the most redistributive.
Furthermore, we also observe a denser positioning of the countries with respect to the x axis
values. Especially, at the outer ends of the targeting dimensions we find relative consistency:
the most targeted systems (Denmark, the Czech Republic and Belgium) achieve
comparatively strong redistribution while the least pro-poor systems (Greece, Italy) have
comparatively weak redistributive impacts. Conversely, a somewhat more widespread
redistribution portrait is obtained with EU-SILC than with LIS data. For example, there are
more countries with redistribution index at or higher than 45% in EU-SILC than in LIS data.
There are also some other striking differences of specific country outcomes based on EU-
SILC in comparison to LIS data:
• Italy takes a somewhat different position (more towards the “universalist” side of the
spectrum although this is a clear misnomer in the Italian case, or for that matter the
other south European countries). The southern European countries still remain
important drivers of the negative relationship.
• The Czech Republic takes a striking position in having a strongly targeted and highly
redistributive system. A similar, though less pronounced, move is observed for
Belgium. Altogether, Denmark, the Czech Republic and Belgium appear as the most
redistributive and the most targeted systems.
• The UK features as having a far less targeted system than in the LIS based analysis.
Other countries show up as less targeted systems (e.g. Denmark, Sweden, Ireland), but
to a lesser extent than the UK.
• Hungary and Slovenia appear as having the most redistributive systems, together with
Sweden, whereas the redistributive power of Polish is recorded as much lower.
What are the major factors driving differing country positions when the EU-SILC data is
used? Differences in classification of income and varying information sources are of primary
29
importance, with the latter one being the major contributor. We depict this in Table 1, where
values of concentration and redistribution indices are compared.
As noted before, LIS dataset information for some countries actually stems from national
SILC surveys (see Appendix 1 for more information). We list these countries in the first half
of Table 1. The comparison of concentration and redistribution indices for these countries
proves that essentially the same results are obtainable from both datasets. In particular, the
redistribution indices are close in their values. The surprising exceptions though concern the
concentration indices for Spain and especially Ireland – they refer to large(r) data mismatches
between the two datasets. Given that redistribution indices, which are based on distributional
properties of market and disposable incomes, have no such biases for these two countries, our
conclusion is that the list of social transfers is somewhat different for these two countries
between the datasets – and thus affects the value of the derived concentration indices. This
may concern, for example, the classification of occupational pensions.
For the countries with different data sources in EU-SILC and LIS, biases between the indices
are much larger, in particular, for the concentration indices. Hence, this shows the importance
of the choice of data source in explaining the relationship between targeting and
redistribution. Not only differing data sources and income classifications play a role here, but,
also differences in terms of the availability of no or only partial tax information. This applies
to French, Hungarian, Italian and Slovenian data, pointing to the need of cautiousness when
interpreting the results.
Table 1. Concentration (ranked on gross incomes) and redistribution indices in EU-SILC and “old”
LIS data
30
6. Discussion: making sense of the disappearing paradox
The disappearance of the negative relationship between targeting and redistributive impact as
established by Korpi and Palme is driven by two factors. First, as Kenworthy (2011) has
already demonstrated, the observed relationship for the original set of countries in the Korpi
and Palme study has weakened over time. This raises the question: what has changed in those
countries? Second, and arguably more crucially, it is the inclusion of new countries that
causes the relationship to become a very weak one. This raises a second major question: what
is different about the countries which did not feature in the original analysis ?
The Korpi and Palme argument essentially is about the relative size of the electorates
benefiting from and paying for redistributive measures. The studies by Moene and Wallerstein
and by others explicitly test theoretical models that seek to explain varying attitudes to
universal versus selective systems in terms of their pure redistributive impacts across
electorates, particularly the middle class (for an overview see McCarthy and Pontusson,
2009).
From this perspective it is not easy to understand why, for example, the United States has
shifted towards a more universalist position. Perhaps an explanation is to be found in the fact
that the debates that triggered and shaped social policy changes over the 1990s and 2000s
were less about such distributional issues than about the perceived effects of strongly targeted
redistributive policies.
One of the factors that arguably made some targeted systems less politically robust and prone
to spending cuts in the 1980s was the fact that strongly targeted, particularly means-tested
benefits entailed strong work disincentives and other behavioural incentives. The issue was
not so much that these systems catered to a small part of the electorate, far removed from the
median voter and that for this reason they suffered from relative neglect by politicians.
Quite to the contrary, means-tested systems that catered to the poor were at the heart of
political debates. In the United States, the main means-tested system (AFDC) became the
focus of quite heated political debates during the 1980s. Charles Murray’s Losing Ground
(1984) launched a virulent attack on this final safety net provision as it was then in place. That
system was identified as the main culprit in creating an underclass of chronically welfare
dependent single mothers. While the book and its claims became the object of equally virulent
criticism from the left, welfare reform took centre stage in the political debate. Clinton ran his
first campaign in a slogan to ‘end welfare as we know it’.
