& ‘Sifting the Wheat from the Chaff’: A Two …...Corporatist ustria Belgium Finland France Italy Social Democratic Denmark Sweden Norway Undeﬁned Germany Ireland Japan Netherlands

©

The Author(s)Journal Compilation ©

Blackwell Publishing Ltd,

Garsington Road, Oxford OX

DQ , UK and

Main Street, Malden, MA

, USA

S

P

& A

0144–5596V

. 41, N

. 1, F

2007,

. 1–28

Blackwell Publishing LtdOxford, UKSPOLSocial Policy & Administration

-

© Blackwell Publishing Ltd.

February

Original Article

&

, V

.

, N

.

, F

&

, V

.

, N

.

, F

‘Sifting the Wheat from the Chaff ’: A Two-dimensional Discriminant Analysis of Welfare State Regime Theory

Clare Bambra

Abstract

Welfare state modelling has long been an important strand within comparative social policy.

However, since the publication of Esping-Andersen’s ‘Worlds of Welfare’ typology, welfare state

classification has become particularly prominent and a multitude of competing typologies and

taxonomies have emerged. Each of these is based on different classification criteria, and each is

trying to capture what a welfare state actually does. The result is that the literature is in a state

of confusion and inertia as it is unclear which of these rival systems is currently the most accurate

and should be taken forward, and which are not and should perhaps be left behind. This article

extends Bonoli’s two-dimensional analysis of welfare state regimes by using multivariate analysis

of variance and discriminant analysis to compare and contrast the various classifications on

universal criteria. It also examines the usefulness of the two-dimensional approach itself and

suggests how it can be enhanced to benefit future attempts at holistic welfare state modelling. The

article concludes that there are some welfare state classifications that are more useful than others,

especially in terms of reflecting a two-dimensional analysis: it thereby ‘sifts the wheat from the

chaff ’ in terms of welfare state regime theory.

Keywords

Bonoli; Esping-Andersen; Multivariate; Regimes; Social expenditure; Welfare state

Introduction

Welfare state modelling has long been an important strand within comparativesocial policy, serving as a means of reducing the complexity of cross-nationalwelfare state provision (Wilensky and Lebaux

; Cutright

; Titmuss

; Wilensky

). However, it is only since the publication of Esping-Andersen’s ‘Worlds of Welfare’ thesis in

that the welfare state classifica-tion literature has become particularly prominent (Pierson

). Basedprimarily on the examination of labour market decommodification, Esping-Andersen proposed a threefold welfare state typology whereby Western

Address for correspondence:

Clare Bambra, Wolfson Research Institute, Durham University, QueensCampus, Stockton, TS

BH. Email: [email protected]

©

The Author(s)Journal compilation ©

Blackwell Publishing Ltd

S

P

& A

,

V

.

, N

.

, F

countries fell into one of three regime ideal-types: Liberal, Conservative orSocial Democratic. Esping-Andersen’s work has provoked an extensive ongo-ing debate in the literature, about which principles should be used to classifywelfare states (Alber

; Korpi and Palme

; Castles

; Abrahamson

; Kautto

; Bambra

a,

b); in which regimes particularcountries belong (Ginsburg

; Leibfried

; Castles and Mitchell

;Ferrera

); the number of different regime types (Leibfried

; Castlesand Mitchell

; Ferrera

; Bonoli

; Pitruzzello

; Bambra

b);the methodology of regime construction (Kangas

; Ragin

; Pitruzzello

; Bambra

); and the nature of gender stratification within differenttypes of welfare state (Lewis

; Orloff

; Sainsbury

; Bambra

).As a result of these criticisms a number of rival welfare state typologies

have emerged, each based on different classification criteria, and each tryingto capture in its own way what a welfare state actually does. Some haveextended the remit of the welfare state modelling literature to include aspectsof gender stratification and defamilization (Lewis

; Siaroff

; Lewisand Ostner

; Esping-Andersen

; Sainsbury

; Korpi

;Bambra

; Pascall and Lewis

), while others have examined the roleof welfare state services (Bambra

a,

b). In addition, the methodo-logical critique of Esping-Andersen’s work (for an overview, see Bambra

)has led to the production of numerous welfare state taxonomies, many ofthem based on the reworking of Esping-Andersen’s data, yet still suggestingalternative regimes and country classifications (Kangas

; Ragin

;Shalev

; Obinger and Wagschal

; Pitruzzello

; Wildeboer Schut

et al. ). This means that within the contemporary comparative socialpolicy literature, even among those typologies and taxonomies that only examineincome protection or the labour market aspects of welfare state provision,there are a number of contrasting claims for the existence of three (Esping-Andersen , ; Ragin ; Shalev ; Wildeboer Schut et al. ),four (Leibfried ; Castles and Mitchell ; Kangas ; Ferrera ;Bonoli ; Korpi and Palme ), and even as many as five different typesof welfare state regimes (Obinger and Wagschal ; Pitruzzello ).

The result is that the welfare state modelling literature is in a state ofconfusion and inertia as it is unclear which of these competing systems ofclassification is currently the most accurate or useful and which are less so.It is the purpose of this article to establish some clarity in this regard bycomparing the classifications and determining which are currently of themost utility: it will thereby ‘sift the wheat from the chaff ’ in terms of welfarestate regime theory.

Typologies and Taxonomies

Esping-Andersen’s original () analysis of decommodification levels ineighteen countries produced an initial threefold typology of welfare states:Liberal (Australia, Canada, Ireland, New Zealand, UK, USA), Conservative(Finland, France, Germany, Italy, Japan, Switzerland), and Social Demo-cratic (Austria, Belgium, Denmark, Finland, Norway, Sweden).1 In usingdecommodification, that is ‘the extent to which individuals and families can

© The Author(s) Journal compilation © Blackwell Publishing Ltd

S P & A, V. , N. , F

maintain a normal and socially acceptable standard of living regardless oftheir market performance’ (Esping-Andersen : ), his approach was anattempt to examine what a welfare state does, rather than how much moneyit is afforded (Esping-Andersen : ). Conversely, though, many of thecriticisms directed at his typology were made on the basis that it had not infact adequately measured what welfare states do, nor indeed managed tocorrectly classify many countries. This led to the construction of alternativetypologies, and taxonomies, each based on slightly different measures ofwelfare state labour market protection and each producing slightly differentresults (see tables and ). Typologies are theoretically informed categoriza-tions of welfare states while taxonomies are purely empirical.

