Regimes and Cultures of Social Security: Comparing Institutional Models through Nonlinear PCA

Abstract

This article reassesses the link between the structural and cultural aspects of social security. Do Esping-Andersen’s ‘Three Worlds’ exist empirically if one considers a comprehensive set of formal institutions simultaneously? And if so, do such regimes coincide with coherent differences in people’s value orientations in this field, or informal cultures? In order to answer these questions, nonlinear principal components analysis was applied to a group of countries at the core of the original Esping-Andersen typology. Nonlinear PCA seems to be a promising tool for comparative research because the technique is able to handle discrete data and nonlinear relationships, and the number of variables can exceed the number of countries. The outcomes of the analyses suggest that the ‘Three Worlds’ of formal social security had a firm empirical basis in the 1990s, and that the typology remains largely valid today, albeit with some qualifications. Furthermore, three different informal ‘cultures of social security’ emerged, with country clusters quite similar to those of the structural regime typology.

Full text

9
Regimes and cultures of social security:
Comparing institutional models through nonlinear PCA
J Cok Vrooman
The Netherlands Institute for Social Research|SCP, The Netherlands
c.vrooman@scp.nl
The final version of this paper was published in International Journal of Comparative Sociology, vol. 53 no. 5-6: p. 444–477 (2012).
Abstract
This article reassesses the link between the structural and cultural aspects of social security. Do
Esping-Andersen’s ‘Three Worlds’ exist empirically if one considers a comprehensive set of
formal institutions simultaneously? And if so, do such regimes coincide with coherent differences
in people’s value orientations in this field, or informal cultures? In order to answer these
questions, nonlinear principal components analysis was applied to a group of countries at the core
of the original Esping-Andersen typology. Nonlinear PCA seems to be a promising tool for
comparative research because the technique is able to handle discrete data and nonlinear
relationships, and the number of variables can exceed the number of countries. The outcomes of
the analyses suggest that the ‘Three Worlds’ of formal social security had a firm empirical basis in
the 1990s, and that the typology remains largely valid today, albeit with some qualifications.
Furthermore, three different informal ‘cultures of social security’ emerged, with country clusters
quite similar to those of the structural regime typology.
Keywords
Comparative social policy
Nonlinear principal components analysis
Social security institutions
Welfare attitudes
Welfare regimes

Over the past two decades, Esping-Andersen’s distinction between liberal, social-democratic and
corporatist welfare regimes has dominated much of the comparative study of social policy
(Castles et al., 2010). While several new research avenues are currently opening up (see e.g.
Ferragina and Seeleib-Kaiser, 2011), two classic issues are not yet fully settled. In spite of the
abundance of empirical work – already designated as a ‘welfare modelling business’ by
Abrahamson in 1999 – most tests of the typology covered its theoretical elements only partially.
In addition, the relation- ship between the structural aspects of Esping-Andersen’s regime types
and the ‘culture of welfare states’ remains puzzling, as previous research suggested a certain
correspondence, but no straight- forward connection (Svallfors, 2010). This article seeks to shed
some light on these issues for a group of countries that are central to the original Esping-
Andersen typology. Nonlinear principal components analysis is applied to a comprehensive set of
formal social security and labour market traits, and to various indicators of informal social
security cultures. The next two sections provide an overview of the theoretical and empirical
literature on regimes, and on their relationship with cultural distinctions at national level; based
on this overview, two research questions are formulated. The third section outlines the
methodology, measures and data used. The empirical outcomes on social security regimes and
cultures are then presented; these are discussed in the concluding section.
Limited coverage of the ‘Three Worlds’
In The Three Worlds of Welfare Capitalism, Esping-Andersen (1990) identified a typology of
divergent models of social security and the labour market that emerged in industrialized nations
after the Second World War. He regards these three ‘regimes’ as structurally different ways of allo-
cating welfare production between state, market and households (Esping-Andersen, 1999: 73). The
liberal1 regime features limited collective schemes but a well-developed private insurance system
for the middle classes. De-commodification is low, and social stratification tends to take shape in
three separate layers: welfare clients and the working poor; the middle classes; and the privileged.
The social-democratic regime is theoretically far more extensive. Social benefits are available for
all inhabitants, at a level appealing to the middle classes; private insurance is less important. The
elaborate social security schemes require high taxation and near-full employment of both sexes,
partly realized through a large services sector. De-commodification is high and the system aims to
reduce social inequalities. The corporatist (or ‘conservative’) regime is theoretically well devel-
oped, but on a more selective basis than the social-democratic model. Rights are tied to paid con-
tributions and past labour experience, thus favouring older ‘insiders’; occupational groups have
separate benefit schemes which correspond to their social position (e.g. generous provisions for

civil servants). Families with children are well protected through collective insurance, which
results in low female labour market participation. De-commodification is moderate, and in
terms of stratification the corporate regime seeks to reproduce the social inequality between
status groups.
Although Esping-Andersen’s classification became highly influential over the years, various
alternatives and extensions have been proposed (see Arts and Gelissen, 2010; Ferragina and
Seeleib-Kaiser, 2011).2 The typology has also been criticized for its focus on the traditional eco-
nomic perils faced by the male breadwinner employee (Stephens, 2010) and its neglect of new
social risks (e.g. single parenthood, the reconciliation of work and family life; see Bonoli, 2006),
gender issues, and welfare production within families (Orloff, 1996; Sainsbury, 1999; see however
Esping-Andersen, 2009). The same applies to the position of employers operating under different
production regimes, or ‘varieties of capitalism’ (Hall and Soskice, 2001; Kitschelt et al., 1999;
Leibfried and Mau, 2008).
In order to substantiate his typology, Esping-Andersen (1990) presented various comparisons of
different sets of OECD countries. His data mainly concerned income replacement schemes, labour
market characteristics and employment incentives. These related not only to institutional traits
(coverage, levels, durations and conditionality of benefits; tax rates; social expenditure; wage pres-
sure), but also to several outcome indicators (unemployment and labour market participation rates;
the structure of employment). Yet Esping-Andersen’s empirical work received a fair amount of
methodological criticism. One objection concerned the limited validity, reliability and accessibility
of his data (Bambra, 2006; Scruggs and Allan, 2006, 2008). Another objection related to his use of
unnecessarily ‘soft’ or less adequate statistical techniques: tabular presentations, additive indices
and regression analyses (Shalev, 2007). Later authors tested the typology more rigorously by gen-
erating new data and by applying more advanced methods (mostly cluster or factor analysis); see
e.g. Kangas (1994), Ragin (1994), Obinger and Wagschal (1998), Gough (2001), Hicks and
Kenworthy (2003), Saint-Arnaud and Bernard (2003), Powell and Barrientos (2004), Castles and
Obinger (2008) and Schröder (2009). In their overview, Arts and Gelissen (2010: 577) conclude:
‘[Esping-Andersen’s] typology has at least some heuristic and descriptive value, but a case can be
made for extending the number of welfare state regimes, perhaps to four or even five. […] Some
cases are (much) more impure than others, and some are clearly hybrids’.
However, there are several reasons for doubting whether the foregoing empirical analyses pro-
vide an adequate test of the typology. First, the institutions included in these studies often did not
cover the theoretical regime traits completely. For instance, Ragin (1994) analysed pension bene-
fits, Gough (2001) looked at social assistance schemes, and Kangas (1994) focused on health
insurance. Such partial analyses can neither corroborate nor rebut Esping-Andersen’s classifica-
tion, which refers to the entire configuration of social security and labour market institutions.

Ferragina and Seeleib-Kaiser (2011) argue that many deviations from the typology reported in the
literature are simply the result of including traits from other policy fields (education, health care)
that should be discarded on theoretical grounds.
A second problem concerns the limited number of indicators in most studies. Bonoli’s classifi-
cation (1997) is an extreme example: he attempted to test the typology by plotting two variables,
social expenditure and the share of contribution-funded social security (updated by Arcanjo, 2011).
But in other cases too, often only between six and ten variables were analysed, mostly relating to
expenditure data. This hardly does justice to the wealth of attributes that Esping-Andersen associ-
ated with his regime types.
A third objection is that most empirical assessments of the typology analyse continuous varia-
bles (e.g. expenditure and benefit levels), on the assumption of linear correlation. In reality, rele-
vant regime traits can consist of discrete categories (e.g. various benefit conditions), or may be
related in a nonlinear way.
These limitations imply that a new analysis, with a large set of traits covering the entire regime,
and use of a technique more suited to the identification of different types, could have added value.
Our first research question is therefore: Do Esping-Andersen’s ‘Three Worlds’ exist empirically if
one considers a comprehensive set of formal institutions simultaneously?
The connection between regimes and culture
Esping-Andersen’s typology stems from ‘power resources theory’ (Korpi, 1983, 1989; Stephens,
1979). This approach posits that regimes are the outcome of the distribution of power across social
classes, their interests and degree of organization, and the political struggles and coalitions which
occur at historical turning-points.3 Following this line of reasoning, regimes are thought to be sta-
ble because they solve problems across class and interest cleavages. Thus the Swedish social-
democratic regime is considered to be the result of high worker mobilization (strong trade unions),
a broad alliance of social democrats and well-organized farmers (who traded an extension of ben-
efit schemes in return for farm price subsidies), and successful incorporation of the new middle
class into social security (Esping-Andersen, 1990).
In power resources theory the values, norms and beliefs of actors are generally regarded as a
cultural spin-off of the structurally diverging regime features. Living under a specific regime may
cause people ‘to adhere to a particular conception of a moral community’ (Arts and Gelissen,

2001), with different notions on equality, equity, solidarity, justice principles, etc. Van Oorschot
et al. (2008: 1–25) situate this assumption in the more general materialist tradition in sociology
(Marx, and to some extent Bourdieu), where culture is treated as a reflection of the social system.
It contrasts with the idealist perspective (Sorokin, Lévi-Strauss), which posits that culture deter-
mines the social structure through the socialized convictions and symbolic acts of individuals. In
the welfare state literature, the ‘national value hypothesis’ provides an example (see Coughlin,
1980; Lipset, 1963). Here, diverging value orientations rooted in national history are thought to
bring about different types of welfare regimes. Accordingly, countries where egalitarian attitudes
are comparatively strong tend to foster extensive welfare states, while individualistic nations – the
USA being the prime example – only institute collective provisions if external circumstances (such
as the Great Depression) force them to do so.
Both approaches imply a rather one-sided and over-deterministic perspective on the relationship
between the structure and culture of modern welfare states. Power resources theory largely rejects
the notion that ideas alone can shape welfare regimes. The national value hypothesis, on the other
hand, postulates homogeneous and stable cultures, has difficulty in explaining policy changes, and
disregards the role of interests, powers and political negotiation in welfare regime development
(Brooks and Manza, 2007). Pfau-Effinger (2005), Larsen (2006) and Van Oorschot et al. (2008)
therefore suggest that welfare regimes and cultures be regarded as dynamically interrelated and
relatively autonomous. This is in line with Castles’s (1993) ideas on ‘families of nations’ and, more
generally, with the ‘new institutionalism’ that emerged in recent decades. That approach combines
the historical analysis of institutions with economic theory on property rights and transaction costs,
and with sociological theory on the social construction of cognitions and morality (see Brinton and
Nee, 1998; Hall and Taylor, 1996; Morgan et al., 2010; North, 1990). In new institutionalism struc-
tural regimes and cultural orientations tend to coincide for three reasons (Vrooman, 2009). First,
modern social security rules on the one hand, and values, norms and conventions on the other, may
both be thought of as ‘humanly devised constraints that shape human interaction’ (North, 1990: 3).
They are institutions that aim to regulate behaviour through a specific allocation of rights and
duties and related conditions and sanctions. Although modern social security is largely formalized
in national and international law, informal elements are still important because they fill in the gaps
that legislation does not cover. Formal and informal rules thus tend to serve as functional exten-
sions of one another; they are two sides of the same institutional model, and are therefore likely to
correspond. Furthermore, coherent views on the way society should be typically played a role in
the vesting of regimes. Those regimes do not merely reflect the power-weighted interests of their
founders (see Coleman, 1990) and the idiosyncrasies of the bargaining process, but also the convic-
tions of the parties involved. Such ideologies may persist to the present day in the form of a ‘col-