What ensued was a major shift in social policy. Statutory time limits on social assistance
benefit duration were introduced. This move was accompanied by the expansion of a targeted
benefit of an altogether different nature: the Earned Income Tax Credit. The Earned Income
Tax Credit (EITC) had been introduced in the US in 1975 as an exemption from employees’
social security payments for poor working households with children. The system was
subsequently expanded and the 1993 reform in particular turned the scheme into the country’s
pre-eminent anti-poverty program for families of active working age. The expansion of EITC
was accompanied by the introduction of new training and employment schemes and several
31
increases in the minimum wage. This combination represented a paradigmatic shift in
American social policy. Empirical studies show that the expansion of EITC, in combination
with other policy reforms and several increases in the minimum wage, produced some striking
initial results, including marked increases in labour market participation and declines in
poverty among some segments of the population, especially single-parent households (Hotz
and Scholz, 2003; Eissa and Hoynes, 2004).
The EITC has become America’s pre-eminent welfare programme. Spending has increased
and appears to enjoy relatively broad and robust political support. This raises the question
why this it. The system is less strongly targeted than before, which probably in part accounts
for the shift of the US in the graphs. The system now caters to larger sections of the
electorate, including the (lower) middle class, and this may account for that expansion. But an
equally if not more important factor may be the fact that the system is perceived to encourage
and reward work; it enjoys greater overall legitimacy and that may explain why spending on
EITC has risen dramatically (Kenworthy, 2011).
Activation and the prevention of long-term dependence among able-bodied people at active
age has also became a major issue in Europe and an increased policy emphasis on activation
has become evident, certainly at the level of policy rhetoric, and gauging by some indicators
also in terms of actual policy (Barbier and Ludwig-Mayerhofer 2004; Kenworthy, 2008;
Dingeldey, 2007; Eichhorst and Konle-Seidl, 2008, Aurich, 2009; Immervoll, 2010; Marchal
and Van Mechelen, 2013). That said, the truth is that we still lack reliable indicators of actual
activation intensity, mainly because implementation aspects are so difficult to measure (e.g.
effective sanctioning or effective availability and take-up of training places, subsidized jobs
etc.). To reduce work disincentives, earnings disregards have been introduced for people
dependent on social assistance or unemployment benefits who make a (partial) transition from
complete benefit dependency to part-time work. People on targeted benefits have also become
the focus of intensified monitoring, activation and sometimes sanctioning efforts (Marx and
Nelson, 2012).
Most importantly, perhaps, means-tested benefits are no longer exclusively aimed at people
not in work, but also at those in work in low-paid jobs. The French RsA scheme is a good
example of a new style means-tested benefit scheme that offers integrated support for the non-
employed and (part-time) low paid workers alike. The scheme also has entirely different work
incentives. The RSA (Revenu de Solidarité Active), was introduced in France in 2008 the
specific aim of remodelling the incentive structure of people on social assistance, and
particularly to make work or returning to education a more lucrative financial prospect. The
previous minimum income system (Minimum Integration Income - RMI) was based on a one
for one trade-off of benefit for earned income, so that EUR 100 earned led to a deduction of
EUR 100 from benefit paid. Under RSA a “62% slope” is applied so that earning EUR 100
leads to a EUR 38 deduction in benefit paid. Efforts have also been made to encourage
beneficiaries of RSA into employment, for example with assisted employment contracts and
(improved) insertion mechanisms. In addition, the RSA has simplified the provision of social
protection by combining several previously separate schemes into a single sum. A household
with no earned income is eligible for the “basic RSA” while the “in-work RSA” acts as a top-
up for people with low earnings.
The point here is that targeted, means-tested systems look totally different today from the
systems in place in the 1980s. Whereas the old systems were the focus of harsh welfare
32
critiques, especially from the right, the new targeted systems are lauded as the essential
gateways of welfare to work.
The original Korpi and Palme analysis included Australia, Canada, Finland, France, Germany,
the Netherlands, Norway, Sweden, Switzerland, the United Kingdom and the United States.
The additional countries included in our LIS based analysis roughly fall into three categories:
a) three Southern European countries (Greece, Italy, Spain); b) post-communist countries now
part of the EU (Czech Republic, Estonia, Hungary and Slovenia) and c) advanced economies
not included in the original analysis: Austria, Ireland, Israel and Luxembourg.
The southern European countries included in our analysis – Greece, Spain and Italy – drive
much of the observed weak positive relationship between targeting and redistributive impact.
In terms of the targeting measure used in this study the southern European countries rank as
having the least amount of targeting. Looking at the share of transfers going to the bottom
quintile of the income distribution (ranked by gross income), the targeting coefficient even
underestimates the pro-richness of the transfers there (Figure A5). In the LIS based analysis,
the share going to the bottom quintile in Italy, Greece and Spain is just over 10 per cent. This
is three times less than the 35 per cent of transfers flowing to the poorest in the countries with
the highest concentration coefficients: the UK, Denmark and Australia.
A distinct feature of the southern European welfare systems is a high degree of categorical
differentiation, mostly by occupational category. This is referred to as “dualistic social
insurance” (Ferrera, 2010). While some segments of the population, and particularly the
workforce, are relatively well catered for, other segments essentially receive little or nothing.
Having co-evolved with a highly segmented, breadwinner biased labour markets, the southern
European welfare states exhibit a strong degree of internal segmentation: generous benefits
for core/regular workers, modest benefits for peripheral/irregular workers (what Jessoula et al.
(2003) call the ‘mid-siders’) and meagre entitlements, if any at all, for those with no formal
labour market attachment. There is a sizeable literature showing that welfare systems there
tend to reinforce or at least replicate socio-economic and occupational inequalities (Ferrera,
2010; Matsaganis et al., 2003). The southern European countries also remain relatively unique
among the “old” European countries in not having nationally organized social safety nets,
except in Portugal where it was introduced in 1997. In Italy and Spain social assistance
remains a regional matter and benefit levels vary quite considerably (Van Mechelen and
Marchal, 2012).