On the basis of the cross-classification of the same eighteen nations interms of aggregate expenditure levels and degrees of benefit equality, Castlesand Mitchell () argued for the existence of a ‘Radical’ (Liberal subgroup)fourth regime (UK, Australia and New Zealand). Institutional analysis (Korpiand Palme ), cluster analysis (Kangas ), BOOLEAN comparativeanalysis (Ragin ), factor analysis (Shalev ), and principal componentanalysis (Wildeboer Schut et al. ) also supported the existence of a fourthregime type (see table ). Cluster analysis by Obinger and Wagschal ()and Pitruzzello () suggested another subgroup, this time within the Con-servative regime, and thereby laid claim to a possible fivefold welfare stateregime typology (table ). However, by extending the variety of countriesused, Leibfried (), Ferrera () and Bonoli () identified anotherdistinctive ‘Latin’, or ‘rudimentary’, fourth regime type (Spain, Portugal,Greece and, to a lesser extent, Italy and France – table ).

Tables and show the different results of each of the welfare stateclassifications, both in terms of the regimes asserted by each typology andthe constituent countries within them. A number of countries are positionedin the same regime type in almost all of the classifications. For example,Canada and the USA are positioned in the least generous (Liberal) regimetype in all of the typologies and taxonomies; similarly, Denmark, Norwayand Sweden always appear together in the most generous regime type (SocialDemocratic) regardless of which indicators are used to construct the typology.Furthermore, Greece, Portugal and Spain are considered to be the corecountries of the Latin regime, and Germany is the one exemplar of theConservative ideal-type model. The positioning of the other countries, however,is a more disputed matter. For example, Australia, Ireland, New Zealand andthe UK are positioned in either the Liberal or the Radical regime type;Austria, Belgium, Finland and the Netherlands are placed in either theConservative or the Social Democratic regime type; and Italy and France areplaced in either the Conservative or the Latin regime type (see tables and ).Most contentious is the case of Switzerland, which is placed in three differentregime types: Liberal (Castles and Mitchell ; Ragin ; Shalev ;Korpi and Palme ; Obinger and Wagschal ), Conservative (Esping-Andersen ; Ferrera ; Pitruzzello ), and Latin (Bonoli ).

Table also shows the variety of factors used to construct each of thedifferent welfare state typologies. All of these typologies are designed to capturethe income maintenance aspects of welfare state provision, yet they all do it

©

The A

uthor(s)Journal com

pilation © B

lackwell P

ublishing Ltd

S

P

&

A

, V

. , N. , F

Table

Welfare state typologies

Author Measures Welfare state regimes

Esping-Andersen ()

countries

*DecommodificationLiberal

AustraliaCanadaIrelandNew ZealandUKUSA

Conservative

FinlandFranceGermanyJapanItalySwitzerland

Social Democratic

AustriaBelgiumNetherlandsDenmarkNorwaySweden

Leibfried ()

countries

*Characteristics*Rights*Basic income

Anglo-Saxon

AustraliaNew ZealandUKUSA

Bismarck AustriaGermany

Scandinavian

DenmarkFinlandNorwaySweden

Latin Rim

FranceGreeceItalyPortugalSpain

Castles and Mitchell ()

countries

*Aggregate welfare expenditure*Benefit equality

Liberal

IrelandJapanSwitzerlandUSA

Conservative

GermanyItalyNetherlands

Non-Right Hegemony

BelgiumDenmarkNorwaySweden

Radical

AustraliaNew ZealandUK

© T

he Author(s)

Journal com

pilation © B

lackwell P

ublishing Ltd

S

P

&

A

, V

. , N. , F

Ferrera ()

countries

*Coverage*Replacement rates*Poverty rates

Anglo-Saxon

IrelandUK

Bismarck

AustriaBelgiumFranceGermanyLuxembourg NetherlandsSwitzerland

Scandinavian


Southern

GreeceItalyPortugalSpain

Bonoli () countries

*Social expenditure as % of GDP*Social expenditure financed via contributions

British

IrelandUK

Continental

BelgiumFranceGermanyLuxembourgNetherlands

Nordic


Southern

GreeceItalyPortugalSpainSwitzerland

Korpi and Palme ()

countries

*Social expenditure as % of GDP*Luxembourg income study*Institutional characteristics

Basic Security

CanadaDenmarkIrelandNetherlandsNew ZealandSwitzerlandUKUSA

Corporatist

AustriaBelgiumFranceGermanyItalyJapan

Encompassing

FinlandNorwaySweden

Targeted

Australia

Source: Based on Arts and Gelissen ().

Author Measures Welfare state regimes

Table

(Continued )

©

The A

uthor(s)Journal com

pilation © B

lackwell P

ublishing Ltd

S

P

&

A

, V

. , N. , F

Table

Welfare state taxonomies

Author Method Welfare state regimes

Kangas () countries

Cluster analysisLiberal

CanadaUSA

Conservative

AustriaGermany ItalyJapanNetherlands

Social Democratic


Radical

AustraliaIrelandNew ZealandUK

Ragin () countries

BOOLEAN comparative analysis

Liberal

AustraliaCanadaSwitzerlandUSA

Corporatist

AustriaBelgiumFinlandFranceItaly

Social Democratic

DenmarkSwedenNorway

Undefined

GermanyIrelandJapanNetherlandsNew ZealandUK

Shalev () countries

Factor analysisLiberal

CanadaJapanSwitzerlandUSA

Conservative

AustriaBelgiumFranceIrelandItaly

Social Democratic

DenmarkFinland NorwaySweden

Undefined

AustraliaGermanyNetherlandsNew ZealandUK

© T

he Author(s)

Journal com

pilation © B

lackwell P

ublishing Ltd

S

P

&

A

, V

. , N. , F

Obinger and Wagschal ()

countries


CanadaJapanSwitzerlandUSA

European

BelgiumFinlandGermanyIreland NetherlandsUK

Conservative Austria France Italy

Social Democratic

Denmark NorwaySweden

Radical

AustraliaNew Zealand

Pitruzzello () countries


CanadaIrelandUKUSA

Conservative

GermanyNetherlandsSwitzerland

Conservative–Bismarckian

AustriaFinlandFranceItalyJapan

Social Democratic

BelgiumDenmarkNorwaySweden

Radical

AustraliaNew Zealand

Wildeboer Schut et al. ()

countries

Principal component analysis

Liberal

AustraliaCanadaUKUSA

Conservative

BelgiumFranceGermany

Social Democratic

DenmarkNorwaySweden

Undefined

Netherlands

Source: Based on Arts and Gelissen ().

Author Method Welfare state regimes

Table

(Continued )

© The Author(s)Journal compilation © Blackwell Publishing Ltd

S P & A, V. , N. , F

using different indicators: decommodification (Esping-Andersen ); basicincome (Leibfried ); aggregate expenditure (Castles and Mitchell );poverty rates (Ferrera ; Korpi and Palme ); or social expenditure(Bonoli ; Korpi and Palme ). It could be, therefore, that each welfarestate typology is measuring a different aspect of welfare state provision andthereby drawing attention to different aspects of what welfare states do.2

However, many of the factors used to devise the diverse typologies are verysimilar, such as decommodification (Esping-Andersen ) and replacementrates and benefit coverage (Ferrera ), while others such as social expendi-ture as a percentage of GDP are shared by different typologies (Bonoli ;Korpi and Palme ). Perhaps, therefore, these diverse typologies andtaxonomies are ultimately measuring merely slightly different aspects of thesame underlying dimensions of the welfare state.