lective memory’ among current policymakers and the general public. Finally, once regimes emerge,
people’s attitudes towards welfare provision may block fundamental regime shifts. Path depend-
ency is partly rooted in the cultural ‘rules of the game’, which make the existing regime seem a
natural order, and may affect social policy through voting behaviour, opinion polls and media
coverage (Brooks and Manza, 2007). However, as stressed in policy feedback theory (see Hacker,
2004; Pierson, 1993; Weaver, 2010), the opposite may also occur. The way in which social policies
are designed (e.g. targeted at certain groups; phasing-in processes; visibility of contributions and
benefits) and framed (e.g. by referring to ‘ticking demographic time bombs’ or to certain classes of
‘undeserving poor’) can also shape popular support for, or rejection of, reform measures.
From the new institutionalist perspective, the degree of correspondence between regimes and
cultures is an empirical matter. It may be that under liberal regimes ideas of personal responsibility,
individual freedom and aversion to state intervention are more widespread than elsewhere; that the
population of corporatist welfare states is more inclined towards conservative attitudes (an organic
view of society which preserves status group differences; traditional family values and gender
roles); and that people in social-democratic countries more frequently favour social equality and
state responsibility for every citizen’s welfare. But this is not necessarily so. The connection
between welfare structure and culture may differ between countries, and it may change over time
– though typically not always in a single direction, and not without a struggle.
Scrutinizing the empirical research on the relationship between regime types and welfare
cultures actually reveals it to be rather ambiguous (see e.g. Andress and Heien, 2001; Arts and
Gelissen, 2001; Blekesaune and Quadagno, 2003; Blomberg and Kroll, 1999; Bonoli, 2000;
Jacobsen, 2011; Jæger, 2006, 2009; Kluegel and Miyano, 1995; Mehrtens, 2004; Papadakis and
Bean, 1993; Svallfors, 2003; Walker, 2008). In his overview article, Svallfors (2010: 245) con-
cludes that ‘support for equality, redistribution, and state intervention is strongest in the social-
democratic regime, weaker in the conservative regime, and weakest in the liberal regime.
However, we do not find any clear regime clustering of countries. Differences and similarities
between countries show interpretable patterns, but they are too complex to be summarized as
“worlds of welfare attitudes”…’ This rather confused picture may be due in part to the great diver-
sity of nations, data sources and attitudes in the earlier work. In addition, countries were some-
times assigned to regime types to which they probably do not belong, or purer specimens of the
various regimes were grouped together with more diluted examples.4 This may have distorted the
correlation with the cultural domain. Finally, as in the regime analyses, welfare attitudes were
often studied under the possibly erroneous assumption of linear relationships between continuous
variables. Yet it seems unlikely that these differences and potential specification errors are so
large and consistent that they can fully explain the lack of correspondence found earlier. The main
reason may lie in a one-dimensional selection of value orientations. In most analyses these relate

to the preferred level of public spending, degree of redistribution, solidarity and similar aspects.
Indicators such as these chiefly capture the contrast between countries operating a small welfare
state (the liberal type) and countries with more extensive systems (social-democratic and corpo-
ratist types). This implies that the cultural differences within the latter group were hardly ever
included in the earlier empirical work, and therefore could not come to light. Theoretically these
should imply a contrast between universalistic, redistributive orientations in social-democratic
regimes, and particularistic, status-related attitudes in corporatist countries. Examples would be
diverging ideas on income transfers, the prestige of civil servants, gender roles, etc. Re-analysing
the relationship between formal regime types and informal cultures with a more balanced set of
attitudes may therefore be worthwhile. Our second research question is therefore: To what extent
do regimes coincide with coherent differences in people’s value orientations, or informal social
security cultures?
Methodology, data and measures
Nonlinear principal components analysis
Shalev (2007: 289) states that a proper analysis of Esping-Andersen’s typology should indicate
whether ‘welfare state regimes actually do “hang together”; and if they do, whether countries
indeed cluster in three distinct subgroups on underlying policy dimensions’. In order to achieve
this he suggests combining multidimensional scaling and factor analysis.5 Nonlinear principal
components analysis fulfils these requirements. Like classic PCA, the technique reduces variables
to a limited number of uncorrelated dimensions, but it also entails a process of optimal
quantification. Categories are assigned numerical values in such a way as to maximize the
accounted-for variance in the transformed variables (see Gifi, 1990; Linting et al., 2007; Michailidis
and De Leeuw, 1998). Nonlinear PCA has several advantages over classic PCA. It can handle
nominal, ordinal and numerical data; it does not assume a linear relationship between variables;
and the technique is ‘eminently suited for analyses in which there are (many) more variables than
objects’ (Meulman et al., 2004: 67). The latter implies that nonlinear PCA is not sensitive to the
‘small-N problem’ often encountered in comparative analysis (e.g. Goldthorpe, 2000): the items
may outnumber the generally limited country sample.6 Nonlinear and classic PCA produce identi-
cal results if variables are treated as numerical in the quantification process and if there are no
missing data (Gifi, 1990: 151–191; Linting et al., 2007). Examples of the application of nonlinear
PCA include studies on personality disorders (Eurelings-Bontekoe et al., 1996), adolescents’ lei-
sure activities (Zeijl et al., 2000), effects of obstetric complications (Arseneault et al., 2002) and
organizational cooperation in crisis situations (Svedin, 2009). In spite of its potential, the technique
has not often been applied in comparative research; but Jehoel-Gijsbers and Vrooman (2008) used

it to assess social exclusion among older people in EU countries, and Soede and Vrooman (2008)
to compare pension schemes. In the SPSS software package that was used the procedure is known
as categorical principal components analysis (CatPCA).
Nonlinear PCA has attractive features for exploring Esping-Andersen’s typology and related
cultural orientations, because here the number of countries and theoretical dimensions is small,
while the set of indicators is large and consists partly of discrete variables (nominal or ordinal
scale). When applying this technique, countries and institutional traits which share many features
will obtain similar scores on the underlying principal components, whereas cases and categories
with little in common will be positioned a long way apart. This makes it possible to ascertain the
prevalence of distinct regimes and cultures of social security.
Data and measures on regimes
With regard to the regime types, eleven national systems of social security and labour market regu-
lation were considered for the first half of the 1990s. This is probably the best period in which to
test the validity of Esping-Andersen’s regime typology. In later years the picture may have become
less clear, as many governments not only tried to contain the costs of social security according to a
‘logic of retrenchment’ (Pierson, 1996), but also attempted to strengthen work incentives and to
recalibrate their systems to meet changing social circumstances (Castles, 2010; Kautto, 2010;
Levy, 2010; Palier, 2010). The countries selected were at the core of the original ‘Three Worlds’
typology.7 Esping-Andersen (1990) places three of these countries in the corporatist group
(Belgium, France and Germany), four in the liberal cluster (Australia, Canada, the USA and the
United Kingdom), and three in the social-democratic group (Sweden, Norway and Denmark). The
Netherlands was added as an example of a hybrid regime, which theoretically combines corporatist
and social-democratic traits. An additional analysis was performed on a more limited set of varia-
bles, relating to 2008–2010 and including a larger set of countries. This enabled us to gain an
insight into the stability of the Esping-Andersen typology over time, and to assess how sensitive
the outcomes of the main analysis might be to case selection.
Nonlinear PCA was performed on 54 institutional characteristics. Data for the regime analysis
were obtained from international sources, mostly the OECD, EC and ILO. The selection of varia-
bles was based on Esping-Andersen’s (1990: 26–29) theoretical demarcation of regime traits, and
overlapped his measures of de-commodification, welfare conservatism, liberalism and socialism,
tax burdens and social expenditure. There were however some differences:

–
information was added on surviving dependants’ pensions, child benefits, social assistance,
disability, retirement ages and fiscal labour incentives. These are part of Esping-Andersen’s
theoretical framework, but were largely discarded from his tables and indices;8
–
the ‘breadwinner bias’ and the scant attention for the position of employers in the original
typology were addressed by including new data on leave arrangements, annual holidays,
activating labour market policies, the statutory minimum wage, the coverage of collective
labour market agreements, and employers’ contributions;
–
tax rates, child benefits and unemployment replacement rates were differentiated by house-
hold type and income level, in order to achieve a more balanced picture;9
–
Esping-Andersen’s indicators relating to outcomes (e.g. unemployment and participation
rates) were discarded. In our view these are not part of the regimes as such, but are the
combined result of the institutional configuration and various other country differences
(demography, national resources, the economic cycle).
Nonlinear PCA makes it possible to analyse the entire dataset simultaneously. Appendix 1 provides
a detailed overview of the regime indicators.
Data and measures on cultures of social security
In order to assess the correspondence of the regime typology and informal social security cultures,
nonlinear PCA was also performed on data from the 1996–1999 waves10 of the International Social
Survey Programme. The same countries were included as in the main regime analysis, with the
exception of Belgium, which did not take part in ISSP at the time.11 We selected 16 indicators of
liberal, social-democratic or corporatist value orientations, presumed to be in line with the regime
typology (see Appendix 2). Most items refer to the attribution of rights and duties in society and
may therefore be regarded as representative of people’s commitment to informal institutions
(Vrooman, 2009). A few variables were however simply preferences that would be in line with the
structural differences between the regime types.
The contrast between liberal regimes on the one hand and corporatist and social-democratic
regimes on the other is assumed to be captured in items relating to:
–
governmental responsibility for the standard of living of older people and the unemployed;
–
governmental responsibility for the provision of jobs;
–
governmental responsibility for income redistribution;
–
aversion to wide differentials at the top of the income distribution.

People living under liberal regimes may be expected to be less in favour of an interventionist
government than those in countries with more extensive regimes. They are also less likely to
oppose very high top incomes, as these reflect the individual’s market value and freedom to make
the most of his or her bargaining position. Such orientations may theoretically be regarded as
typifying a liberal social security culture.
Other variables might be indicative of social-democratic orientations. These include:
–
aversion to income differentials between lower and higher occupations;
–
a limited role for meritocratic performance in setting the wage level;
–
postmaterialist work orientations;
–
a preference among females for public sector employment.
As social-democratic regimes explicitly aim to reduce income inequality through vertical redis-
tribution, inhabitants of countries operating such a system may be expected to favour small income
differentials to a greater extent than elsewhere. In the same line of reasoning, it is likely that meri-
tocratic principles will be deemed a less relevant wage criterion: a better job performance should
not necessarily entitle someone to a (much) higher income.
The high degree of de-commodification pursued by the social-democratic regime could trans-
late into comparatively strong support for postmaterialist work values: if every citizen is assured
of a decent minimum income, people may more readily accept a job for motives other than finan-
cial gain. As noted earlier, the social-democratic regime requires high labour market participation
by both sexes; for women this is mainly realized in the large public services sector. It may therefore
be assumed that females have a comparatively strong preference for working in this sector.
Characteristic orientations theoretically belonging to the corporatist regime type could relate to:
– traditional gender roles;
– seniority as a wage criterion;
– the prestige of state officials;
– the acceptability of breaking the formal tax and benefit rules.
Given the prevalence of the breadwinner model in corporatist regimes, it is assumed that people
are more likely than elsewhere to prefer women to stay at home, thus legitimizing the actual state
of affairs. In similar vein, the strong emphasis on acquired rights and positions in these systems
could imply that it is considered self-evident to reward people according to seniority. Furthermore,
corporatist social insurance schemes were originally designed to reproduce status differences
under the auspices of the emerging nation-states, personified by the civil authorities. A high pres-
tige of state officials would be consistent with the extensive social security actually provided for
this group. Finally, it may be presumed that bending the formal tax and benefit rules is more