The four post-communist economies take strikingly varying positions on the two main
dimensions under focus here. In the LIS based analysis, the Czech Republic appears to have a
relatively targeted system while Poland is to be found on the other end of the spectrum;
33
redistributive impacts, however, are very similar in both countries. (Note that market income
inequality remains comparatively low in the Czech Republic, but high in Hungary or Poland.)
We observe these countries at one point in time, capturing their current position in ongoing
trajectories of change. After the collapse of the Soviet Union these countries experienced a
major transition shock, marked by massive labour shedding and declining wages. (The Czech
Republic, a country that takes a marked position in these analyses, but also in other
comparative studies on inequality and poverty in relation to welfare state spending, was less
affected.) A number of CEE governments implemented targeted safety nets and expanded
pensions, unemployment and family benefits, be it in an ad hoc way, and driven to some
extent by organised interests (Cook, 2010). As the 1990s progressed many CEE started to
restructure their welfare states, taking a turn towards liberalization. The extent and intensity of
liberalization differed, however, with countries like the Czech Republic maintaining the most
solidaristic welfare provision. Hungary and Poland did embark on the path of liberalisation
while the Baltic states, initially strongly affected by the transition, first lagged and then took
the most radical turn towards market-conforming reforms. Universal subsidies and family
benefits, for example, were partially replaced by means-tested benefits targeted at the poor.
Concurrently, there has been a partial return to pre-communist Bismarckian traditions. For
example, financing of pensions and social insurance was moved from state budgets to payroll
taxes. Mandatory capitalized pillars were added to pay-as-you-go pension systems.
At present, there is a considerable degree of diversity among the post-socialist countries and it
is important to underscore the point that these do not belong to a singular ‘regime’. This
diversity is also evident from measures of social spending, taxation and benefit generosity
(Cerami and Vanthuysse, 2010; Cook; 2010). Institutional indicators of minimum income
protection also point to very significant differences (Van Mechelen & Marchal, 2012;
Goedemé, 2012). It is also clear that parts of welfare systems there remain in a state of flux
(Marx and Nelson, 2013).
7. Conclusion
This paper has re-examined Korpi and Palme’s influential claim that “the more we target
benefits at the poor, the less likely we are to reduce poverty and inequality”. We find that this
tenet no longer holds as a robust empirical generalisation. This paper has replicated their
original analysis for a broad set of rich countries and finds that by and large the relationship
between targeting and redistributive impact is a very weak one across countries, suggesting
that the extent of targeting per se may not matter as much as we have assumed since Korpi
and Palme. For what it matters, targeting tends to be associated with higher levels of
redistributive impact, especially when overall effort in terms of spending is high.
A key point of this paper, however, is that empirical specification, data source and country
selection matter in very significant ways – and this perhaps holds a lesson for comparative
welfare state research in a wider sense.
The point of departure for this paper has been Korpi and Palme’s original empirical
specification so as to achieve maximum initial comparability When applying this
methodology, i.e. exactly replicating the original study for more recent years, we find that the
relationship looks totally different today, with countries taking very diverse positions on both
the targeting and the redistribution dimensions. No clear relationship appears across countries.
34
Some countries, notably Denmark, have transfers systems that are strongly targeted at the
poor and yet yield strong redistributive impacts, contrary to the original Korpi and Palme
thesis. At the other end the picture looks equally diverse, if not more so. Weak targeting is in
some cases clearly associated with relatively weak redistributive impacts. The best
performing countries in terms of redistributive impact employ ‘targeting within
universalism”. But this in itself is no guarantee for strong redistribution: quite a number of
countries that target in equal measure, but not necessarily with the same level of spending
effort, have poorly performing systems in terms of redistributive impact.
As indicated, outcomes are to some extent sensitive to operationalisation, data source and,
especially, country selection. These matter to varying degrees.
The choice of the ranking variable does matter to some extent, as others have also
demonstrated, but not in a way that substantially changes the picture. Ranking by market
income instead of gross income yields stronger redistributive effects, because households, like
pensioners, solely reliant on transfers are assumed to have zero market incomes in the absence
of transfers. This produces stronger redistributive impacts almost by definition.
Rather more substantially relevant is the fact that the categorisation of income components
matters. It is striking that the picture shifts in a significant way once occupational pensions are
treated as transfers rather than as part of market income. In view of the extent of state
regulation and in some cases support of occupational pensions a good case can be made that
these are more properly seen as transfers than as market income. If we do so, stronger
targeting tends to be associated with a larger redistributive impact.
It also matters whether we consider transfers in total or whether instead we do the analysis at
a more disaggregated level. A decomposition analysis shows that the relationship between
targeting and redistribution is different for family transfers as compared to old-age benefits, or
other benefits for the working aged. Only for family benefits do we still find a weak negative
relationship between targeting and redistributive impact. The spread of countries is quite
considerable however.
Another striking sensitivity result is the comparison between LIS and SILC. Analysis on the
SILC data clearly yields a picture strongly suggestive of a positive relationship between
targeting and redistributive impact. This is in part because countries covered in both datasets
take different positions in some cases. Different country positionings are due to three major
factors: differences between the data sets in the way incomes are classified, differences in the
the underlying data sources, and the inclusion or exclusion of tax information.