The Two Dimensions of Welfare

Indeed, Bonoli () argues that the welfare state typology literature can bedivided into two halves: one of these examines ‘how much’ (i.e. the quantity ofwelfare provision), while the other examines ‘how’ (the Bismarck–Beveridgefunding dichotomy). Bonoli argues that welfare state typologies (such as thoseoutlined in table ) measure one of these two underlying aspects of thewelfare state more than the other: they are one-dimensional categorizations.For example, he asserts that Esping-Andersen’s () typology measures the‘how much’ issue whereas the typology of Ferrera () encapsulates the‘how’ aspect more thoroughly. Therefore, regardless of the particular indi-vidual factors used by each typologist, there is in fact only one of two under-lying dimensions being measured. This leads Bonoli to use just two factors(social expenditure as a percentage of GDP and the proportion of contribu-tion financing and tax financing of social expenditure) to draw up a two-dimensional typology. He argues that, taken together, these two factorsrepresent both the ‘how much’ – the quantity issue – and the ‘how’ – theBeveridge tax-funded universalism versus the Bismarck contribution-basedsocial insurance system. As table shows, his typology is very similar to thosedrawn up using a more varied or numerous array of factors.

Bonoli’s work is very important in the context of this article as it suggeststhat, rather than emphasizing the differences between how the variouswelfare state typologies have been constructed and thereby resigning theliterature to one of indecision and incomparability, it is possible to use histwo-dimensional approach to compare all of the different typologies andtaxonomies on the same terms. It is therefore possible to determine which,if any, of the competing theories of welfare state regimes are the most usefulin terms of accounting for welfare state variation in one-dimensional (howmuch or how) and/or a two-dimensional manner (how much and how).

This article therefore utilizes a two-dimensional approach to examine thecompeting typologies and taxonomies that exist within the contemporarycomparative welfare state literature. The aims are fourfold: firstly, to deter-mine which of the income maintenance-based classifications are the mostuseful in terms of accounting for welfare state variance; secondly, which of


S P & A, V. , N. , F

the typologies measure one dimension more than the other and whichdimension has dominance overall in welfare state modelling; thirdly, toexplore which of the two underlying dimensions distinguishes most betweendifferent welfare state regimes; and finally, to compare the results of the moretheoretically derived typologies with the more empiricist taxonomies. Addi-tionally, the value of Bonoli’s two-dimensional approach will be given furtherimplicit empirical exploration. Overall, the analysis will ‘sift the wheat fromthe chaff ’ and enable certain typologies and taxonomies to be given moreprominence within discussions about welfare regimes in the comparativesocial policy literature.

Methods

Comparative social policy in general, and the construction of welfare statetypologies in particular, has seldom been underpinned by robust methodology(Kangas ; Ragin ; Shalev ; Pitruzzello ; Gough ;Bambra ). This has meant that many welfare state typologies are theo-retically rather than empirically informed. For example, Esping-Andersen’sThree Worlds typology has been extensively criticized on the basis of itsreliance on averaging and additive indexes (Fawcett and Papadopoulos ;Arts and Gelissen ; Bambra ). Some commentators, such as Kangas() or Ragin (), have responded to these problems by using methodssuch as cluster analysis or BOOLEAN comparative analysis (see table ) todevelop alternative, more empirically based, welfare state groupings. How-ever, while these taxonomies are clearly a methodological enhancement, theymerely serve to develop new, rival, welfare state classifications. These methodsare unable to test the relative merits of existing welfare state typologiesand thereby serve merely to further fuel discussions about the relative place-ment of certain countries and the number of different welfare state regimes(see tables and ). The use of statistical techniques that enable the moreextensive comparison and testing of the different welfare state typologies (andindeed, taxonomies), such as analysis of variance (ANOVA), multivariate analysisof variance (MANOVA), and discriminant analysis (DA), has long been advoc-ated within the welfare state modelling literature (Pitruzzello ).

MANOVA is an extension of ANOVA (analysis of variance) that enablesthe comparison of the mean values of different groups (in this case welfare stateregimes) for a variety of dependent variables (Weinfurt ; Tabachnick andFidell ). ANOVA calculates an F-statistic that is the ratio of the variancebetween the different groups (considered to be due to the independentvariable – group membership) divided by the variability within each of thegroups (considered to be due to chance). A large and significant F-statisticindicates that there is more variability between the groups than within themand that therefore the group means are not equal (Pallant ). In otherwords, it shows that the groups differ significantly in terms of the dependentvariable. Post-hoc tests, such as the Scheffe test, compare each group with allother groups to determine where the differences lie. ANOVA thereby enablesa comparison of whether the regimes (groups) in each of the different welfarestate typologies and taxonomies actually differ from one another in terms of


S P & A, V. , N. , F

the chosen dependent variables. In addition, through the eta2 effect size statistic,ANOVA enables comparisons to be made between the rival classifications onthe basis of how much of the variance in the dependent variable is accountedfor by the grouping or independent variable. In other words, eta2 helps todetermine which of the diverse welfare state typologies/taxonomies is themost useful because it accounts for the most amount of variation.

MANOVA extends this analysis to include more than one dependentvariable (Field ). It compares the differences in the means of the groupfor the dependent variables both individually, through separate ANOVAs, andtogether, by combining the dependent variables into one new linear com-posite dependent variable (Pallant ). Descriptive discriminant analysis (DA)extends the MANOVA analysis as it explores the underlying dimensions ofthe data and determines which weighted combination of scores on the twovariables best distinguishes between the different groups (Field ; Silvaand Stam ). The weightings of variables form a new composite variable:the discriminant function – a linear combination of the weightings and scores onthis variable (Cramer ). In this way, DA transforms the original set ofvariables into one or more new functions. DA can also be used to predict groupmembership (Silva and Stam ). Reflecting the MANOVA approach, anon-stepwise method of entry was used in the DA (Field ).

The choice of dependent variables is therefore very important in terms ofthe analysis that ANOVA, MANOVA and DA can provide. Statistically, thedependent variables need to be conceptually related but not highly cor-related (multicolinearity); be normally distributed (both univariate and multi-variate); have a linear relationship; and exhibit homogeneity of variance andcovariance (Pallant ). More importantly, though, the chosen variablesneed to reflect as far as possible the wide variety of factors and considerationsused to originally establish the different typologies which, as table shows,range from stratification (Esping-Andersen ) to poverty rates (Ferrera; Korpi and Palme ) and benefit equality (Castles and Mitchell). This is by no means an easy task especially when, in addition to thesevariations, the availability of cross-national data is limited. Two dependentvariables are used in this article: social expenditure as a percentage of GDP;and employer and employee contributions as a percentage of total socialsecurity receipts. These variables provide what Bonoli () refers to as atwo-dimensional approach to welfare state classification (see the discussionabove) as together they reflect both the quantity of welfare state provision(social expenditure as a percentage of GDP) and how that provision is funded(employer and employee contributions as a percentage of total social securityreceipts).