acceptable under corporatist systems. Due to the more generous benefits and higher tax rates, the
potential gains of defection are greater than in liberal countries; and since corporatist regimes aim
to serve particularistic interests, it may seem more natural for people to shun their duties than in
social-democratic regimes (where benefit fraud and tax evasion are more likely to be considered an
offence against the public interest).
In the empirical analysis some of these orientations had multiple indicators; and a few variables
were derived from the original responses. Thus, the share of people expressing a preference to
work for the government or civil service was split by gender. A high level of support among males
is assumed to indicate a high status of civil servants (theoretically a corporatist feature); while a
high ratio of female versus male support is presumed to imply a feminine bias towards working in
the public sector (characteristic of social-democratic regimes).
The inequality aversion items were based on a set of questions where respondents had to rate
the earnings they thought should apply for nine different occupations, ranging from an unskilled
factory worker to a Supreme Court judge and the chairman of a large corporation. For each country,
a distribution of preferred income differentials was constructed by concatenating the responses for
the various occupations.12 Two inequality measures were calculated for this variable. The Theil-
coefficient was used to assess the preferred earning differentials between lower and higher occupa-
tions; and the Gini-coefficient – which is less susceptible to extreme values – was used as an
indicator for the income inequality aversion regarding the six higher-level occupations. The same
set of questions was also used to generate an indicator of the prestige of state officials: the ranking
of the mean preferred income of a Cabinet minister among the nine occupations.
Empirical regimes
In the nonlinear PCA of regime traits, most variables were scaled at the ordinal level; however,
four had low fit and were treated as nominal.13 Cut-off points for continuous variables were based
on gaps and concentrations in their distribution; the impact of this is discussed in a sensitivity
analysis below. The initial categorization and component loadings of the variables are listed in
Appendix 1.
Interpretation of the dimensions
The first two components account for 62 percent of the total variance. Adding more dimensions
leads to a limited increase in the proportion of explained variance and does not produce any new
substan- tive insights.14 The component loadings suggest three major clusters of variables. The first
group has high negative loadings on the first dimension, but fairly low ones on the second
(numbers 1–26 in Appendix 1). Substantively, the horizontal axis is largely associated with variables

measuring whether a system is residual or extensive. Lower scores indicate frugal benefits of
limited duration combined with strict entry regulation through means-testing. As a consequence,
expenditure and the funding required are low. In Esping-Andersen’s terms this is mainly a
combination of de-commodification, welfare liberalism, the social wage and employment
incentives through taxation. More specifically, a scaling on the negative side of the horizontal axis
corresponds to low replacement rates in the first and fifth year of unemployment. Child benefits are
meagre and earnings-related occupational pensions substitute only a small part of previous income.
There is no link between the duration of the insurance period and the level of collective old age
pensions. Earnings-related unemployment benefit is usually paid for a short period, with a
maximum of one year. In disability schemes, the risque professionnel (medical impediments due to
occupational activity, accidents or diseases) is only insured for employ- ees, not for the self-
employed, trainees, interns, etc. Coverage of the risque social is limited to people who are almost
fully incapacitated (80–100%). With regard to leave arrangements, a low score on the first
dimension points to a small proportion of working women with entitlement, a low level of mater- nity
benefits, and the absence of earnings-related leave schemes. A large part of total social security
spending concerns means-tested social assistance; child benefits and collective old age and surviv-
ing dependants’ pensions also are conditional on the absence of other sources of income or wealth.
Taxes and social security contributions make up a relatively small part of GNP. Average tax rates
for standard workers are low, as are the marginal rates for single people and double-earners and
employers’ contributions. Tax breaks for work-related expenses are limited.
For two variables in this group (numbers 5 and 26 in Appendix 1) the component loadings –
though rather low – are contrary to expectations. While a high statutory minimum wage and the
existence of a specific orphan’s pension are theoretically traits of more extensive regimes, these in
fact coincide with residualist system characteristics.
The second major group of variables show high negative component loadings on the vertical
axis, and mostly positive values on the horizontal axis (numbers 27 to 39 in Appendix 1). It com-
prises features which Esping-Andersen theoretically linked to corporatist regimes: welfare con-
servatism; a close connection between employment or past labour experience and benefit
entitlements; and work disincentives for females. Welfare conservatism is expressed in a high
degree of occupationalism. There are separate schemes for different groups of employees, aimed at
maintaining their standard of living; and expenditure on social security for civil servants tends to
be higher than elsewhere. This corresponds with a considerable share of contribution-based fund-
ing, reflecting a strong emphasis on social insurance principles. On the labour market, employment
contracts are largely covered by collective agreements.

The corporatist selectivity in protecting ‘insiders’ on the labour market is evident in disability
schemes. Benefits are high for employees who become incapacitated due to their work; under some
risque professionnel schemes, they may even continue to receive their full salary. On the other
hand, the coverage of the risque social is limited. Typically, those with early onset disability, the
self-employed, students, housewives and unemployed persons who become incapacitated are not
entitled to any disability benefits. These traits coincide with fairly good (semi-)collective pensions
for widows and orphans of employees, with no limits on duration. Surviving dependents of persons
without an employment contract, however, have to rely on their own means or on social assistance.
Child benefits are generally linked to family size, with no specific targeting of low-income groups.
High child benefits are granted to large families which cannot be considered needy, while poor
lone-parent families do not receive special protection. Facilities for mothers are comparatively
limited. Maternity and paternal leave is fairly long, but is largely unpaid. It is combined with low
coverage of formal childcare facilities and high tax allowances for couples with children.
Nonetheless, the additional tax advantages single-earner families receive (over single persons)
that are not related to children are limited. This suggests that labour disincentives are not linked
to having a partner, but rather to the presence of children.
A third group of variables combines high positive component loadings on the second dimen-
sion with positive loadings on the first. These traits, numbered 40–49 in Appendix 1, are theo-
retically associated with social-democratic regimes: generous provisions with strict entry
regulation, activation and elaborate leave arrangements. Universalism is most evident in pen-
sions. All inhabitants are entitled to a collective old age benefit, at a comparatively generous
minimum level; and in addition most employees are covered by a compulsory earnings-related
pension scheme. These favourable provisions for older people are combined with a rather high
statutory retirement age (65 years or more) and limited opportunities for full early retirement.
Entry to disability schemes is also restricted: a minimum incapacity threshold is applied, even
where disabilities and illnesses are work-related. The safety net is wide: after five consecutive
years of unemployment, social assistance benefits for families with children are comparatively
high. This is however accompanied by considerable efforts to get people back to work, as indi-
cated by the substantial expenditure on activating labour market policy. Working people enjoy
extensive leave arrangements; the statutory minimum annual leave is more than four weeks. As
a corollary to the good level of provisions, tax and contribution rates tend to be high, especially
for single earners with children.
One variable (number 50 in Appendix 1) does not belong to any of these groups. In most coun-
tries, entitlements to disability schemes covering the risque social are earnings-related; but the UK,
Australia and Denmark operate flat-rate benefits, resulting in a component loading between the

liberal and social-democratic groups of variables. The four remaining variables were scaled as
multiple-nominal and do not have component loadings (see note 13).
All in all, the component loadings indicate a clear interpretation of the two dimensions, substan-
tiating a hypothesis put forward by Shalev (2007: 291–292). On the horizontal axis, we find an
underlying continuum stretching from residualist (negative loadings) to extensive (positive load-
ings) regime characteristics. This resembles what is sometimes referred to as the scope of social
security, or the social security effort. Yet the first dimension reflects much more than relative
expenditure, the commonly adopted effort indicator (e.g. Wilensky, 1975): it also indicates de-
commodification, welfare liberalism and fiscal incentives to work. On the vertical axis we mainly
see a contrast between ‘particularistic’ and ‘universalistic’ traits; hence, this can be regarded as a
principal component indicating the degree of universalism.
Regime clusters
Figure 1 shows the object scores that were obtained on the two principal components. There are
three distinct country clusters and one intermediate case. First, there is a liberal group, comprising
the United States, Australia, Canada and the UK. These countries score low on scope, and are
scaled about average on the vertical axis, mostly with a bias toward the universalistic side. Based
on the characteristics studied here, the USA was the most ‘residual’ country during the first half of
the 1990s, closely followed by Australia. Canada and the United Kingdom attained somewhat
higher scores on the first dimension and may be regarded as less typical representatives of the
liberal type. This was due to higher replacement rates in unemployment and social assistance
schemes than in the USA and Australia, better income provisions covering the cost of children, and
higher minimum collective old age pensions for persons who had never worked. Tax and contribu-
tion rates in Canada and the UK were correspondingly higher as well.
The second cluster consists of three countries which Esping-Andersen designated as corporatist.
The Belgian, French and German systems are much wider in scope than the liberal regimes, and
these countries achieve fairly extreme scores on the particularistic side of the second dimension.
Although it is the most homogeneous cluster, there are a number of differences within it: France
had the most residual system in the early 1990s, and Germany was somewhat less corporatist than
the other two countries. Each case also diverges in some respects from the theoretical expectations.
In Belgium, unemployment insurance benefits were lower than is customary in corporatist coun-
tries, while France had fairly substantial social assistance regulations, typical of liberal regimes.

Figure 1. Optimal scaling of 11 countries on formal regime traits (nonlinear PCA, object scores).
Sources: see Appendices 1, 4.
German disability benefit schemes were comparable in many respects to what one would expect
in a social-democratic regime.
The Nordic countries of Europe form a third group. These countries score higher than the
corporatist countries on the first dimension, while on the second they surpass the liberal cluster.
This means that countries operating a social-democratic regime have the most extensive and
universalistic schemes. In the first half of the 1990s, Sweden and Denmark were the purest rep-
resentatives of this type, although both had a specific bias: the Swedish system was more exten-
sive, while the Danish regime was more universalistic. Norway comes out lower than Sweden on
the first dimension and lags behind Denmark on the second, suggesting that it was a less typical
example of the social-democratic regime type at the time. The smaller scope of the Norwegian
system is mainly due to slightly lower replacement rates and more modest contribution rates.
The lower universalism score chiefly results from Norway’s voluntary earnings-related occupa-
tional pensions (generating lower coverage) and the smaller amount spent on activating labour
market policy.
The Dutch system emerges as a hybrid. In the early 1990s, the Netherlands was rather similar to
the Nordic countries on the ‘scope’ dimension: slightly less extensive than the regimes of Denmark
and Sweden but comparable to Norway and outranking the corporatist countries. This stems mainly

from the generous benefits and correspondingly high costs and tax rates. The Netherlands achieves
an average score on the second axis, about halfway between the social-democratic and corporatist
groups. While two decades ago the scores on the social assistance and old age pension variables
were quite ‘Scandinavian’, in a number of other respects the Dutch system resembled a corporatist
regime: the privileged position of the breadwinner; the large share of social insurance contribu-
tions; the generous schemes for civil servants; and the disincentives to the female and older labour
supply. Dutch social security also showed a number of unique features, such as the absence of a
distinction between social and occupational disability benefits. The scaling confirms Esping-
Andersen’s (1999: 88) characterization of the Netherlands as ‘a Janus-headed welfare regime,
combining both social-democratic and conservative attributes’.
Cluster analyses of country scores
The foregoing analysis raises a number of methodological issues. The first concerns the classi-
fication of countries into the various regime types. The ellipses in Figure 1 are based on the sign
of the scores on the two underlying dimensions and their distances from the origin. To obtain a
more substantiated classification, hierarchical cluster analyses of the country scores were per-
formed. A two-cluster solution results in a contrast between the liberal nations (Australia,
Canada, UK, USA) and the rest. With three clusters specified, the non-liberal group is split into
corporatist (Belgium, France, Germany) and social-democratic (Denmark, Norway, Sweden)
ones. The Netherlands is included in the latter, but emerges as a distinct case in a four-cluster
solution. The classification was corroborated in a more formal way as well, by applying latent
class cluster analysis. This technique allows a test of the differences in the goodness of fit of
various cluster solutions (see Vermunt and Magidson, 2002). A two-cluster model once again
shows a distinction between liberal and non-liberal regimes. The fit, however, is not signifi-
cantly different from the baseline model, which contains a single cluster (Δ log-likelihood2-1=
4.5/3 d.f., p = .21). A three-cluster model distinguishes the liberal, social-democratic (plus the
Netherlands) and corporatist countries. Its fit is significantly better than both the baseline and the
two-cluster model (Δ log-likelihood3-1= 12.8/6 d.f., p = .04; Δ log-likelihood3-2= 8.3/3 d.f.,
p = .04). Due to the small number of cases, LCA models with four or more clusters are not identi-
fied and could not be tested.
Sensitivity analysis: Nonlinear and classic PCA
A second issue relates to the merits of nonlinear PCA compared to the classical method. In order
to assess this, a sensitivity analysis15 was carried out on a subset of ten items, covering various
regime aspects; clearly, classic PCA cannot be performed on the entire dataset because the number