Finally, there is the key issue of country selection. The fact that the relationship originally
established by Korpi and Palme breaks down is due to two factors. First, the countries
originally included in the study have shifted their position. Second, and more important, there
is now data available for a far larger set of countries. It is the addition of these new data points
which causes the relationship to break down completely.
We have tried to make substantive sense of this breakdown of this by focusing on two
questions: a) what may have changed in the countries originally included in the analysis; b) in
what respect are welfare states now additionally included different? The key issue underlying
both questions is this: what does this mean for what we think to know about the dynamics of
political support for welfare programmes. Can programmes that cater in a targeted way to
35
relatively small segments of the electorate maintain sufficient political support so that they do
not become “poor”, i.e. underfunded programmes?
Recall that the Korpi and Palme argument essentially is about the relative size of the
electorates benefiting from and paying for redistributive measures. We have argued here that
the debates that triggered and shaped social policy changes over the last decades were less
about such distributional issues than about the incentive effects of redistributive policies. The
issue was not so much that strongly targeted systems catered to a small part of the electorate,
far removed from the median voter. The political weakness rather derived from the perceived
work and family formation incentives. Nowadays strongly targeted (means-tested) benefits
are no longer exclusively aimed at people not in work, but also at those in work in low-paid
jobs. Whereas the old systems were the focus of harsh critiques, especially from the right, the
“new” targeted systems (for example EITC in the US, RSA in France or WTC in the UK) are
lauded as the essential gateways of welfare to work. These systems enjoy relatively broad and
robust political support.
Another part of the answer lies in the fact that we now have data for a larger set of countries.
The Southern European countries drive much of the observed positive relationship between
targeting and redistributive impact. The post-socialist countries take quite varying positions
on both the targeting and the redistribution dimensions. Interestingly, all countries with
positive concentration indices were missing in the original Korpi and Palme study, effectively
limiting the range of observations from countries with strong targeting to countries with weak
targeting.
This paper raises further questions. Why does a similar degree of strong targeting, as captured
by the concentration index, produce stronger redistributive outcomes in Denmark as
compared to the United Kingdom and Australia? More broadly the question is: why are
similar levels of targeting associated with different redistributive outcomes across countries?
Our analysis also points to the potentially dissimilar impact of targeting across different types
of transfer provisions, for example child benefits as compared to pensions. To what extent all
this is a result of our modelling choices, of compositional factors (e.g. the distribution of
beneficiaries across the income spectrum) or actual design features of systems is still to be
established. We have also brought to light that the impact of taxes, both in general and
specifically those on transfers, requires further exploration.
36
References
Allan, J. P., & Scruggs, L. (2004). Political Partisanship and Welfare State Reform in
Advanced Industrial Societies. American Journal of Political Science, 48(3), 496-512.
Alvarez P.B., (2001), ‘The Politics of Income Inequality in the OECD: The Role of Second
Order Effects’, LIS-working paper, No. 284, 29p.
Atkinson A., (1993), 'Work Incentives', in A.B. Atkinson and G.V. Mogensen (eds.), Welfare
and Work Incentives: A North European Perspective, Clarendon Press, Oxford.
Aurich, P. (2009), "Levels, directions and divisions of change towards activation in Europe",
ISA World Congress 2010, RC 19.13 ,Comparing Activation Models and Policy', July
11-17, Gothenburg, Sweden.
Baldwin P., (1990), The Politics of Social Solidarity: Class Bases of the European Welfare
State 1875-1975, Cambridge University Press, Cambridge.
Bandelj & Mahutga (2010). How Socio-Economic Change Shapes Income Inequality in Post-
Socialist Europe. Social Forces, Vol. 88, No. 5, pp. 2133-2162.
Barr N., (1998), The Economics of the Welfare State, Oxford University Press, Oxford.
Barbier, J.-C. and Ludwig-Mayerhofer, W. (2004) 'Introduction: the many worlds of
activation', European Societies, Vol. 6, No. 4, pp. 423-36.
Barth, E and Moene, K. (2009) The Equality Multiplier. NBER Working Paper 15076.
Cambridge Massachusetts: National Bureau of Economic Research.
Beramendi Alvarez, P. (2001) The Politics of Income Inequality in the OECD. The Role of
Second Order Effects. Luxembourg Income Study Working Paper No. 284.
Bergh, (2005), ‘On the Counterfactual Problem of Welfare State Research: How Can We
Measure Redistribution?’, European Sociological Review, 21 (4), 345-357.
Borck, Rainald. 2007. “Voting, Inequality and Redistribution.” Journal of Economic Surveys
21(1): 90-109.
Brady, David and Rebekah Burroway. "Targeting, Universalism and Single Mother Poverty:
A Multi-Level Analysis Across 18 Affluent Democracies." Demography 49 (2012):
719-746.
Bradley, D., Huber, E., Moller, S., Nielsen, F., & Stephens, J. D. (2003). Distribution and
redistribution in postindustrial democracies. World Politics, 55(02), 193-228
Brooks, C., & Manza, J. (2006). Social Policy Responsiveness in Developed Democracies.
American Sociological Review, 71(3), 474-494.:
Cantillon B., Marx I., Van den Bosch K., (2003), ‘The puzzle of egalitarianism: relationship
between employment, wage inequality, social expenditure and poverty’, European
Journal of Social Security, 5(2), p. 108-127.