Although much of the welfare state modelling literature has been abouttrying to get beyond aggregate measures of funding and provision (see forexample Esping-Andersen : , or indeed table ), the social expenditureas a percentage of GDP variable is in fact highly correlated (see table ) witha number of the other indicators that have been used to construct welfarestate typologies. For example, the correlation between social expenditure asa percentage of GDP and poverty rates is −. (p < .). Furthermore,social expenditure as a percentage of GDP is used as one of the main measures

© T

he Author(s)

Journal com

pilation © B

lackwell P

ublishing Ltd

S

P

&

A

, V

. , N. , F

Table

Correlations between social expenditure as a percentage of GDP and other quantitative indicators of welfare state provision

Measure Associated author(s) Data source(s) Social expenditure as a percentage of GDP (mean –)

Correlatione P-value

Decommodificationa Esping-Andersen () Esping-Andersen () . .Bambra () . .

Replacement ratesb,c Ferrera () OECD () . .Esping-Andersen ()

Poverty ratesb,d Ferrera () OECD () −. .Korpi and Palme ()

a countries.b countries.cLong-term unemployment replacement rates (mean –).dPoverty rates (mean –).eCalculated using Spearman’s rho.


S P & A, V. , N. , F

in a number of welfare state typologies, such as Castles and Mitchell ()or Korpi and Palme (), and employer and employee contributions as apercentage of total social security receipts is used by Bonoli () andreflects Leibfried (). Therefore, although by no means exhaustive, thesetwo robust variables do offer a fairly indicative universal overview of themeasures that have been used in welfare state classifications (Bonoli ).

The data for the social expenditure as a percentage of GDP variable wereobtained from the OECD () and cover countries: Australia, Austria,Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy,Japan, Norway, the Netherlands, New Zealand, Portugal, Spain, Sweden,Switzerland, the UK, and the USA. These countries (see table ) representall of those used by the various different welfare state typologies (with theexception of Luxembourg, which has not been included in the analysis as itis only considered by Ferrera and Bonoli ). The social expenditureas a percentage of GDP variable is the mean calculated for the period – to minimize the influence of any year-on-year fluctuation. The data forthe contributions as a percentage of total social security receipts variablewere recalculated from the ILO ‘Cost of Social Security Database’ ().Subsequently, the data for this variable are older (ranging from to )and generally only represent one year’s worth of data. This reflects the factthat, although more recent data are available for this variable for the sixteenEuropean countries (EU ), for the five non-European countries the ILO() was the only source. Therefore, to ensure data comparability betweenthe European and non-European countries this data source was used for allthe countries. It should be noted that these data relate to different time periodsfrom those used in the original typologies and taxonomies. All analysis wascarried out using SPSS version ..

Results

The results of the one-way between groups ANOVAs are presented in tables, , and .3 Overall, they show that the differences between welfare stateregimes were statistically significant in the majority of the typologies andtaxonomies tested. The significant ANOVAs all show large (>.) eta2 effectsizes (Weinfurt ).

The results of the one-way ANOVAs for social expenditure as a percent-age of GDP (tables and ) show that the Ferrera () typology has thehighest eta2 score of .. This indicates that per cent of the variance inthe dependent variable – social expenditure – is accounted for by the in-dependent variable – Ferrera’s welfare state regimes. Similarly, the results intable show that the typologies of Esping-Andersen (), Leibfried (),Castles and Mitchell () and Bonoli () all achieve statistically signi-ficant differences (p < .) between welfare state regimes in terms of socialexpenditure and they all offer large effect sizes accounting for between percent (Esping-Andersen) and per cent (Leibfried, Castles and Mitchell) ofvariance. Indeed, only Korpi and Palme’s () typology does not achievestatistical significance (p = .), indicating that there is no significant differ-ence between their welfare state regimes in terms of mean social expenditure.


S P & A, V. , N. , F

The results for the taxonomies (table ) are similar as all bar one (Shalev) achieve statistical significance (p < .) and the eta2 values, rangingfrom . to ., indicate strong effect sizes. The Wildeboer Schut et al.() taxonomy offers the largest eta2 value of ., which suggests that theirwelfare state regime classification accounts for per cent of variance insocial expenditure as a percentage of GDP. Post-hoc comparisons using theScheffe test indicated that the Liberal (Anglo-Saxon, British, etc.) regimetype differed significantly (p < .) from the Social Democratic (Nordic,Scandinavian, etc.) regime type consistently across all the typologies andtaxonomies that achieved overall significance. This reflects the fact that in allthe welfare state models the Liberal-style regime had the lowest value forsocial expenditure as a percentage of GDP (ranging from . to .) whilethe Social Democratic-style regime had the highest (. to .). Thepost-hoc tests also suggested that in a number of cases (Esping-Andersen,

Table

Data used in ANOVA and MANOVA calculations

Social expenditure as a percentage of GDPa

Employer and employee contributions as a percentage of

total social security receiptsb

Australia . .Austria . .Belgium . .Canada . .Denmark . .Finland . .France . .Germany . .Greece . .Ireland . .Italy . .Japan . .Netherlands . .New Zealand . .Norway . .Portugal . .Spain . .Sweden . .Switzerland . .UK . .USA . .

aOECD () mean –.bILO () data for (except Greece and France, ; New Zealand, USA and Belgium, ; Australia and Norway, ; Canada, Japan and UK, ).


S P & A, V. , N. , F

Ferrera, Ragin, Wildeboer Schut et al.) the Liberal-style regime differed sig-nificantly from the Conservative-style (Bismarck, Corporatist, etc.) regime.None of the welfare state models indicated a significant difference between theother regime types, with the exception of Kangas (), where a significantdifference was also found between the Social Democratic regime type andthe Radical regime.

The results of the one-way ANOVAs for employer and employee contri-butions as a percentage of total social security receipts (tables and ) showa similar pattern for the typologies. Ferrera () again has the highest effectsize of .. On this variable all the other typologies achieve statistical signi-ficance and large effect sizes (. and above), including the Korpi and Palme

Table

ANOVA results for social expenditure as a percentage of GDP: typologies

Author Groups N Means F-statistic P-value Eta Scheffea

Esping-Andersen ()

. Liberal . . . . *. Conservative . *. Social .

DemocraticLeibfried () . Anglo-Saxon . . . . *

. Bismarck .. Scandinavian .. Latin Rim .