of observations is too small. A nonlinear PCA of these 10 variables results in component loadings
very similar to those of the original solution. However, due to the smaller number of variables it is
easier to fit: the explained variance (75%) is higher than before (see Table 1).
If we apply classic PCA to these 10 items (with the same categorizations), the explained vari-
ance drops to 67 percent and the component loadings decrease; but their structure – and therefore
the interpretation of the dimensions – remains the same. In a third analysis, the six variables that
may be regarded as continuous were entered as raw scores in a classic PCA. Three of the items then
have considerably lower component loadings on the second dimension, and the explained variance
is 10 percentage points below that obtained in the nonlinear solution (65%). The country clusters
remain visible in both classic PCAs, but the dispersion within the liberal and corporatist groups
increases; and the Netherlands now inclines towards the corporatist side. Thus, the merits of non-
linear PCA for comparative research seem to go beyond its ability to analyze a set of variables that
outnumbers the country sample. The outcomes of the sensitivity analysis suggest that it may result
in a considerable fit improvement compared with classic PCA, and in a clearer identification of
country clusters.
Stability over time and case selection
The outcomes depicted in Figure 1 are a snapshot of the institutional configuration as it prevailed in
the 1990s. This of course raises the question of whether the regime classification would be appropri-
ate today. For various reasons some convergence is likely to have occurred, as countries faced simi-
lar challenges (the need for retrenchment; globalization and international competition) and may
have learned from each other through policy exchange and international coordination. As a result
the distinction between the three types may be less clear-cut in later years. However, from a theoreti-
cal standpoint it is equally possible that the structural differences were preserved in a process of
path-dependent change. The previous analyses were moreover limited to countries that were consid-
ered rather ‘pure’ examples in the original Esping-Andersen typology (with the exception of the
Netherlands). It may of course be that the inclusion of other, possibly more mixed countries would
blur the picture. The results of an additional nonlinear PCA (Appendix 3) provide some information
on the potential vulnerability of the typology to the chosen time frame and to case selection. The 10
variables used in the sensitivity analysis were updated16 to 2008–2010 for the same countries as
before, plus six new ones: New Zealand and Ireland (possibly liberal), Iceland and Finland (possibly
social-democratic), and Austria and Switzerland (possibly corporatist). Thus, all the countries in
Esping-Andersen’s original de-commodification and stratification indices are included, except
Japan and Italy.17 The 17 recent observations were added as supplementary objects to the nonlinear
PCA of the reduced 1990s dataset. This implies that these cases do not influence the component

loadings, but are fitted into the solution afterwards (Meulman et al., 2004: 50); both periods are
therefore measured against the same criteria. In 2008–2010, most countries turn out to be scaled in
the quadrant to which they belong theoretically. This indicates consistency and persistence of regime
differences, rather than overall convergence: the typology does not collapse. But this conclusion
requires some qualification. If we look at the 11 ‘old’ countries, the contrasts over time remain fairly
stable on the first dimension (liberal versus non-liberal),18 but diminish on the second dimension.
France, Belgium and Germany turn out to be less selective, while Sweden and Denmark emerge as
less universalistic than before.19 Although structural differences remain, this suggests some conver-
gence between the corporatist and social-democratic systems of Western and Northern Europe.
Secondly, while Ireland and New Zealand certainly belong to the liberal group in 2008–2010, and
Iceland is placed squarely in the social-democratic cluster, as expected, three other ‘new’ countries
emerge less clearly from the analysis. The Finnish system seems a rather hybrid form of the social-
democratic and corporatist types, while Austria and Switzerland are placed at some distance of the
corporatist countries.20 However, this is not a new result. Finland and Switzerland are also listed as
less pure cases in Ferragina and Seeleib-Kaiser’s (2011) meta-analysis of empirical regime typolo-
gies. Esping-Andersen (1990) classified Austria as social-democratic in terms of de-commodification
(see also Bambra, 2006), but as conservative with regard to stratification. Esping-Andersen’s typol-
ogy therefore seems to be still relevant today, albeit with some reservations; and it probably stretches

beyond the cases selected in Figure 1. Even so, it is not engraved in stone; and a similar analysis
based on more items, time points and countries could shed more light on the actual evolution of
regimes. This, however, is not the aim here.
Cultures of social security
In the nonlinear PCA of 16 value orientations, we used the categorized mean country scores on
each variable as input. These were scaled at ordinal level.21 Appendix 2 lists the categorizations
and the component loadings that were obtained.
Interpretation of the dimensions
The first two principal components account for 72 percent of the total variance.22 Three distinct
groups of variables emerge from the component loadings. The first dimension shows a cluster of
five variables with negative loadings (items 1–5 in Appendix 2). Theoretically, these are in line
with a liberal welfare regime. A low score on the x-axis indicates that comparatively few people
think the government should be responsible for providing a decent standard of living for the elderly
and the unemployed, for providing jobs and for reducing income differentials. This corresponds
with a limited aversion to inequality at the top of the income distribution, as indicated by the sub-
stantial value of the Gini-coefficient calculated over the preferred earnings for six higher-level
occupations.
The second cluster of variables has high positive loadings on both dimensions (items 6–10). In
principle, these value orientations fit in well with the social-democratic regime. The earnings ine-
quality people think is acceptable between lower and higher-level occupations is comparatively
low, and meritocratic wage criteria are considered less relevant than elsewhere. Furthermore, a
relatively high proportion of people would still enjoy their jobs if they did not need the money.
Postmaterialism also shows in the fact that relatively few respondents think a job should offer good
career opportunities. Such attitudes coincide with a comparatively strong inclination by women to
work in (semi-)government jobs.
The third group of items (numbers 11–16 in Appendix 2) combines high negative loadings on
the second dimension with moderately positive ones on the first. Theoretically, these character-
istics are mainly associated with the corporatist regime. A fairly high proportion of the popula-
tion agree that family life would suffer if women worked full-time. People believe more than
elsewhere that seniority should be a wage criterion. The high ranking of a Cabinet minister in the
preferred earnings distribution suggests a considerable prestige of state officials; this is corrobo-
rated by the comparatively high proportion of men who would prefer to work in the public sec-
tor. Bending the rules is more accepted than elsewhere: a fairly high share of respondents think

that tax evasion or obtaining benefits by giving false information is not – or only slightly –
wrong.
Summarizing, the first dimension measures (non-)liberal orientations: the degree to which the
population favour individual freedom or collective solidarity in social security and on the labour
market. It chiefly shows contrasts relating to preferred government (non-)intervention in this
field, and to the (non-)acceptance of income inequality. These are associated with some other
juxtapositions: meritocratic versus non-meritocratic wage allocation, materialism versus postma-
terialism, and low versus high preferred earnings of a Cabinet minister.
The second dimension is an amalgam of social-democratic and corporatist orientations. It
mainly reflects a distinction between support for vertical redistribution and certain related
values (postmaterialism, non-meritocratic wage allocation) and preferences (female public
sector employment) on the one hand; and the inclination to make status distinctions (civil
servants, seniority, traditional gender roles) plus lenient rule interpretation on the other.
Cultural clusters
The scale values of the countries on the principal components are presented in Figure 2. In line
with the regime analysis, the first dimension reveals a contrast between countries with liberal
value orientations and the rest, while on the second dimension, countries with social-demo- cratic
and corporatist orientations are set apart. These groups of countries have different coherent sets
of informal institutions and preferences, suggesting rather fundamental cultural differences.
A liberal social security culture emerges in the USA, Canada, Great Britain and Australia. In the
1990s fewer people than elsewhere thought the government should be responsible for ensuring a
decent standard of living for the elderly and unemployed, for the provision of jobs and for reducing
income inequality. They were also less averse to income differentials at the top. Within this group
of countries, the scale values of the United States and Great Britain are consistent with their posi-
tions in Figure 1. The USA had the most residual social security regime in the 1990s, and this is
backed up by the value orientations of its population, which were generally the most liberal of all
the countries studied. In line with the UK’s less pure social security regime, the British were more
in favour of government interventions and smaller income differentials than the inhabitants of
other liberal countries. Yet the positions of Canada and Australia in Figure 2 are not entirely in line
with the regime scalings. The Canadian social security system was not a perfect example of the
liberal type; but the Canadians’ value orientations were quite liberal. In Australia the opposite
occurs. Here, the residual social security regime concurred with orientations that were less liberal
than in the USA and Canada. This can only partially be explained by the fact that some data for
Australia were missing.23

Sweden and Norway are scaled close to each other, with high scores on the positive side of
both the first and the second dimension. Typical for their social-democratic culture is that a
large proportion of people had a strong aversion to earnings inequality between higher and
lower occupations; and a majority wanted the government to combat this. To a lesser extent
than elsewhere, people felt that job performance ought to be the central reward criterion.
Postmaterialist work orientations were widespread, and females had a stronger preference for
working for the government than in other countries. Although according to Figure 1 Norway was
a less clear representative of the social-democratic regime type, the value orientations of its
population seem very similar to those which prevailed in Sweden (which in both graphs emerges
as a rather pure case). Denmark, on the other hand, was a fairly clear exemplar of the regime
type, but seemed less social-democratic in its culture. The comparatively low Danish score on the
first axis can be explained partly by missing data on two variables. If scores similar to the other
Scandinavian countries are imputed, Denmark moves closer to Norway and Sweden, but remains
a less typical case, especially on the second dimension. In the 1990s, the Danish resembled their
fellow Nordic people in a comparatively strong rejection of traditional family roles, benefit fraud
and tax evasion. But on postmaterialism they scored considerably lower than Swedes and
Norwegians, and the female preference for working in the public sector was also less
pronounced. In two respects the Danish were rather corporatist: many people thought seniority
should be an important earnings criterion, and a fairly high proportion of men favoured working
in the government sector.
The third cluster comprises France and Germany, which are scaled quite close together. These
countries attain high negative scores on the second dimension of Figure 2. Typical for their corpo-
ratist culture at the time was the comparatively strong endorsement of traditional gender roles, the
value placed on seniority as a key wage criterion, and the rather lenient views on tax evasion and
social security fraud. The preferred earnings of a Cabinet minister were comparatively high and
males often favoured working for the government, both suggesting a considerable prestige of state
officials.

Figure 2. Optimal scaling of 10 countriesa on value orientations (nonlinear PCA, object scores).
a DE: Excluding the new Bundesländer after German reunification.
Sources: See Appendices 2, 4.
In cultural terms the Netherlands may once again be considered a hybrid. On the first dimen-
sion, the Dutch appear to be less collectivistic than inhabitants of the other countries in Western
Europe and Scandinavia. This is largely due to the fact that two variables were missing: if
‘corporatist’ or ‘social-democratic’ responses are imputed, the Netherlands attains a position
close to France. The second dimension, however, presents a very mixed picture. The Dutch
favoured postmaterialism in the 1990s, but to a lesser degree than the Swedes and Norwegians;
and the female preference for working in the public sector lagged far behind that in Scandinavia.
Furthermore, while the Dutch attained a ‘corporatist’ level on traditional gender roles and the
acceptability of tax evasion, they recorded fairly low scores on the preference of males for
working in the collective sector, on the importance of seniority as a wage criterion, and on the
acceptability of benefit fraud. Thus it is likely that the Netherlands in fact occupies a somewhat
higher position on the first dimension than Figure 2 suggests, and has a truly hybrid score on
the second.