Cerami, A. and Vanhuysse, P. (eds), Post-Communist Welfare Pathways. Theorizing Social
Policy Transformations in Central and Eastern Europe, Basingstoke: Palgrave
Macmillan: 164-180
Cook, L. (2010), ‘Eastern Europe and Russia’, in Castles, F et al. (eds) Oxford Handbook of
the Welfare State; Oxford: Oxford University Press
37
Corak, M., Lietz, C. and Sutherland H. (2005) The Impact Of Tax And Transfer Systems On
Children In The European Union, EUROMOD Working Papers EM4/05, EUROMOD
at the Institute for Social and Economic Research
Corcoran, S. and Evans, W.N. (2010) Income Inequality, the Median Voter, and the Support
for Public Education. NBER Working Paper 16097. Cambridge Massachusetts:
National Bureau of Economic Research.
Deleeck, H. et al. (1983) Het Matteüseffect. De ongelijke verdeling van de sociale
overheidsuitgaven in België. Antwerpen: Kluwer.
Dingeldey, I. (2006), ‘Between workfare and enabling– Different Paths to Transformation of
the Welfare State: A comparative analysis of activating labour market policies’, in
European Journal of Political Research 46 (6): 823-851.
Eichhorst, Werner., Gerard, M., Kendzia, M., Mayrhuber, C., Nielsen, C., Rünstler, G. and
Url, T. (2011), ‘Report No. 42: Pension Systems in the EU – Contingent Liabilities and
Assets in the Public and Private Sector’, IZA Research Report No 42.
Eichhorst, W. and Konle-Seidl, R. (2008), Contingent Convergence: A comparative analysis
of activation policies, IZA DP 3905
Eissa, N. and Hoynes, H. (2004) ‘Taxes and the Labor Market Participation of Married
Couples: The Earned Income Tax Credit’, Journal of Public Economics 88: 1931-1958.
European Commission (2013), Communication to the European Parliament, the Council, the
European Economic and Social Committee and the Committee of the Regions, Towards
Social Investment for Growth and Social Cohesion – including implementing the
European Social Fund 2014-2020, COM (2013) 83 final
EUROSTAT (2007). ‘EU-SILC User Database Description Version 2007-1 from 01-03-09’,
European Commission, EUROSTAT; Luxembourg.
Ferrera, M. (2010), ‘The South European Countries’, in Castles, F et al. (eds) Oxford
Handbook of the Welfare State; Oxford: Oxford University Press
Ferrarini, T., Nelson, K. And Höög, H. ‘From universalism to selectivity: old wine in new
bottles for child benefits in Europe’, in Marx, I., Nelson, K. eds. (2012), Minimum
Income Protection in Flux, Basingstoke: Palgrave Macmillan
Figari, F., Paulus, A. and H. Sutherland (2011), ‘Measuring the size and impact of public cash
support for children in cross-national perspective’, Social Science Computer Review
29(1), 2011
Forma, P. (1997) ‘The Rational Legitimacy of the Welfare State: Popular Support of ten
Income Transfer Schemes in Finland’, Policy and Politics, Vol. 25, No.3, 235-249.
Förster, M. and Mira d’Ercole, M. (2005) Income Distribution and Poverty in OECD
Countries in the Second Half of the 1990s. OECD Social Employment and Migration
Working Papers, No. 22. Paris: Organisation for Economic Co-operation and
Development.
Fuchs, M. (2009) 'Social assistance - no, thanks? The non-take-up phenomenon and its
patterns in Austria, Germany and Finland after 2000'. European Centre Policy Brief
September 2009. Vienna: European Centre.
Goedeme, T. ‘Minimum income protection for Europe’s elderly. What and how much has
been achieved’, in Marx, I., Nelson, K. eds. (2012), Minimum Income Protection in
Flux, Basingstoke: Palgrave Macmillan
38
Goodin R., (1988), Reasons for Welfare: The Political Theory of the Welfare State, Princeton
University Press, Princeton.
Goodin R., LeGrand J., (1987), Not Only the Poor; The Middle Classes and the Welfare State,
Unwin Hyman, London.
Horn, Daniel. 2011a. “Income Inequality and Voter Turnout.”GINI Discussion Papers no. 16.
Hotz, V.J. and Scholz, J. K. (2003) ‘The Earned Income Tax Credit’, in R. Moffit (ed) Means-
tested Transfer Programs in the U.S.. University of Chicago Press.
Jäntti, M., Kangas, O. And Ritakallio, V. (1996) ‘From Marginalism to Institutionalism:
Distributional Consequences of the Transformation of the Finnish Pension Regime’,
Review of Income and Wealth, Vol. 42, No. 4, pp. 473-491.
Jessoula, M., Granziano, P. R., & Madama, I. (2010). 'Selective Flexicurity' in Segmented
Labour Markets: The Case of Italian 'Mid-Siders'. Journal of Social Policy, 39(4), 561-
583.
Jesuit, D. K. and Mahler, V. A. (2010). Comparing Government Redistribution Across
Countries: The Problem of Second-Order Effects. Social Science Quarterly, 91: 1390–
1404.
Kakwani, N. (1977a), “Measurement of tax progressivity: an international comparison”, in
Economic Journal, vol. 87, pp.71-80.
Kangas, O. (1995) ‘Attitudes on Means-Tested Social Benefits in Finland’, Acta Sociologica,
Vol. 38, Issue 4, pp. 299-310.