. Liberal . . . . *. Conservative .. Non-right .

hegemony. Radical .

Ferrera () . Anglo-Saxon . . . . *. Bismarck . *. Scandinavian .. Southern .

Bonoli () . British . . . . *. Continental .. Nordic .. Southern .

Korpi and Palme ()

. Basic security . . . – –. Corporatist .. Encompassing .. Targeted .

Note: All figures are rounded to two decimal places.aOnly significant group differences ( p < .) are displayed, e.g. * translates as the means of groups and differed significantly from one another. Scheffe test is suitable for unequal group sizes. It is considered to be one of the more conservative post-hoc tests and so the Bonferroni adjustment was not made (Cramer and Howitt ).


S P & A, V. , N. , F

model (p = .; eta2 = .). However, the results for the taxonomies aremuch less positive as only three (Kangas, Obinger and Wagschal, Pitruzzello)of the six models show evidence of a statistically significant difference betweenwelfare state regimes on this variable. Obinger and Wagschal’s model has thelargest effect size of ., suggesting that welfare state regime membershipaccounts for per cent of the variation in employer and employee contri-butions as a percentage of total social security receipts. The post-hoc Scheffe

Table

ANOVA results for social expenditure as a percentage of GDP: taxonomies

Author Groups N Mean F-statistic P-value Eta Scheffea

Kangas () . Liberal . . . . *. Conservative . *. Social .

Democratic. Radical .

Ragin () . Liberal . . . . *. Corporatist . *. Social .

DemocraticShalev () . Liberal . . . – –

. Conservative .. Social .

DemocraticObinger and Wagschal ()

. Liberal . . . . –. European .. Social .

Democratic. Radical . . Conservative .

Pitruzzello () . Liberal . . . . *. Conservative .. Social .

Democratic. Radical .. Conservative- .

BismarckWildeboer Schut et al. ()

. Liberal . . . . *. Conservative . *. Social .

Democratic



S P & A, V. , N. , F

tests show a great deal of variety in the models as to which regimes differsignificantly from one another, as while the taxonomy Scheffe tests (table )reveal significant differences between Conservative and Radical, there is noclear pattern of difference in the typologies (table ). The typologies showfour significant post-hoc differences: Liberal and Conservative (Esping-Andersen, Ferrera); Liberal and Latin (Leibfried); Conservative and Radical(Castles and Mitchell); Conservative and Social Democratic (Bonoli, Ferrera).Generally, these patterns suggest that significant differences in contributionsas a percentage of total receipts exist between the higher- and the lower-scoring regimes in each typology. In all the typologies (table ) and taxono-mies (table ), the Conservative regimes scored the highest (ranging from

Table

ANOVA results for employer and employee contributions as a percentage of total social security receipts: typologies


Esping-Andersen ()

. Liberal . . . . *. Conservative .. Social Democratic .

Leibfried () . Anglo-Saxon . . . . *. Bismarck .. Scandinavian .. Latin Rim .


. Liberal .. Conservative .. Non-right . . . . *

hegemony. Radical .

Ferrera () . Anglo-Saxon . . . . *. Bismarck . *. Scandinavian .. Southern .

Bonoli () . British . . . . *. Continental .. Nordic .. Southern .

Korpi and Palme ()

. Basic security . . . . –. Corporatist .. Encompassing .. Targeted .



S P & A, V. , N. , F

. to .) and the Liberal (. to .) and Radical the lowest (.to .).

The MANOVA results, in which the two dependent variables are com-bined (table ), reflect the patterns evident in the single dependent variableANOVAs. Most importantly, Ferrera’s typology with an eta2 of . accountsfor the most variance in the combined dependent variable, just as it did forboth of the dependent variables in their separate ANOVAs. Similarly, theKorpi and Palme typology once more failed to achieve statistical significance,

Table

ANOVA results for employer and employee contributions as a percentage of total social security receipts: taxonomies


Kangas () . Liberal . . . . *. Conservative .. Social .

Democratic. Radical .

Ragin () . Liberal . . . – –. Corporatist .. Social Democratic .

Shalev () . Liberal . . . – –. Conservative .. Social Democratic .


. Liberal . . . . *. European .. Social .

Democratic. Radical .. Conservative .

Pitruzzello () . Liberal . . . . *. Conservative . *. Social .

Democratic. Radical .. Conservative-Bismarck .


. Liberal . . . – –. Conservative .. Social .

Democratic



S P & A, V. , N. , F

suggesting that in the combined MANOVA analysis, the means scores oftheir welfare state regimes do not differ. There are also some interestingdifferences between the single ANOVAs and the MANOVA results. Forexample, Esping-Andersen’s typology accounted for per cent of the vari-ance in social expenditure as a percentage of GDP (table ) and per centof employer and employee contributions (table ). However, in the MANOVAanalysis of the combined variable Esping-Andersen’s typology only accountedfor per cent of the variance (table ). Conversely, Shalev’s typology, whichdid not achieve a statistically significant ANOVA for either variable, accountedfor per cent of the variance in the combined dependent variable. These resultsreflect the fact that the MANOVA analysis takes into account the correlationbetween the dependent variables. Also, the overall MANOVA results suggestthat some typologies and taxonomies perhaps reflect one variable – and there-fore one dimension of welfare provision – more than the other, while othersreflect both variables and therefore both dimensions of the welfare state.

Discriminant analysis (DA) offers the opportunity to examine these under-lying dimensions in more detail and determine which weightings of variables(functions) discriminate the most between the different groups in each typologyand taxonomy. Table shows the key statistics produced by the DA for eachof the welfare state typologies and taxonomies. Around half of the typologiesand taxonomies reflect two underlying dimensions (reflected by the numberof statistically significant discriminant functions) but the other half reflectonly one, suggesting that only one dimension distinguishes between the groups.Those typologies and taxonomies that have variate correlations which arepositive for both of the variables (social expenditure as a percentage of GDP,

Table

MANOVA results: typologies and taxonomies

Author F-statistic P-valuea Eta

Esping-Andersen () . . .Leibfried () . . .Castles and Mitchell () . . .Ferrera () . . .Bonoli () . . .Korpi and Palme () . . –Kangas () . . .Ragin () . . .Shalev () . . .Obinger and Wagschal () . . .Pitruzzello () . . .Wildeboer Schut et al. () . . .

Note: All figures are rounded to two decimal places.aCalculated using Pillai’s Trace, which is considered to be more robust for smaller samples and unequal group sizes (Tabachnick and Fidell : ).