Cluster analyses of country scores
The classification is largely confirmed by the outcomes of hierarchical cluster analyses of the
country scores.24 A two-cluster solution distinguishes between the two corporatist countries plus
the Netherlands on the one hand, and the representatives of the liberal and social-democratic
regime on the other. In a three-cluster model the first group consists of France, Germany and the
Netherlands, the second comprises Norway, Sweden and Denmark, and the third contains Australia,
Canada, Great Britain and the USA. A four-cluster solution splits the latter group in two (the North
American countries versus the rest).
Correlation between regimes and cultures
All in all, the correspondence of these orientations with regime types is found to be greater than in
the previous work. On the first dimension, the correlation of the country scores in Figures 1 and 2
is 0.84; on the second it is 0.85. The analyses therefore suggest that during the 1990s Esping-
Andersen’s regime typology largely coincided with three different cultures of social security, at
least in its key countries. In revealing this relationship, the incorporation of indicators singling out
corporatist and social-democratic orientations was essential.
Conclusions
Esping-Andersen’s ‘Three Worlds’ indicate general, qualitatively different and coherent systems of
formal institutions. Their empirical status is open to debate, as the many alternative classifications
of regime types published over the last two decades demonstrate. Previous analyses in this field
had several drawbacks. They tended to cover only a part of the regime traits (e.g. pensions or social
assistance), to be based on a small number of variables, and to use multivariate techniques that
were not entirely suitable. The first nonlinear principal components analysis performed here sought
to accommodate these issues. By applying this technique, an extensive dataset covering 54 regime
traits of eleven countries during the first part of the 1990s was reduced to two main dimensions.
Generally speaking, the results confirm Esping-Andersen’s typology. The component loadings of
the variables fall into a number of consistent groups. These allow for a clear interpretation of two
underlying dimensions, one referring to the scope of social security, the other to the degree of uni-
versalism. The grouping of countries on the principal components largely corresponds to what was
expected, and is corroborated by cluster analyses of the country scores. Scrutiny of a subset of vari-
ables indicates that the clustering remains fairly intact today, even if more countries are added.
Compared with classic PCA, the nonlinear technique results in a better fit and a clearer identifica-
tion of regime types. While the first question addressed in this article – ‘Do Esping-Andersen’s

‘Three Worlds’ exist empirically if one considers a comprehensive set of formal institutions?’ –
may therefore be answered in the affirmative, some qualifications need to be made. Seven coun-
tries may be regarded as fairly characteristic exponents of the regime types in the early 1990s:
Sweden and Denmark for the social-democratic regime type, which is extensive and universalistic;
Belgium, Germany and France for the corporatist model (extensive-particularistic), and the USA
and Australia for the liberal group (residual, and about average in terms of the degree of universal-
ism). Yet three countries belonged less clearly to these clusters. In the social-democratic group,
Norway was not as extensive as Sweden, and not as universalistic as Denmark. The United
Kingdom and Canada had social security regimes with a wider scope than the other liberal coun-
tries. The Netherlands emerged as a true hybrid, occupying a position between the social-demo-
cratic and corporatist countries. Moreover, the outcomes relating to the reduced dataset for 2008–
2010 indicate that the dispersion on the second dimension may have decreased over time: the
corporatist group seems less selective, the social-democratic cluster not as universalistic as it used
to be. Of the countries that were added in that analysis, Iceland, Ireland and New Zealand end up
in the group to which they belong theoretically; but in line with previous research, Austria, Finland
and Switzerland are less typical cases.
In the 1990s, the link between formal regime types and informal social security cultures turned
out to be fairly straightforward. A similar typology emerged from a nonlinear PCA relating to value
orientations. More than elsewhere, people in countries with a liberal social security regime favoured
individual freedom above collective solidarity. Fewer people wanted the government to intervene in
order to guarantee minimum income standards and to provide jobs. To a lesser extent than elsewhere
the government was held responsible for reducing income inequality, and there was less aversion to
income differentials at the top. This liberal culture was more pronounced in the USA and Canada
than in Australia and Great Britain. The social-democratic regime type mostly coincided with
social-democratic value orientations: people more often favoured a strong role for the government,
endorsed postmaterialist work values and preferred small earnings differentials between lower and
higher occupations. In line with the actual situation, women would choose to work in the public
sector to a greater extent than women elsewhere. The social-democratic culture emerged less clearly
in Denmark than in Sweden and Norway. In two countries operating a corporatist regime, value
orientations also conform to the expected pattern. The French and Germans more often favoured
traditional gender roles, attributed a comparatively elevated status to state officials, more often
thought seniority should be an essential earnings criterion, and had a comparatively lenient interpre-
tation of formal rules. Finally, the Janus-headed Dutch social security regime was supported by a set
of orientations that resembled those of the corporatist countries in some respects and those of the
social-democratic group in others; its culture was therefore hybrid as well. Thus, the answer to the
second research question – do structural regimes coincide with different ‘cultures of social security’

– is largely positive for the countries studied here: the correlations of the country scores on the two
principal components of Figures 1 and 2 equal 0.84 and 0.85, respectively.
Of course, this does not imply that these issues are now settled once and for all. The analyses
could be extended to other traits and countries – Mediterranean, Eastern Asiatic, new EU member
states – and include more years, in order to assess the path of change in different regime types and
cultures. It also may be that part of the correlation found here is due to the fact that countries oper-
ating the same regime are similar in other respects as well, for example in their demography, reli-
gious background or political organization. Even more important, the present analysis looks only
at if regime types and social security cultures exist and correspond to any extent, not why they do
so. Is it simply that regimes produce their own cultural legitimacy, or is the relationship far more
complex and dynamic, as postulated in new institutionalism? A more elaborate historical analysis
could shed light on these matters. Preferably, the evolution of the various aspects of regimes and
cultures of social security identified here would then be linked to socioeconomic, demographic and
technological developments; to the negotiated responses of policy actors facing these challenges;
and to social outcomes in terms of de-commodification, stratification and social rights.
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit
sectors.

Appendix 1. Component loadings and variables in the regime analysis
#
Component
loadings
Type of
scheme
Variable Unit of
measurement
Input categories (before scaling)‡
dim 1 dim 2 1 2 3 4 5
1
−0.94
0.30
LP
Level of maternity benefit
% Earnings
100
90
79–84
53–70
<50
2
−0.93
0.34
LP
Tax exemptions for work-related expenses
(single earner, APW*)
% Average
income
8–9 4–5 2–3
0
3
−0.84
−0.39
LP
Duration of paid maternity and parental
leave
Weeks, level
>25 w.,
% earnings
13–25 w.,
% earnings
<13 w.,
% earnings
Flat rate
None
4
−0.83
0.28
LP
Coverage of maternity leave % of all
employed
women
>90
60
<50
5
−0.49
−0.08
LP
Statutory minimum wage
Level
High
Low
None
6
−0.93
−0.10
FU
Direct/indirect taxes + contributions
% GNP
49–53
44–47
36–38
29–30
7
−0.86
−0.26
FU
Average tax/contribution rate (single earner
+ children, APW*)
%
39–41
26–28
23–24
18–19
8
−0.77
−0.16
FU
Marginal tax/contribution rate (single
person, 1.33 APW*)
%
62–69
51–53
46–49
41–43
<35
9
−0.66
0.20
FU
Employer’s contributions: average surcharge
(single earner + children, APW*)
%
33–34
15–18
8–11
0
10
−0.32
−0.02
FU
Marginal tax/contribution rate (double
earner + 2 children, 1.33 APW*)
%
>61
49–52
39–42
<36
11
−0.93
0.33
UA
Duration of earnings-related unemployment
benefit (not means-tested)
Years
>3.5 yrs
variable
< 1 yr
12
−0.85
−0.33
UA
Social assistance single person, 5 yrs
unemployed (NRR*)
%
62–65
46–52
27–39
<10
13
−0.80
−0.12
UA
Social assistance couple, 5 yrs unemployed
(NRR*)
%
74–83
57–68
46–50
<15
14
−0.77
−0.02
UA
Unemployment benefit 1st month, couple +
children (2/3 APW, NRR*)
%
>83
75–80 60–69
15
−0.76
−0.14
UA
Unemployment benefit 1st month, couple +
children (APW, NRR*)
%
>78
70–75
60–69
16
−0.75
0.00
UA
Means-tested social assistance in social
security expenditure
%
3–5
7–9
>15
17
−0.79
−0.27
DA
Disability coverage, risque professionnel*
Target group
All working
All
employees
Some
employees
High negative loading on dimension 1

Appendix 1.
(Continued)
#
Component
Type of Variable Unit of Input categories (before scaling)‡
loadings
scheme
measurement
dim 1
dim 2
1
2
3
4
5
18
−0.76
0.09
CH
Average child benefit level across 18 family
types
Monthly
amount in
Euros
>430
± 270
180–230 135–180
<135
19
−0.72
0.16
CH
Universality of child benefits
Type of benefit
Universal
+ income
related
Universal
Income
related
20
−0.69
0.10
CH
Child benefit level, non-poor lone-parent
families
Relative to
national average
(= 100)
130–150
100
<80
21
−0.64
0.25
CH
Total level of provisions for children
Ranking overall
child package
High
Medium
Low
22
−0.72
0.33
OA
Means testing of collective retirement
pensions
Extent
None
Partial
(Almost)
full
23
−0.46
0.29
OA
Equivalence of insurance period and benefit
level
Occurrence
Yes
No
24
−0.42
0.21
OA
Level of earnings-related occupational
pensions (GRR*)
%
65–70
60–61
<54
25
−0.69
0.07
SV
Means testing widow’s pensions
Extent
None
Partial
(Almost)
full
26
−0.20
0.17
SV
Separate orphan’s pension
Occurrence
Yes
No
High
negative loading on dimension 2 (dimension 1: usually positive)
27
0.12
−0.94
DA
Disability coverage, risque social* Target group
Inhabitants:
flat rate
Inhabitants Employees
+
All Some
employees
employees
Employees:
% earnings
Self-
employed
28
0.45
−0.53
DA
Disability benefit level, risque professionnel* Level
Part of
wages
Full wage
29
0.22
−0.93
LP
Coverage of collective labour agreements
% Employees
<47
60–69
71–86
>90
30
0.04
−0.82
LP
Additional tax deductions for couples with
children
Monthly
amount in
Euros
0
7–20 48–62
>74

Appendix 1.
(Continued)
#
Component
loadings
Type of
scheme
Variable Unit of
measurement
Input categories (before scaling)‡
dim 1
dim 2
1
2
3
4
5
31
0.03
−0.72
LP
Children (0–3 years) in formal child care
%
>48
34–45
15–29
<10
32
0.46
−0.70
LP
Total duration of maternity and parental
leave
Years
>1.5 yrs
0.5–1.5 yrs <0.5 yrs
33
0.53
−0.59
LP
Additional tax advantages for one-earner
couples (not related to children)
% Gross
income
>9
3–5
0
34
0.27
−0.87
OC
Level of protection for civil servants US$ per capita
pop. 15–64 yrs
<115
215–270
± 415
>560
35
0.12
−0.81
OC
Separate schemes for occupational groups
Occurrence
None
Some
Many
36
0.36
−0.84
CH
Child benefit level, non-poor couples + 4
children
Relative to
national average
(= 100)
<100
125–170
>200
37
0.44
−0.80
CH
Child benefit level, poor lone-parent families
Relative to
national average
(= 100)
± 240
150–175
100–135
38
0.03
−0.83
FU
Contribution-funded social security
% of total
funding
<28
40–42
58–64
>71
39
−0.35
−0.76
SV
Collective pension for widows of employees
Benefit type
None
Temporary
Long-term
High positive loading on dimension 2 (dimension 1: positive)
40
0.19
0.86
OA
Collective retirement pension for all
inhabitants
Occurrence
No
Yes
41
0.53
0.68
OA
Minimum collective retirement pension for
non-employees (average single, couple)
Monthly
amount in US$
<725
± 850
900–1100
42
0.30
0.58
OA
(Semi-) collective schemes for full early
retirement
Occurrence
Yes, below
age 60
Yes, 60–65
No
43
44
0.60
0.54
0.51
0.77
OA
UA
Coverage of earnings-related occupational
pensions
Social assistance couple + 2 children, 5 yrs
%
%
5–25
49–65
45–55
71–83
60–80
95–100
100
45
0.41
0.71
UA
unemployed (NRR*)
Social assistance lone-parent family + 2
%
42–50
57–65
69–72
90
46
0.38
0.63
LP
children, 5 yrs unemployed (NRR*)
Statutory retirement age (men)
Age in years
<65
65
>65
47
0.50
0.57
CH
Child benefit level, poor couples + 4
children
Relative to
national average
(= 100)
>200
± 140
80–100