Kenworthy, L., & Pontusson, J. (2005). Rising Inequality and the Politics of Redistribution in
Affluent Countries. Perspectives on Politics, 3(03), 449-471. :
Kenworthy, L. (2008) Jobs with Equality, Oxford: Oxford University Press
Kenworthy, L. (2011) Progress for the Poor. Oxford: Oxford University Press (forthcoming)
Korpi, W. (1980) ‘Social Policy and Distributional Conflict in the Capitalist Democracies. A
Preliminary Comparative Framework’, West European Politics, Vol. 3, No. 3, pp. 296-
316.
Korpi, W. and Palme, J. (1998) ‘The Paradox of Redistribution and Strategies of Equality:
Welfare State Institutions, Inequality, and Poverty in the Western Countries’, American
Sociological Review, Vol. 63, No. 5 ,p. 661-687.
Lambert, P. (2001), The Distribution and Redistribution of Income, Third Edition, Manchester
University Press.
Larsen, C. (2008) ‘The Institutional Logic of Welfare Attitudes. How Welfare Regimes
Influence Public Support’, Comparative Political Studies, Vol. 41, No.2,pp. 145-168.
Lindbeck A., Nyberg S., Weibull J., (1999), “Social Norms and Economic Incentives in the
Welfare State”, Quarterly Journal of Economics, 114(1): 1-35.
Lindert, P.H. (2004) Growing Public: Social Spending and Economic Growth since the
Eighteenth Century. Cambridge: Cambridge University.
Mahler, V. A. (2008), Electoral turnout and income redistribution by the state: A cross-
national analysis of the developed democracies. European Journal of Political
Research, 47: 161–183.
39
Marx, I., Nelson, K. eds. (2013), Minimum Income Protection in Flux, Basingstoke: Palgrave
Macmillan
Matsaganis, M., Ferrera, M., Capucha, L., & Moreno, L. (2003). Mending Nets in the South:
Anti-poverty Policies in Greece, Italy, Portugal and Spain. Social Policy and
Administration, 37(6), 639-655.
McCarty, N. and Pontusson, J. (2009) ‘The Political Economy of Inequality and
Redistribution’, in W. Salverda, B. Nolan and T. Smeeding (eds) Oxford Handbook of
Economic Inequality. Oxford: Oxford University Press.
Meltzer, A. and Richard, S. (1981) A rational theory of the size of government. Journal of
Political Economy, Vol. 89, No. 5, pp. 914-927.
Merton, R. (1968) ‘The Matthew effect in science’, Science, Vol. 159, No. 3810, pp. 56-63.
Milanovic, B . (2000). “The Median-Voter Hypothesis, Income Inequality and Income
Redistribution.” European Journal of Political Economy 16(2-3): 367-410.
Moene, K. and Wallerstein, M. (2001) ‘Inequality, Social Insurance and Redistribution’,
American Political Science Review, Vol. 95, No. 4,pp. 859-874.
Moene K., Wallerstein M., (2002), ‘Income Inequality and Welfare Spending: A
Disaggregated Analysis’, Unpublished paper. North-western University, Evanston, IL,
February.
Moene, K. and Michael Wallerstein, M. (2003) ‘Earnings Inequality and Welfare Spending’,
World Politics , Vol. 55, No. 4, pp485-516.
Murray, C. (1984), Losing ground: American social policy, 1950-1980, New York: Basic
Books.
Nelson, K. (2004). Mechanisms of poverty alleviation. Journal of European Social Policy, 14,
371–390.
Nelson, Kenneth (2007). "Universalism versus Targeting: The Vulnerability of Social
Insurance and Means-Tested Minimum Income Protection in 18 Countries, 1990-2002."
International Social Security Review 60: 33-58.
Nolan, B. and Marx, I. (2008) ‘Poverty and Social Exclusion’, in W. Salverda, B. Nolan and
T. Smeeding (eds) Oxford Handbook of Economic Inequality. Oxford: Oxford
University Press.
OECD, (2004), OECD Employment Outlook 2004, OECD Publications, Paris.
OECD (2008) Growing Unequal: Income Distribution and Poverty in OECD Countries.
OECD Publications, Paris.
OECD (2011) Divided we Stand. Why Inequality Remains Rising. OECD Publications Paris.
Olivera, Javier 2012 Preferences for redistribution in Europe Gini Discussion Paper No 67
Okun A.M., (1975), Equality vs Efficiency: The Big Trade-Off , Brookings Institution Press,
Washington D.C.
Pierson, P. (1994) Dismantling the Welfare State? Reagan, Thatcher, and the Politics of
Retrenchment. Cambridge: Cambridge University Press.
Ringen S., (1987), The Possibility of Politics, Clarendon Press, Oxford.
40
Skocpol, T. (1991). Targeting within universalism: Politically viable policies to combat
poverty in the United States. In C. Jencks & P. E. Peterson (Eds.), The urban underclass
(pp. 411–436). Washington, DC: The Brookings Institution.
Scruggs L., (2004), Welfare State Entitlements Data Set: A Comparative Institutional
Analysis of Eighteen Welfare States, Version 1.1. Beschikbaar op:
http://www.sp.uconn.edu/~scruggs/wp.htm.
Titmuss, R. (1969), Essays on the Welfare State, second edition with a new chapter, London:
Unwin University Books.