S P & A, V. , N. , F

employer and employee contributions as a percentage of total social expend-iture receipts), such as function for Esping-Andersen, suggest that it is thecombination of the two variables that discriminates between the groups.Those typologies that have significant discriminant functions with one negativecorrelation and one positive correlation with the variables, such as function for Leibfried, indicate that it is the difference between the variables thatseparates the groups. For each significant discriminant function, the variablewith the higher canonical correlation coefficient contributes the most to groupseparation. The group variate centroids (mean function scores for eachgroup) show which groups the discriminant function distinguishes between.So, for example, in the case of Esping-Andersen’s typology (see table )there is only one significant discriminant function (function p = .), it isthe combination of the variables that distinguishes between the groups as bothcorrelations are positive (r = .; r = .), social expenditure contributesthe most to group separation, r = . (although the contribution of the othervariable is still large, r = .), and the group variate centroids suggest thatthe discriminant function discriminates the Social Democratic group fromthe other two groups, particularly the Liberal group (as the differencebetween these two scores, . and −., is the largest).

The DA shows that the welfare state classifications of Esping-Andersen,Ragin, Shalev, Pitruzzello and Wildeboer Schut et al. all reflect only oneunderlying dimension: in all cases except Shalev, it is the combination of thevariables that distinguishes between the groups; social expenditure contributesthe most to group separation (although in the cases of Esping-Andersen andPitruzzello the contributions correlation is also large). Also, the variate centroidssuggest that the function discriminates the most between the Liberalgroup and the Social Democratic group (Esping-Andersen, Shalev, Ragin);the Liberal group and the Radical group compared to the other groups(Pitruzzello); and between the Liberal group and the other groups (WildeboerSchut et al.). The typologies of Leibfried, Castles and Mitchell, Ferrera,Bonoli, Kangas, and Obinger and Wagschal encapsulate a two-dimensionalapproach as each has two significant discriminant functions. In each case, oneof the discriminant functions shows social expenditure as contributing mostto the discrimination between the groups and the other shows contributionsas a percentage of total social security receipts as contributing most to groupdiscrimination. The variate centroids suggest that the social expenditure-weighted discriminant function discriminates most between the Liberal group and the other groups (Leibfried, Bonoli, Kangas), while the discriminantfunction weighted more by contributions as a percentage of total social secur-ity receipts distinguishes most between the Social Democratic group andthe other groups (Leibfried, Ferrera, Bonoli, Kangas, Obinger and Wagschal).However, in the fourfold regime classification of Castles and Mitchell, thecontributions-weighted variate centroids distinguish most between the Radicalgroup (−.) and the others, particularly the Conservative group (.).Again, as expected from the MANOVA, the DA of the Korpi and Palmetypology was not significant.

To summarize, the results of the single ANOVAs showed that out of thetypologies, Ferrera’s accounted for the highest amount of variance for each

©

The A

uthor(s)Journal com

pilation © B

lackwell P

ublishing Ltd

S

P

&

A

, V

. , N. , F

Table

Discriminant analysis results: typologies and taxonomies

Author Discriminant functions

P-valuea Canonical variate correlations Group variate centroids

Social expenditure Contributions

Esping-Andersen ()

. . . −. –. . .

Leibfried () . . . −. . . . . −. . . . −. .


. . . −. . . −. . −. . . . −. −.

Ferrera () . . . −. . −. −. . −. . . . −. .

Bonoli () . . . −. . −. . . . −. −. −. . −.

Korpi and Palme ()

. .

Kangas () . . . −. . . −. . −. . . . −. −.

Ragin () . . . −. . . .

Shalev () . . −. −. −. . .


. . . −. . −. −. . . . . −. −. . . .

Pitruzzello () . . . −. . . −. . .


. . . −. . . .

aCalculated using Wilks-Lambda.


S P & A, V. , N. , F

of the two variables, and that differences between particular groups variedby typology. The MANOVA confirmed Ferrera’s position as the highest-scoring typology and Korpi and Palme as the lowest; and the DA suggestedthat some welfare state classifications were two-dimensional, while those thatwere one-dimensional reflected the social expenditure as a percentage of GDPvariable more than the contributions as a percentage of total social securityreceipts variable. Furthermore, the DA provided clarity, unattained via thesingle ANOVAs, that both variables distinguished most between the Liberalgroups and the other groups, or the Social Democratic groups and the others.

Finally, as the welfare state taxonomies were in part developed to be amore methodologically robust and empirical way of classifying welfare statesthan the typologies (Kangas ; Ragin ; Shalev ; Pitruzzello ),it is worth considering whether the taxonomies account for more of thevariance in the two dependent variables and in the combined MANOVAvariable than the typologies. One-tailed Mann-Whitney U tests were used tocompare the average eta2 values of the typologies and the taxonomies foreach of the variables and the combined MANOVA variable: no significantdifferences were found (social expenditure p = .; employer and employeecontributions p = .; combined p = .). This suggests that overall thetaxonomies did not account for significantly more variance than the typologies.

Discussion

One of the central aims of this article was to discover which of the variouscompeting welfare state classifications, both typologies and taxonomies, werethe most currently useful and which the least. In one respect the results haveenabled this aim to be fulfilled, as it is clear that some typologies account formore welfare state variance than others. However, the amount of variance inthe two variables – social expenditure and employee and employer contribu-tions – accounted for by a number of the typologies and taxonomies is infact very similar. Ferrera’s () typology achieved the highest eta2 values inboth of the single ANOVAs (. for social expenditure and . for employerand employee contributions), and it also scored the highest eta2 value in theMANOVA. It would therefore be simple to conclude that the Ferrera typologyaccounts for the most variance and is therefore the most useful of the com-peting welfare state classifications – the ‘wheat’, to follow through on themetaphor. However, the results are in fact much less clear-cut than this as anumber of the classifications achieved eta2 values very similar to Ferrera’s(and not significantly different from one another). For example, in the socialexpenditure single ANOVA, the typologies of Leibfried (), Castles andMitchell (), and Bonoli () also achieved eta2 values in the .s, asdid the taxonomies of Kangas (), Ragin () and Pitruzzello ().Indeed, on this variable Ferrera was not actually the highest-scoring classifi-cation, as the Wildeboer Schut et al. () taxonomy achieved an eta2 valueof .. Similarly, in the single ANOVA for employer and employee contri-butions, Leibfried (), Bonoli () and Obinger and Wagschal () allobtained similar eta2 scores to Ferrera. On the whole, the MANOVA resultswere likewise, with a number of typologies scoring similarly. This suggests that,


S P & A, V. , N. , F

far from having one single typology that stands out from the others, there arein fact a number of typologies that offer comparable levels of variance expla-nation. This poses somewhat of a dilemma as it means that, as a result ofthis endeavour, the welfare state modelling literature is actually little nearerto determining which individual classification model is the most useful.