Appendix 1.
(Continued)
#
Component
Type of
Variable Unit of
Input categories (before scaling)‡
loadings
scheme
measurement
dim 1
dim 2
1
2
3
4
5
48
49
0.54
0.61
0.46
0.40
FU
DA
Marginal tax/contribution rate (single earner
+ 2 children, 1.33 APW*)
Disability threshold, risque professionnel*
%
Occurrence
<36
No
47–51
Yes
>62
Other
50
−0.23
0.58
DA
Disability benefit level, risque social*
Benefit type
% Earnings
Flat rate
51
−
−
DA
Disability threshold, risque social*†
% Work
incapacity
<50
50–64
65–79
80–100
52
−
−
LP
Weeks of leave (statutory minimum)†
Weeks
None
<4 weeks
4 weeks
>4 weeks
53
−
−
LP
Expenditure on active labour market
policy†
% GNP
<0.3
0.5–0.6
1.0–1.1
>1.3
54
−
−
SV
Collective provision for widows of
residents†
Benefit type
None
Flat rate,
low
Flat rate,
high
Plot of component loadings:
Legend:
CH = Costs of children; DA = Disability schemes; FU = Funding; LP = Labour market and participation (dis)incentives; OA = Old age retirement
pensions; OC = Occupational schemes; SV = Surviving dependants’ pensions; UA = Unemployment and social assista nce.
* APW = average production worker; NRR = net replacement rate; GRR = gross replacement rate; Risque professionnel = disability related to
occupational activity; Risque social = disa bility not related to occupational activity.
† Multiple nomina l scaling level (all other variables: single ordinal).
‡ In order to enhance the i nterpretability of co mponent loadings, some variables run from high to low; this do es not affect sc aling results.
Sources: see Appendix 4.

Appendix 2. Component loadings and variables in the analysis of informal cultures
# Component loadings Type of
orientation
Variable/questionnaire item Unit of measurement Input categories (before scaling)‡
dim 1
dim 2
1
2
3
4
5
High negative loading on dimension 1
1
2
−0.88
−0.83
−0.40
0.15
GOV
GOV
Should it be the government’s
responsibility to provide a decent
standard of living for the unemployed?
Should it be the government’s
% Definitely/probably should be
% Definitely/probably should be
91–93
>96
78–82
94
65–68
87–91
48
3
−0.83
0.27
GOV
responsibility to provide a decent
standard of living for the old?
Should it be the government’s
% Definitely/probably should be
86–92
70–73
58
52–55
38
4
−0.71
−0.15
GOV
responsibility to provide a job for
everyone who wants one?
It is the responsibility of the government
% (strongly) agree
60–69
47–57
35
5
−0.88
0.17
INA
to reduce the differences in income
between people with high incomes and
those with low incomes
Dispersion measure based upon:
Gini-coefficient of respondents’
0.19
0.28–0.29
0.34–0.36
0.37–0.38
‘About how much should [six higher
occupations*] earn each year before
taxes?’
preferred earnings of six higher
occupations*
High positive loading on dimension 2 (dimension 1:
positive)
6
7
0.69
0.65
0.60
0.60
PMW
PMW
I would enjoy having a paid job even if I
did not need the money
‘Good opportunities for advancement’
(how important respondent thinks this
is in a job?)
% (Strongly) agree
% (Very) important
47–55
85–89
60–70
77–81
73–75
51–53
8
9
0.65
0.62
0.60
0.58
FEM
MER
I would choose: working in a private
business – working for the government
or civil service
In deciding how much people ought to
% Government/civil
service [ratio female : male
respondents]
% essential
1.06–1.12
39–46
1.18–1.21
25–33
1.28–1.32 1.91–2.04
21–23
10
0.61
0.47
INA
earn, how important should it be how
well he or she does the job?
Dispersion measure based upon: ‘About
Theil-coefficient of respondents’
0.13
0.11–0.12
0.03–0.06
how much should [lower vs higher
occupations] earn each year before
taxes?’
preferred earnings, 3 lower vs 6
higher occupations†
High negative loading on dimension 2 (dimension 1: positive)
11 0.17 −0.94 TGR All in all, family life suffers when the
woman has a full-time job
% (Strongly) agree 22–25 31–32 38–42 46–54

Appendix 2.
(Continued)
# Component loadings Type of
orientation
Variable/questionnaire item Unit of measurement Input categories (before scaling)‡
dim 1
dim 2
1
2
3
4
5
12
0.20
−0.88
LEN
A person gives the government
% Not/a bit wrong
2–5
14
21
incorrect information about himself
to get government benefits he is not
entitled to
13
0.32
−0.87
LEN
A taxpayer does not report all of his
% Not/a bit wrong
16–20
24–27
39–42
48
income in order to pay less income
taxes
14
0.34
−0.69
PST
I would choose: working in a private
% Government/civil service
16–20
23–28
39
75–79
business – working for the government
[male respondents]
or civil service (males)
15
0.73
−0.50
PST
How much should [a cabinet minister in
Scale of mean preferred earnings
4.0
6.0
7.0
9.0
the national government] earn?
of nine occupations† in country
[cabinet minister ranking score]
16
0.16
−0.31
SEN
In deciding on pay for two people doing
% Essential/very important
27–31
40–46
74
the same kind of work, how important
should (it) be how long the person has
been with the firm?
Plot of component loadings:
Legend:
FEM = Female preference for public sector employment; GOV = Governmental responsibility for a decent standard of living, income redistribution,
and providing jobs; INA = Income ineq uality aversion; LEN = Le nient interpretation o f formal rules; PMW = Postmaterialist work orientation; MER
= Meritocratic performance as a wage criterion; PST = Prestige of state officials; S EN = Seniority as a wage criterion; TGR = Traditi onal gender
role orientation.
* Preferred earnings for: lawyer; do ctor in general practice; Cabinet minister in the national government; Supreme Court Judge; owner-manager of a
large factory; chairman of a large national corporation.
† Preferred earnings for: unskilled factory worker; shop assistant; skilled factory worker; plus the six higher occupations above.
‡ In order to enhance the i nterpretability of co mponent loadings, some va riables run from high to low; this does not affect scaling results.
Sources: see Appendix 4.

Appendix 3. NPCA object scores of 17 countries* (10 formal regime traits, 2008–2010)
Dim I
Dim II
Liberal group
United States
−1.52
0.62
Australia
−1.22
0.33
Canada
−1.21
0.52
New Zealand
−1.18
1.05
United Kingdom
−0.77
0.78
Ireland
−0.52
0.45
Social-democratic group
Norway
0.76
0.71
Iceland
0.42
0.68
Denmark
0.26
0.56
Sweden
0.87
0.22
Finland
0.67
−0.03
Corporatist/hybrid group
France
0.65
−0.75
The Netherlands
0.53
−0.71
Belgium
0.67
−0.49
Germany
−0.29
−0.12
Austria
0.28
0.09
Switzerland
0.12
0.57
* Fitted as supplementary objects to the 1990s NPCA solution.
Underlined: New countries in the analysis.
Italics: Scores in unexpected direction.
Sources: See Appendix 4.
Appendix 4.
Data sources
Var #
Regime analysis 1990s
(see Appendix 1)
Regime analysis
2008–
2010
(see Appendices 1, 3)
Informal cultures
analysis 1990s
(see Appendix 2)
1
Luxembourg Income Study
ILO (2011)
ISSP 1996
2
OECD (1995a)
----
ISSP 1996
3
OECD (1995b)
----
ISSP 1998
4
Luxembourg Income Study
----
ISSP 1999
5
OECD (1994)
----
ISSP 1999
6
OECD (1995a)
OECD (2011)
ISSP 1997
7
OECD (1995a)
----
ISSP 1997
8
OECD (1994)
----
ISSP 1997
9
OECD (1995a)
----
ISSP 1999
10
OECD (1994)
----
ISSP 1999
11
OECD (1997a)
OECD (2007)
ISSP 1998
12
OECD (1997a)
----
ISSP 1998
13
OECD (1997a)
----
ISSP 1998
14
OECD (1997a)
OECD (2011)
ISSP 1997

Appendix 4.
(Continued)
Var #
Regime analysis 1990s
(see Appendix 1)
Regime analysis
2008–
2010
(see Appendices 1, 3)
Informal cultures
analysis 1990s
(see Appendix 2)
15
OECD (1997a)
----
ISSP 1999
16
Eardley et al. (1996)
----
ISSP 1997
17
DHHS (1991); EC (1990)
----
----
18
Bradshaw et al. (1993); DHHS (1991)
----
----
19
Bradshaw et al. (1993); DHHS (1991)
----
----
20
Bradshaw et al. (1993)
----
----
21
Bradshaw et al. (1993)
----
----
22
OECD (1988)
----
----
23
DHHS (1991)
----
----
24
OECD (1992)
----
----
25
DHHS (1991)
ILO (2011)
----
26
DHHS (1991)
----
----
27
DHHS (1991); EC (1990)
----
----
28
DHHS (1991); EC (1990)
----
----
29
OECD (1997b)
ICTWSS (2011)
----
30
Bradshaw et al. (1993); OECD (1995a)
----
----
31
OECD (2001)
----
----
32
OECD (1995b)
----
----
33
OECD (1995a)
----
----
34
ILO (1992)
----
----
35
DHHS (1991)
----
----
36
Bradshaw et al. (1993)
----
----
37
Bradshaw et al. (1993)
----
----
38
ILO (1992)
OECD (2011)
----
39
DHHS (1991)
----
----
40
DHHS (1991)
----
----
41
DHHS (1991); CPB (1995)
----
----
42
De Kemp (1992); OECD (1988)
OECD (2009)
----
43
OECD (1992)
----
----
44
OECD (1997a)
----
----
45
OECD (1997a)
OECD (2011)
----
46
De Kemp (1992); OECD (1988)
----
----
47
Bradshaw et al. (1993)
----
----
48
OECD (1994)
----
----
49
DHHS (1991); EC (1990)
ILO (2011)
----
50
DHHS (1991); EC (1990)
----
----
51
DHHS (1991); EC (1990)
----
----
52
OECD (1994)
----
----
53
OECD (1993)
----
----
54
DHHS (1991)
----
----
Note: Sources of variables included in sensitivity analysis of reduced dataset are in italics.