Toth, I., Horn, D. and M. Medgyesi (2013), Rising inequalities: will electorates go for higher
redistribution, forthcoming in Salverda et al. (eds), OUP.
Vandenbroucke, F, Diris, R., Verbist, G. (2013), Excessive social imbalances and the
performance of welfare states in the EU, mimeo
Van Kerm, P. (2009), ‘sgini: Generalized Gini and Concentration coefficients (with factor
decomposition) in Stata', v1.0, CEPS/INSTEAD, Differdange, Luxembourg.
Van Lancker Wim, Ghysels Joris, Cantillon Bea.- An international comparison of the impact
of child benefits on poverty outcomes for single mothers Antwerpen: , 2012.- 28 p. -
(CSB working paper/University of Antwerp, Herman Deleeck Centre for Social Policy ;
12/03)
Van Mechelen, N., and J. Bradshaw (2013), ‘Child Poverty as a Government Priority: Child
Benefit Packages for Working Families, 1992-2009’, in I. Marx, and K. Nelson (2012),
Minimum Income Protection in Flux. Basingstoke: Palgrave Macmillan.
Van Mechelen, N., and S. Marshal (2013), ‘Struggle for Life: Social Assistance Benefits,
1992-2009’, In I. Marx, and K. Nelson (eds.), Minimum Income Protection in Flux.
Basingstoke: Palgrave Macmillan. Macmillan.
Van Oorschot, W. (2002) ‘Targeting Welfare: on the Functions and Dysfunctions of Means-
testing in Social Policy’ in P. Townsend and D. Gordon (Eds.), World Poverty: New
Policies to Defeat an Old Enemy. Bristol: The Policy Press.
Verbist, G. (2004) Redistributive Effect and Progressivity of Income Taxes: An International
Comparison across the EU using EUROMOD, EUROMOD Working Paper EM5/04,
The Microsimulation Unit, Cambridge University.
Whiteford, Peter (2007) "Targeting, Redistribution, and Poverty Reduction in OECD
Countries." Unpublished.
Whiteford, Peter (2008) "How Much Redistribution Do Governments Achieve? The Role of
Cash Transfers and Household Taxes." Chapter 4 in Growing Unequal? Paris: OECD
Whiteford, Peter (2009) "Transfer Issues and Directions for Reform: Australian Transfer
Policy in Comparative Perspective." Unpublished. Social Policy Research Center,
University of New South Wales.
41
Appendix 1- Comparing LIS and EU-SILC datasets
The EU-SILC data is collected using a common framework across the participating European
countries and therefore encompasses a high degree of survey (design) comparability. The LIS
data (i.e. the Luxembourg Income Study Database) is a harmonised microdata with the
underlying information sources variable across the participating countries. As shown in a
table below, LIS data for a number of our countries (i.e. Austria, Estonia, Greece, Spain,
Ireland, etc.) is derived from the national EU-SILC surveys – seemingly, the most common
type of LIS information source for the European countries as of mid 2000’s. The other
frequent source of information is household budget surveys (e.g. France, Poland, Slovenia), as
well as diverse income and expenditure surveys. LIS data for Denmark is derived from the
income tax register data.
Given that LIS and EU-SILC have the same information sources for some of their countries,
any analysis based on one or the other dataset should essentially yield the same for the latter
countries. Some differences are, however, likely to occur. For example, imputation policy is
different across the two databases. LIS does not implement own imputations, whereas there is
a requirement to “impute missing values in the income variables where that can be reasonably
done” for the EU-SILC data (EUROSTAT, 2007). Furthermore, classification of incomes is
somewhat different across the two datasets. The table below presents the structure of gross
and disposable incomes in the LIS and EU-SILC data in more detail and highlights
differences in individual income components.
In general, the broadest income categories are essentially the same in both datasets and
concern employment, capital incomes or private transfers. Larger differences, though, lie in
the classification of smaller income components, making comparisons of certain income
types, as social transfers, between the two datasets cumbersome. We give an example using
the old-age pension classification.
The EU-SILC refers to “old-age benefits” as all types of benefits (i.e. disability, survivor,
retirement pensions) if paid to a person after a (country specific) standard retirement age. The
original identification of the benefit payment (i.e. due to disability) is not kept when pooling
information under old-age benefits. A different income typology is applied in LIS data. Here,
transfers are classified by type (i.e. universal, assistance, insurance based) and need (i.e. old-
age, disability, unemployment, education, etc.), but without a reference to a person’s age.
Hence, “old-age pensions”, an EU-SILC term of social transfers, is not directly obtainable in
42
the LIS data 13. Similar classification divergences to the ones of “old-age pensions” are noted
for the other types of social transfers, such as disability pensions, family allowances, or
unemployment benefits.
Table 2. Composition of gross and disposable incomes in EU-SILC and LIS databases
EU-SILC LIS
+ Gross employee cash or near cash income Regular and casual (monetary and non-monetary) paid employment
+
+ Company car (not in 2005) income
Gross cash benefits or losses from self-employment, Self-employment income (farm and non-farm: profit from business
+ +
including royalties & household production activities)
Interests, dividends, profit from capital investments
+ in unincorporated business (minus associated
Capital income: interest and dividends, voluntary individual
expenses); +
pensions, rental income, royalties (minus interest paid).
Income from rental of a property or land (minus
+
associated expenses);
+ Old-age benefits Assistance benefits: general social assistance; old-age
+ Housing allowances /disability/survivors assistance pensions; unemployment assistance;
+
family/maternity/child assistance; education, housing, heating, food
and medical assistance.