However, while the analysis has been unable to determine conclusivelywhich classification system is the ‘wheat’, it has been more successful in‘sifting out the chaff ’: the classifications of Korpi and Palme () andShalev () were non-significant for the social expenditure single ANOVA,and the taxonomies of Ragin (), Shalev (), and Wildeboer Schutet al. () were non-significant in the employee and employer contributionssingle ANOVA. The MANOVA replicated the non-significant results forKorpi and Palme () but it also produced another, less expected result:despite achieving high eta2 scores in each of the single ANOVAs, Esping-Andersen’s ‘Three Worlds’ typology, by far the most prominent of all welfarestate classification systems, accounted for the lowest amount of variance (apartfrom the non-significant results) in the MANOVA. This suggests perhaps thathis typology is less useful than others when considering a two-dimensionalapproach. Overall, the analysis suggests that the typologies and taxonomiesshown to score low eta2 values or non-significant ANOVAs can perhapstherefore be given less accord within future discussions about the numberand consistency of welfare state regimes.

However, the typologies and taxonomies cannot be discounted or reinforcedon the basis of the ANOVA, MANOVA and eta2 analyses alone. The resultsof the DA must also be considered, particularly in terms of the extent towhich the different welfare state classifications measure one or two dimensionsof welfare state provision. Those typologies and taxonomies shown by theDA to measure only one or other of the two underlying welfare state dimen-sions of ‘how much’ and ‘how’ (Bonoli ) should be given less prominencethan those that are shown to reflect both aspects more adequately. The DAconfirmed that in some cases the typologies and taxonomies reflected onlyone dimension. For example, in the classifications of Esping-Andersen (),Ragin (), Shalev (), Pitruzzello () and Wildeboer Schut et al. (),social expenditure was much more prominent, whereas in other cases, mostnotably Leibfried (), Castles and Mitchell (), Ferrera () and Bonoli(), the DA showed that typologies can encapsulate both aspects of wel-fare state provision. Those that measure both aspects should surely be givenmore weighting in the welfare state modelling literature than those that onlyexamine one.

This is especially the case for those typologies, such as that of Esping-Andersen, for which claims are made that they ‘get beyond aggregate measuresof welfare state expenditure’ by creating alternative means of comparison,such as decommodification. The DA results show that for the majority oftypologies and taxonomies it is the social expenditure as a percentage ofGDP weighted variate that discriminates most between the regime types.When these results are taken together with the correlations between socialexpenditure as a percentage of GDP and other measures of welfare stateprovision, such as decommodification, it suggests that a number of welfare


S P & A, V. , N. , F

state typologies still only reflect expenditure and little more. In common withcriticisms made by those such as Castles and Mitchell () or Bonoli (),it suggests that a lot of the attempts to ‘get beyond’ using expenditure as ameans of comparing and contrasting different welfare states have failed, andthat perhaps this aspect of welfare state modelling still needs to be fullyachieved. The analysis in this article suggests that the attempt to ‘get beyond’aggregate expenditure comparisons should remain as one of the focuses ofcomparative research.

The two-dimensional analysis has also revealed which welfare state regimesdiffer most from one another in terms of either and both dimensions. Theresults of the post-hoc tests in the single ANOVAs suggested that in themajority of the different typologies and taxonomies, the social expenditurevariable (the ‘how much’ dimension) distinguished most between the Liberal-type and the Social Democratic-type regimes, in other words between thehighest and the lowest expenditure regimes in all of the typologies. Perhapsmore interestingly, though, in those classifications that included a Radicalfourth regime type, the single ANOVA for employer and employee contribu-tions revealed significant differences between the Conservative and Radicalregime types. It was, as expected from the literature, the Conservative regimesthat scored the highest in terms of employer- and employee-based funding;however, it was the Radical rather than the Liberal regimes that scored thelowest. These results reinforce one of the common tenets of the welfare stateliterature – the stark differences between the Liberal and Social Democraticregime types (however constituted) – but it also lends further support to thearguments of those such as Castles and Mitchell () that a distinctiveRadical regime exists, one which is more clearly Beveridgean than Bismarckian(Bonoli ) in orientation, even more so than the Liberal-style regime. Tosome extent, the DA reinforces these conclusions as the discriminant func-tions weighted more by social expenditure were shown to discriminate mostbetween the Liberal regime and the other regime types. However, while thepost-hoc tests in the contributions single ANOVA suggested differencesbetween the Conservative and the other welfare state regime types, in theDA, the weighted variate was better at separating the Social Democraticregime type from the others. This perhaps suggests that it is the lack ofcontributions-based funding that is creating the divisions. The DA does,however, go a little further than the single ANOVAs in terms of lendingsupport to proponents of the Radical regime type as in some of the typologiesand taxonomies, such as Pitruzzello (), the social expenditure-weightedvariate also discriminated well between the Liberal and Radical regimes, andin the case of Castles and Mitchell (), the Radical group was distin-guished from the others most by the contributions-weighted variate.

The two-dimensional approach proposed by Bonoli () and used withinthis analysis has therefore produced a means of comparing and contrastingthe different regime classifications that are currently circulating in the com-parative welfare state literature. Although not without its own limitations,which are acknowledged by Bonoli himself, the approach does offer a pro-ductive way forward in terms of enabling the construction of a welfare statetypology which reflects both aspects of the welfare state modelling literature,


S P & A, V. , N. , F

at least as it currently stands in terms of the income maintenance dimension.It is perhaps surprising, then, that the results in this article have not shownBonoli’s own typology, based as it is on the two dimensions alone, to be ofeven more utility. There are at least three reasons for this (but it needs to bereiterated that Bonoli’s typology accounts for a very high amount of variancein the MANOVA with an eta2 of .). On the one hand, Bonoli’s analysiswas much earlier than the one in this article and the data he used were froma different date and source. However, perhaps more influential was the factthat the method used by Bonoli to put countries into regime types reliedmerely on cross-classification (see Bonoli ), and therefore, unlike theMANOVA, it did not take into account any correlation between the twodimensions, and this may have led to incorrect classification. Alternatively, ofcourse, there could be a third underlying dimension of welfare state provisionthat Bonoli’s two-dimensional approach overlooks, such as the mix of cashbenefits and welfare state services (Castles ; Kautto ; Bambraa, b). The two-dimensional approach could therefore be extendedboth methodologically, by using a more robust system of classification suchas cluster analysis, and empirically, by including this third dimension – the‘how spent’ aspect of welfare state provision. Perhaps then the welfare statemodelling literature will finally succeed in obtaining a holistic typology.