Notes
1.
Esping-Andersen uses the expression ‘liberal’ in its classic sense, referring to a regime that maximizes indi-
vidual freedom and limits state intervention. This convention is followed throughout this article, although the
word has acquired a different meaning in American English. As Murray (1997: xii) points out: ‘The correct
word for my view of the world is ‘‘liberal’’ [and] the writers of the nineteenth century who expounded on
this view were called liberals […]. But words mean what people think they mean, and in the United States
the unmodified term ‘‘liberal’’ now refers to the politics of an expensive government and the welfare state.’
2.
The ‘Mediterranean’ group would show weak collective safety nets and a strong emphasis on support
by the family and patron–client relationships (see Ferrera, 1996, 2010). The ‘Eastern Asiatic’ variant
theoretically combines low public spending with social security provided by families, local communities
and firms (Peng and Wong, 2010). Australia and New Zealand’s ‘radical’ system is assumed to provide a
more extensive collective protection than the liberal type (Castles, 1996, 2010). Esping-Andersen (1990,
1999) regards the Mediterranean and Eastern Asiatic systems as variants of the corporatist model, and
the Antipodeans as part of the liberal model.
3.
In this respect, power resources theory diverges from the ‘state-centered’ tradition in comparative
social policy (Amenta, 2003; Skocpol, 1985), which regards welfare programmes as the outcome of the
bureau- cratic and financial capacities of governments, the centralization or fragmentation of
policymaking, and political party systems (e.g. Amenta, 1998; Orloff and Skocpol, 1984; Skocpol,
1992; Tsebelis, 2002).
4.
For example, Arts and Gelissen (2001) classify corporatist Belgium as a representative of the social-
democratic regime, while Jacobsen (2011) places both ‘Mediterranean’ Italy and Spain and the hybrid
Dutch case in the corporatist group, together with Germany and France.
5.
Shalev does not consider Qualitative Comparative Analysis (Kangas, 1994; Ragin, 2000) a suitable alter-
native, as it has thus far produced rather inconclusive regime classifications, due to its dependence on
binary logic. However, fuzzy set approaches might open up new avenues for QCA (Kogut, 2010).
6.
Technically, this is because component loadings and scores for individual objects (cases) are not obtained
from an eigenvalue decomposition of the correlation matrix, or from a singular value decomposition of
the standardized data matrix. Nonlinear PCA minimizes a least-squares loss function, where the loss per-
tains to the decrease of information due to representing the original data on a small number of principal
components (Linting et al., 2007).
7.
In terms of de-commodification and stratification the original typology related to 18 countries, but the data
on the employment structure included fewer countries (see tables 6.1-8.12 in Esping-Andersen, 1990).
8.
Esping-Andersen (1990: 70, 127) only measured the share of social assistance benefits in total public
expenditure (the ‘means-tested poor relief bias’), not the characteristics of these benefits. Disability out-
lays (excluding occupational injuries) were classified as pension expenditure (Esping-Andersen, 1990:
104); but the actual traits of occupational and social incapacity schemes (coverage, levels, disability
thresholds) were not assessed.
9.
For instance, Esping-Andersen’s (1990: 177) conclusion that the tax burden was much heavier in social
democratic nations than in corporatist countries may have been due partly to his selection of the single-
earner family with two children as a reference case. Theoretically, in corporatist countries breadwinner
families are more likely to receive a favourable tax treatment.
10.
This is somewhat later than in the regime analysis, as suitable earlier data were not available for the
countries studied here. No data were gathered for Northern Ireland; the analysis therefore concerns
Great Britain and not the UK. The 1999 waves for Denmark and the Netherlands are not included in the
standard ISSP dataset.
11.
Denmark and the Netherlands did not take part in ISSP 1996 and as a result have two missing variables;
Australia skipped the 1997 round and lacks five variables. Missing observations were treated passively
in the analysis, implying that they do not influence the scaling results.
12.
Obviously, the constructed variable does not represent the entire distribution of preferred incomes. The

questionnaire items relate to occupational earnings at the top and at the bottom, thus neglecting middle
incomes and benefit recipients. Furthermore, countries usually have far more low-skilled employees than
lawyers or Cabinet ministers, while in the constructed variable the latter type of profession dominates
(since it contains responses for six higher and only three lower-level occupations).
13.
For 22 items the correlation of the transformed and original codings is almost perfect (r > 0.985); 11
others show limited deviations from linearity (0.90 < r ≤ 0985). Nonlinearity is somewhat stronger in the
transformation of 17 other items, mainly because some categories were merged; yet here the ordinality
restriction still assures a high correlation with the original codings (0.77 < r ≤ 0.90). Two of the four
variables that were treated as nominal are linearly transformed on the first dimension (0.97 < r < 0.99),
but not on the second. The other two nominally scaled items are nonlinearly transformed on both. An
additional analysis showed that omitting the four nominal variables leads to minor changes in the com-
ponent loadings of the remaining items (≤ .04) and produces very similar country scores.
14.
The third dimension adds 11% to the explained variance, the fourth 7%, and the fifth 6%. These higher
dimensions identify atypical differences between a few countries on specific items. For instance, the third
dimension shows a contrast between Sweden and Norway on the one hand, and the Netherlands on the
other. This mainly results from the absence of a full early retirement scheme and a statutory minimum
wage in Sweden and Norway, and the lack of a collective employee insurance for surviving dependants in
the Netherlands.
15.
For the sake of comparability all solutions were rotated slightly (by 0–9 degrees), with the restriction that
the average component loading on the y-axis should equal zero for the ‘liberal’ items.
16.
Data on the duration of earnings-related unemployment benefit refer to 2005.
17.
Mediterranean and Eastern Asiatic countries were not included in the present analysis, because this
would merely have assessed their degree of liberalism, corporatism and social democracy; a fair analysis
would also have to include variables that are characteristic of these possibly distinct regimes (see note 2).
18.
However, on the first dimension the German score decreased as a result of the shorter duration of non-
means tested unemployment benefit and the increased means-testing of widow’s pensions. Denmark
also ended up lower than before, because the level of maternity benefits declined. It should be noted that
changes on a few items are likely to have a greater impact in this analysis (based on 10 indicators) than
in the original set of 54 variables.
19.
If the more recent observations are treated as active cases in a nonlinear PCA of the same 10 traits in the
original 11 countries the first dimension remains intact. However, as might be expected, the eigenvalue
of the second dimension drops from 0.67 to 0.57. It can still be interpreted as a contrast between univer-
sal and selective traits, but is based on only three of the original five high-loading items.
20.
In the nonlinear PCA Switzerland’s positive score on the second dimension is due to an un-corporatist
low coverage of collective labour market agreements and a limited share of contribution-funded social
security. The hybrid scores of the other two countries mostly result from their atypical position regarding
collective schemes for full early retirement (nonexistent in Austria, available in Finland).
21.
For three-quarters of the variables (12 out of 16) deviations from linearity turned out to be limited: the
correlation of the transformed scores with the original codings exceeds 0.90. The remaining four were
transformed partly in a nonlinear way, mainly due to category merging; but here too the correlation is
quite strong (0.77 < r < 0.90).
22.
Adding a third dimension increases the explained variance to 88%, but the interpretation of this solution
is not substantially different. The new second dimension only measures the adherence to corporatist
orientations, while the third dimension captures the support for social-democratic orientations. Thus, the
original second dimension has merely been split.
23.
Australia’s score on the first dimension cannot be attributed to missing values, because most high-load-
ing variables are valid. If one imputes the USA’s values for missing items, Australia remains about as
collectivistic as Great Britain.
24.
Because no data were available for Belgium, the number of cases was too small to assess the goodness
of fit through latent class cluster analysis (see the cluster analyses of regimes).

References
Abrahamson P (1999) The welfare modelling business. Social Policy & Administration 33(4): 394–415.
Amenta E (1998) Bold Relief: Institutional Politics and the Origins of Modern American Social Policy.
Princeton, NJ: Princeton University Press.
Amenta E (2003) What we know about the development of social policy: Comparative and historical research
in comparative and historical perspective. In: Mahoney J and Rueschemeyer D (eds) Comparative
Historical Analysis in the Social Sciences. Cambridge: Cambridge University Press, 91–130.
Andress H-J and Heien T (2001) Four worlds of welfare state attitudes? A comparison of Germany, Norway,
and the United States. European Sociological Review 17(4): 337–356.
Arcanjo M (2011) Welfare state regimes and reforms: A classification of ten European countries between
1990 and 2006. Social Policy and Society 10(2): 139–150.
Arseneault L, Tremblay RE, Boulerice B and Saucier J (2002) Obstetrical complications and violent delin-
quency: Testing two developmental pathways. Child Development 73(2): 496–508.
Arts W and Gelissen J (2001) Welfare states, solidarity and justice principles: Does the type really matter?
Acta Sociologica 44(4): 283–299.
Arts WA and Gelissen J (2010) Models of the welfare state. In: Castles FG, Leibfried S, Lewis J, Obinger
H and Pierson C (eds) The Oxford Handbook of the Welfare State. Oxford: Oxford University Press,
569–583.
Bambra C (2006) Decommodification and the worlds of welfare revisited. Journal of European Social Policy
16(1): 73–80.
Blekesaune M and Quadagno J (2003) Public attitudes toward welfare state policies: A comparative analysis
of 24 nations. European Sociological Review 19(5): 415–427.
Blomberg H and Kroll C (1999) Do structural contexts matter? Macro-sociological factors and popular
attitudes towards public welfare services. Acta Sociologica 42(4): 319–335.
Bonoli G (1997) Classifying welfare states: A two-dimension approach. Journal of Social Policy 26(3): 351–372.
Bonoli G (2000) Public attitudes to social protection and political economy traditions in Western Europe.
European Societies 2(4): 431–453.
Bonoli G (2006) New social risks and the politics of post-industrial social policies. In: Armingeon K and
Bonoli G (eds) The Politics of Post-Industrial Welfare States – Adapting Post-War Social Policies to New
Social Risks. Abingdon: Routledge, 3–26.
Bradshaw J, Ditch J, Holmes H and Whiteford P (1993) Support for Children: A Comparison of Arrangements
in Fifteen Countries. London: HMSO.
Brinton M and Nee M (1998) The New Institutionalism in Sociology. New York: Russell Sage Foundation.
Brooks C and Manza J (2007) Why Welfare States Persist: The Importance of Public Opinion in Democracies.
Chicago, IL: University of Chicago Press.
Castles F (ed.) (1993) Families of Nations – Patterns of Public Policy in Western Democracies. Aldershot:
Dartmouth.
Castles FG (1996) Needs-based strategies of social protection in Australia and New Zealand. In: Esping-
Andersen G (ed.) Welfare States in Transition: National Adaptations in Global Economics. London:
SAGE.
Castles FG (2010) The English-speaking countries. In: Castles FG, Leibfried S, Lewis J, Obinger H and
Pierson C (eds) The Oxford Handbook of the Welfare State. Oxford: Oxford University Press, 630–642.
Castles FG and Obinger H (2008) Worlds, families, regimes: Country clusters in European and OECD area
public policy. West European Politics 31(1–2): 321–344.
Castles FG, Leibfried S, Lewis J, Obinger H and Pierson C (eds) (2010) The Oxford Handbook of the Welfare
State. Oxford: Oxford University Press.
Coleman JS (1990) Foundations of Social Theory. Cambridge, MA: The Belknap Press of Harvard University
Press.
Coughlin R (1980) Ideology, Public Opinion, Welfare Policy: Attitudes towards Taxing and Spending in
Industrialized Countries. Berkeley, CA: Institution of International Studies.