+ Social exclusion not elsewhere classified
GI = …
The two datasets also have a different treatment/reference of/to certain income components.
For example, in EU-SILC gross incomes include monetary receipts by people under the age of
16. This income type does not exist in LIS. In the latter dataset, on the other hand, income
from household production activities is viewed as self-employment income and included in
household gross incomes. EU-SILC excludes this income type in the derivation of gross and
disposable income types. The same applies to voluntary private pensions, which is not (yet)
included in the calculation of total household gross and disposable incomes in EU-SILC. The
LIS data refers to voluntary individual pensions as a type of capital income and includes it in
the concepts of both gross and disposable incomes. Finally, disposable income in both
datasets is calculated by subtracting withholding income taxes and tax adjustments, as well as
social contributions. In the EU-SILC, however, further subtractions are also made: regular
taxes on wealth and regular inter-household cash transfers paid. Though observed in the LIS
13
The EU-SILC equivalent category of old-age pensions has been created in the LIS data for some of the
countries that derive their LIS information from the EU-SILC survey.
43
data (e.g. such income types as property/municipality taxes, other direct taxes, inter-
household transfers paid), these expenditure types are not used in deriving disposable income.
To what extent these differences in gross and disposable income structures influence their
monetary values? Below, we provide a quick comparison of average household incomes (non-
equivalised and unweighted) across the countries that derive their information from the EU-
SILC survey and across the countries with diverging data information sources. For the first
group of countries, any observed differences should be mainly due to differences in
terminology and data provider’s imputations. For the second group of countries, an
information source can add an additional explanatory power in the observed differences.
We, actually, find no major differences in the two income estimates - average gross and
disposable household incomes – between the LIS and EU-SILC datasets for the countries that
rely on the SILC survey to draw their data. This implies that methodological differences
between the datasets do not cause any significant deviations in income estimates. One should
note, though, that this observation is made for the major aggregate income types as gross and
disposable incomes, while larger deviations can be possible for the other income types or
estimates.
The picture is very different for the remaining countries: a divergent data information source
implies that highly varying income levels are observed between the EU-SILC and LIS
datasets. Actually, only two countries, namely France and Norway, have income estimates,
which are very close in their values. In Germany and Sweden, EU-SILC records both higher
gross and disposable incomes than LIS, with an extreme gap (more than 20 percentage points)
for Swedish disposable incomes. For Denmark, Slovenia and UK, EU-SILC provides higher
mean values both for gross and disposable incomes. This difference is extremely large in the
44
Slovenian case. This is partially explained by the fact that LIS collects only net incomes for
Slovenia, whereas EU-SILC provides with the “true” gross information. Nonetheless, aside
the issue of taxation, the difference in Slovenian disposable income is very high across the
two datasets. Hungarian data is also reported “net” in LIS dataset, making it difficult to
compare to gross incomes of the EU-SILC. Hungarian disposable incomes are higher in LIS
dataset. Overall, this comparison of gross and disposable income shows that divergent income
estimates when obtained from the two datasets are mainly due the diverging underlying
information sources rather than to somewhat different classification of gross and disposable
incomes.
45
Appendix 2 – “Original” studies
Figure A.1: Index of Targeting of Transfer Income and Income Redistribution: 11 OECD countries,
mid 1980s
46
Figure A.2: Redistribution by targeting-universalism: across countries at common points in time
47
Appendix 3 – Other sensitivity checks
Figure A.3. Reduction in inequality due to taxes and transfers compared with size of social transfers
(cash social spending expressed as a share of GDP), mid 2000s.
50
SWE FIN
DNK BEL
NLD HUN DEU
NOR CZE LUX POL AUT
Redistribution index
40
IRL SVN FRA
AUS ITA
GBR
EST
30
ESP
CAN ISR GRC
CHE
USA
20
5 10 15 20
Generosity (cash social spending as % of GDP)
Source: OECD SOCX 2005 for cash social spending as a share of GDP.
Figure A.4. Targeting and generosity (cash social spending as % of GDP), mid 2000s. 20
Generosity (cash spending as % of GDP)
AUT
FRA
ITA
BEL DEU
FIN POL
15
SWE SVN
DNK LUX GRC
HUN ESP
CHE
CZE NLD
NOR
10
GBR
ISR
IRL EST
AUS USA
CAN
5
-.4 -.2 0 .2
Targeting (Concentration index, ranking on gross income)
Source: OECD SOCX 2005 for cash social spending as a share of GDP.
48
Figure A.5. Concentration coefficients versus bottom quintile transfers share
as % of total transfers
AUS
DNK
35
GBR CHE
CAN
IRL BELFIN
CZE NOR ISRNLD
25
SWE USA
EST
DEU SVN
FRA
HUN AUT
LUX
15
ESP GRC
ITA
POL
5
-.4 -.2 0 .2
Targeting (Concentration index, ranking on gross income)
Figure A.6. Concentration indices when ranked by gross and market income
ITA ESP
LUX FRA
SVN HUN
NLD USA
DEU
-.1
ISR EST
FIN SWE
BEL NOR
CHE CZE IRL
CAN
-.3
GBR
DNK
AUS
-.5
49
Table A.1: Gini coefficients and indices of redistribution, generosity and targeting, LIS and EU-SILC
data, mid 2000s
50