CPB (1995) Replacement Rates: A Transatlantic View. The Hague: Central Planning Office.
De Kemp AAM (1992) Ouderen tussen pensioen en bijstand. Rijswijk and Den Haag: Sociaal en Cultureel
Planbureau/ VUGA.
DHHS (1991) Social Security Programs Throughout the World. Washington, DC: US Department of Health
and Human Services.
Eardley T, Bradshaw J, Ditch J, Gough I and Whiteford P (1996) Social Assistance in OECD Countries (2
vols). London: HMSO.
EC (1990) Comparative Tables of the Social Security Schemes in the Member States of the European
Communities. Brussels and Luxembourg: Commission of the European Communities.
Esping-Andersen G (1990) The Three Worlds of Welfare Capitalism. Cambridge: Polity Press.
Esping-Andersen G (1999) Social Foundations of Postindustrial Economics. Oxford: Oxford University
Press.
Esping-Andersen G (2009) The Incomplete Revolution – Adapting to Women’s New Roles. Cambridge:
Polity Press.
Eurelings-Bontekoe EHM, Duijsens IJ and Verschuur MJ (1996) Prevalence of DSM-III-R and ICD-10 per-
sonality disorders among military conscripts suffering from homesickness. Personality and Individual
Differences 21(3): 431–440.
Ferragina E and Seeleib-Kaiser M (2011) Welfare regime debate: Past, present, futures? Policy & Politics
39(4): 583–611.
Ferrera M (1996) The ‘Southern model’ of welfare in Social Europe. Journal of European Social Policy 6(1):
17–27.
Ferrara M (2010) The South European countries. In: Castles FG, Leibfried S, Lewis J, Obinger H and Pierson C
(eds) The Oxford Handbook of the Welfare State. Oxford: Oxford University Press, 616–629.
Gifi A (1990) Nonlinear Multivariate Analysis. Chichester: Wiley.
Goldthorpe JH (2000) On Sociology: Numbers, Narratives and the Integration of Research and Theory.
Oxford: Oxford University Press.
Gough I (2001) Social assistance regimes: A cluster analysis. Journal of European Social Policy 11(2): 165–170.
Hacker JS (2004) Privatizing risk without privatizing the welfare state: The hidden politics of social policy
retrenchment in the United States. American Political Science Review 98(2): 243–260.
Hall PA and Soskice D (eds) (2001) Varieties of Capitalism. The Institutional Foundations of Comparative
Advantage. Oxford: Oxford University Press.
Hall PA and Taylor RCR (1996) Political science and the three new institutionalisms. Political Studies
44(5): 936–957.
Hicks A and Kenworthy L (2003) Varieties of welfare capitalism. Socio-Economic Review 1(1): 27–61.
ICTWSS (2011) Institutional Characteristics of Trade Unions, Wage Setting, State Intervention and Social
Pacts (database, version 3.0). Amsterdam: University of Amsterdam.
ILO (1992) The Cost of Social Security. Geneva: International Labour Office.
ILO (2011) Social Security Database, downloaded June 2011. Geneva: International Labour Office.
Jacobsen TG (2011) Welfare attitudes and social expenditure: Do regimes shape public opinion? Social
Indicators Research 101(3): 323–340.
Jæger MM (2006) Welfare regimes and attitudes towards redistribution: The regime hypothesis revisited.
European Sociological Review 22(2): 157–170.
Jæger MM (2009) United but divided: Welfare regimes and the level and variance in public support for
redistribution. European Sociological Review 25(6): 723–737.
Jehoel-Gijsbers G and Vrooman C (2008) Social Exclusion of The Elderly. A Comparative Study of EU
Member States. Brussels: Centre for European Policy Studies.
Kangas OE (1994) The politics of social security: On regressions, qualitative comparisons, and cluster
analysis. In: Janoski T and Hicks AM (eds) The Comparative Political Economy of the Welfare State.
Cambridge: Cambridge University Press.
Kautto M (2010) The nordic countries. In: Castles FG, Leibfried S, Lewis J, Obinger H and Pierson C (eds)
The Oxford Handbook of the Welfare State. Oxford: Oxford University Press, 586–600.

Kitschelt H, Lange P, Marks G and Stephens JD (eds) (1999) Continuity and Change in Contemporary
Capitalism. Cambridge: Cambridge University Press.
Kluegel JR and Miyano M (1995) Justice beliefs and support for the welfare state in advanced capitalism.
In: Kluegel JR, Mason DS and Wegener B (eds) Social Justice and Political Change: Public Opinion in
Capitalist and Post-Communist States. New York: Aldine de Gruyter, 81–105.
Kogut B (2010) Qualitative comparative analysis of social science data. In: Morgan G, Campbell JL, Crouch
C, Pedersen OK and Whitley R (eds) The Oxford Handbook of Comparative Institutional Analysis.
Oxford: Oxford University Press, 139–179.
Korpi W (1983) The Democratic Class Struggle. London: Routledge and Kegan Paul.
Korpi W (1989) Power, politics, and state autonomy in the development of social citizenship: Social rights
during sickness in eighteen OECD countries since 1930. American Sociological Review 54(3): 309–328.
Larsen CA (2006) The Institutional Logic of Welfare Attitudes; How Welfare Regimes Influence Public
Support. Aldershot: Ashgate.
Leibfried S and Mau S (eds) (2008) Welfare States: Construction, Deconstruction, Reconstruction. Analytical
Approaches. Cheltenham: Edward Elgar.
Levy JD (2010) Welfare retrenchment. In: Castles FG, Leibfried S, Lewis J, Obinger H and Pierson C (eds)
The Oxford Handbook of the Welfare State. Oxford: Oxford University Press, 552–565.
Linting M, Meulman JJ, Groenen PJF and Van der Kooij JJ (2007) Nonlinear principal components analysis:
Introduction and application. Psychological Methods 12(3): 336–358.
Lipset SM (1963) The First New Nation: The United States in Historical Perspective. New York: Basic
Books.
Mehrtens FJ (2004) Three worlds of public opinion? Values, variation, and the effect on social policy.
International Journal of Public Opinion Research 16(2): 115–143.
Meulman JJ, Van der Kooij AJ and Heiser WJ (2004) Principal components analysis with nonlinear optimal
scaling transformations for ordinal and nominal data. In: Kaplan D (ed.) The Sage Handbook of Quantitative
Methodology for the Social Sciences. London: SAGE, 49–70.
Michailidis G and De Leeuw J (1998) The Gifi system of descriptive multivariate analysis. Statistical Science
13(4): 307–336.
Morgan G, Campbell J, Crouch C, Pedersen OK and Whitley R (2010) The Oxford Handbook of Comparative
Institutional Analysis. Oxford: Oxford University Press.
Murray CA (1997) What It Means To Be a Libertarian; A Personal Interpretation. New York: Broadway
Books.
North DC (1990) Institutions, Institutional Change and Economic Performance. Cambridge: Cambridge
University Press.
Obinger H and Wagschal U (1998) Das Stratifizierungskonzept in der clusteranalytischen Überprüfung. In:
Lessenich S and Ostner I (eds) Welten des Wohlfarhtskapitalismus – der Sozialstaat in vergleichender
Perspektive. Frankfurt am Main: Campus Verlag.
OECD (1988) Reforming Public Pensions. Paris: Organisation for Economic Co-operation and Development.
OECD (1992) Private Pensions and Public Policy. Paris: Organisation for Economic Co-operation and
Development.
OECD (1993) Employment Outlook. Paris: Organisation for Economic Co-operation and Development.
OECD (1994) The OECD Jobs Study – Evidence and Explanations; Part II: The Adjustment Potential of the
Labour Market. Paris: Organisation for Economic Co-operation and Development.
OECD (1995a) The Tax/Benefit Position of Production Workers. Paris: Organisation for Economic Co-
operation and Development.
OECD (1995b) Employment Outlook. Paris: Organisation for Economic Co-operation and Development.
OECD (1997a) Making Work Pay: Taxation, Benefits, Employment and Unemployment. Paris: Organisation
for Economic Co-operation and Development.
OECD (1997b) Employment Outlook. Paris: Organisation for Economic Co-operation and Development.
OECD (2001) Employment Outlook. Paris: Organisation for Economic Co-operation and Development.
OECD (2007) Benefits and Wages 2007. Paris: Organisation for Economic Co-Operation and

Development.
OECD (2009) Pensions at a Glance. Paris: Organisation for Economic Co-Operation and Development.
OECD (2011) OECD iLibrary, downloaded June 2011. Paris: Organisation for Economic Co-Operation and
Development.
Orloff AS (1996) Gender in the welfare state. Annual Review of Sociology 22: 51–78.
Orloff AS and Skocpol T (1984) Why not equal protection? Explaining the politics of public social spending
in Britain, 1900–1911, and the United States, 1880s–1920s. American Sociological Review 49(6): 726–750.
Palier B (ed.) (2010) A Long Goodbye to Bismarck? The Politics of Welfare Reform in Continental Europe.
Amsterdam: Amsterdam University Press.
Papadakis E and Bean C (1993) Popular support for the welfare state: A comparison between institutional
regimes. Journal of Public Policy 13(3): 227–254.
Peng I and Wong J (2010) East Asia. In: Castles FG, Leibfried S, Lewis J, Obinger H and Pierson C (eds) The
Oxford Handbook of the Welfare State. Oxford: Oxford University Press, 656–670.
Pfau-Effinger B (2005) Culture and welfare state policies: Reflections on a complex interrelation. Journal
of Social Policy 34(1): 3–20.
Pierson P (1993) When effect becomes cause: Policy feedback and political change. World Politics 45(4):
595–628.
Pierson P (1996) The new politics of the welfare state. World Politics 48(2): 143–179.
Powell M and Barrientos A (2004) Welfare regimes and the welfare mix. European Journal of Political
Research 43(1): 83–106.
Ragin C (1994) A qualitative comparative analysis of pension systems. In: Janoski T and Hicks AM (eds)
The Comparative Political Economy of the Welfare State. Cambridge: Cambridge University Press, 320–
345.
Sainsbury D (ed.) (1999) Gender and Welfare State Regimes. Oxford: Oxford University Press.
Saint-Arnaud S and Bernard P (2003) Convergence or resilience? A hierarchical cluster analysis of welfare
regimes in advanced countries. Current Sociology 51(5): 499–528.
Schröder M (2009) Integrating welfare and production typologies: How refinements of the varieties of capi-
talism approach call for a combination of welfare typologies. Journal of Social Policy 38(1): 19–43.
Scruggs L and Allan J (2006) Welfare state decommodification in eighteen OECD countries: A replication
and revision. Journal of European Social Policy 16(1): 55–72.
Scruggs LA and Allan JP (2008) Social stratification and welfare regimes for the twenty-first century:
Revisiting the three worlds of welfare capitalism. World Politics 60(4): 642–664.
Shalev M (2007) Limits and alternatives to multiple regression in comparative research. Comparative
Social Research 24(1): 261–308.
Skocpol T (1985) Bringing the state back in: Strategies of analysis in current research. In: Evans PB,
Rueschemeyer D and Skocpol T (eds) Bringing the State Back In. Cambridge: Cambridge University
Press, 3–37.
Skocpol T (1992) Protecting Soldiers and Mothers. Cambridge, MA: Harvard University Press.
Soede A and Vrooman C (2008) A Comparative Typology of Pension Regimes. Brussels: Centre for European
Policy Studies.
Stephens J (1979) The Transition from Capitalism to Socialism. London: Macmillan.
Stephens JD (2010) The social rights of citizenship. In: Castles FG, Leibfried S, Lewis J, Obinger H
and Pierson C (eds) The Oxford Handbook of the Welfare State. Oxford: Oxford University Press, 511–
525.
Svallfors S (2003) Welfare regimes and Western opinions: A comparison of eight Western countries. Social
Indicators Research 64(3): 495–520.
Svallfors S (2010) Public attitudes. In: Castles FG, Leibfried S, Lewis J, Obinger H and Pierson C (eds) The
Oxford Handbook of the Welfare State. Oxford: Oxford University Press, 241–251.
Svedin LM (2009) Organizational Cooperation in Crises. Farnham: Ashgate Publishing.
Tsebelis G (2002) Veto Players: How Political Institutions Work. Princeton, NJ: Princeton University Press.
Van Oorschot W, Opielka M and Pfau-Effinger B (eds) (2008) Culture and Welfare State: Values and Social

Policy in Comparative Perspective. Cheltenham: Edward Elgar.
Vermunt JK and Magidson J (2002) Latent class cluster analysis. In: Hagenaars JA and McCutcheon AL (eds)
Applied Latent Class Analysis. Cambridge: Cambridge University Press, 89–106.
Vrooman JC (2009) Rules of Relief; Institutions of Social Security, and Their Impact. The Hague: The
Netherlands Institute for Social Research|SCP.
Walker R (2008) European and American welfare values: Case studies in cash benefits reform. In: Van
Oorschot W, Opielka M and Pfau-Effinger B (eds) Culture and Welfare State; Values and Social Policy in
Comparative Perspective. Cheltenham: Edward Elgar, 117–140.
Weaver K (2010) Path and forks or chutes and ladders? Negative feedbacks and policy regime change.
Journal of Public Policy 30(2): 137–162.
Wilensky HL (1975) The Welfare State and Equality; Structural and Ideological Roots of Public Expenditures.
Berkeley: University of California Press.
Zeijl E, Te Poel Y, Du Bois-Reymond M, Ravesloot J and Meulman JJ (2000) The role of parents and peers
in the leisure activities of young adolescents. Journal of Leisure Research 32(3): 281–302.