Limitations

The research in this article is subject to a number of limitations. Firstly, aspointed out in the discussion section, it only examines two dimensions, whichare themselves based on two indicators – social expenditure as a percentageof GDP and employer and employee contributions as a percentage of totalsocial security receipts. If other indicators, or more indicators, had been usedthe results may have been different. Therefore, although multivariate analysiswas used, this means that it is likely that other, unaccounted-for factors areexerting influence on welfare state variation. Consequently, caution shouldbe applied to the results and their interpretation, not least as the statisticalcorrelations and associations discussed do not necessarily equate with expla-nation or causation. Furthermore, the data used for the contributions vari-able were prone to possible year-on-year variation and they were limited tothe years –. Also, it needs to be noted that the various typologies andtaxonomies were based on data from differing years (e.g. Esping-Andersen’sdata were from ) and so this article cannot comment on the original fitof the classifications; it is only able to examine their current relevance. Thetypologies and taxonomies may well have been more accurate at the timethey were originally constructed (e.g. in the s for Esping-Andersen), andthey may still be currently accurate if all their constituent variables areincluded (e.g. if decommodification indicators were reanalysed for the sfor Esping-Andersen, or institutional characteristics were somehow includedin the reassessment of the Korpi and Palme model).4 It is also important toacknowledge that welfare regime patterns change over time and that thisarticle has only examined the applicability of welfare state classifications atone point in time. It is quite possible that different typologies and taxonomies


S P & A, V. , N. , F

will be accurate at different time points. Finally, most typologies were per-haps intending to measure something other than the two dimensions, andindeed some of them are based on a variety of factors, not all of which arequantifiable. The analysis in this article is perhaps therefore limited by overlyquantifying and thereby marginalizing the more qualitative and theoreticalaspects of typology construction.

Conclusion

Although the analysis has been unable to determine with any certaintywhich individual classification is the most useful, it has provided evidencethat one subgroup of the diverse regime classifications is more useful thanthe others. This is particularly the case when examining whether the classi-fications encapsulate one dimension or two dimensions of the welfare state.It is proposed in this article that those typologies that, in the DA, reflect thetwo underlying dimensions should be given more prominence than thosethat only reflect one. Furthermore, the eta2 results for the ANOVAs and theMANOVA should also be taken into consideration, with those classificationsobtaining higher eta2 values accounting for more variation. This means thatthe four typologies of Leibfried (), Castles and Mitchell (), Ferrera() and Bonoli (), and the two taxonomies of Kangas () andObinger and Wagschal () emerge as the ‘wheat’ while the others, parti-cularly those that have non-significant differences between regime types,particularly Korpi and Palme () and to a lesser extent Shalev (), orwhich only reflect one dimension of the welfare state, such as Ragin (),Shalev (), and Wildeboer Schut et al. () and to a lesser extent Esping-Andersen () and Pitruzzello (), are the ‘chaff ’. In addition, the analysishas reinforced the fact that the differences between certain welfare state regimetypes are more pronounced, particularly those between the Liberal and theSocial Democratic types. However, it has also highlighted and upheld argu-ments for the existence of a distinctive Radical regime type. Methodologic-ally, this article has pioneered the use of ANOVA, MANOVA and DA interms of the comparative welfare state literature and has demonstrated thevalue of these methods in taking the modelling debate forward. Futureresearch should rely more on such sophisticated analytical approaches andexpand empirically upon the two-dimensional approach of ‘how much’ and‘how’ by adding the third dimension of ‘how spent’ in terms of the cashbenefits and welfare state services mix (Bambra a, b).

Notes. Esping-Andersen also reflects upon social stratification and the public–private mix

in welfare state provision. However, it is his decommodification index typologythat is widely cited (see, for example, Arts and Gelissen ).

. It should also be noted that the Korpi and Palme typology is also based on moretheoretical and qualitative data about the institutional characteristics of socialinsurance.

. It is more usual to run the MANOVA first and then follow up with the singleANOVAs.


S P & A, V. , N. , F

. An updated analysis of the full decommodification indicator set by Bambra ()suggests, however, that in the case of Esping-Andersen the threefold typology isno longer accurate. Furthermore, research by Scruggs and Allan () has ques-tioned the original validity of the threefold classification.

ReferencesAbrahamson, P. (), The welfare modelling business, Social Policy & Administration,

, : –.Alber, J. (), A framework for the comparative study of social services, Journal of

European Social Policy, , : –.Arts, W. and Gelissen, J. (), Three worlds of welfare or more? Journal of European

Social Policy, , : –.Bambra, C. (), The worlds of welfare: illusory and gender-blind? Social Policy and

Society, , : –.Bambra, C. (a), Worlds of welfare and the health care discrepancy, Social Policy

and Society, , : –.Bambra, C. (b), Cash versus services: worlds of welfare and the decommodifica-

tion of cash benefits and health care services, Journal of Social Policy, , : –.Bambra, C. (), Decommodification and the worlds of welfare: revisited, Journal

of European Social Policy, , : –.Bonoli, G. (), Classifying welfare states: a two-dimension approach, Journal of

Social Policy, , : –.Castles, F. G. (), Comparative Public Policy: Patterns of Post-war Transformation,

Cheltenham: Edward Elgar.Castles, F. G. and Mitchell, D. (), Worlds of welfare and families of nations. In

F. G. Castles (ed.), Families of Nations: Patterns of Public Policy in Western Democracies,Aldershot: Dartmouth.

Cramer, D. (), Advanced Quantitative Data Analysis, Maidenhead: McGraw-Hill.Cramer, D. and Howitt, D. (), The Sage Dictionary of Statistics, London: Sage.Cutright, P. (), Political structure, economic development, and national social

security programs, American Journal of Sociology, : –.Esping-Andersen, G. (), Citizenship and socialism: decommodification and solidarity

in the welfare state. In G. Esping-Andersen and L. Rainwater (eds), Stagnation andRenewal in Social Policy: The Rise and Fall of Policy Regimes, London: Sharpe.

Esping-Andersen, G. (), The Three Worlds of Welfare Capitalism, London: Polity.Esping-Andersen, G. (), Social Foundations of Post-industrial Economies, Oxford:

Oxford University Press.European Union (EU) (), European Social Statistics: Social Protection Expenditure and

Receipts, Luxembourg: Eurostat.Fawcett, H. and Papadopoulos, T. N. (), Social exclusion, social citizenship and

decommodification: an evaluation of the adequacy of support for the unemployedin the European Union, West European Politics, , : –.

Ferrera, M. (), The southern model of welfare in social Europe, Journal of EuropeanSocial Policy, , : –.

Field, A. (), Discovering Statistics Using SPSS for Windows, London: Sage.Ginsburg, N. (), Divisions of Welfare: A Critical Introduction to Comparative Social Policy,

London: Sage.Gough, I. (), Social assistance regimes: a cluster analysis, Journal of European Social

Policy, , : –.ILO (International Labour Organization) (), The Cost of Social Security. Available

at: http://www.ilo.org/public/english/protection/socfas/research/css/cssindex.htm(accessed April ).

http://www.ilo.org/public/english/protection/socfas/research/css/cssindex.htm


S P & A, V. , N. , F

Kangas, O. (), The politics of social security: on regressions, qualitative compari-sons and cluster analysis. In T. Janoski and A. Hicks (eds), The Comparative PoliticalEconomy of the Welfare State, Cambridge: Cambridge University Press.

Kautto, M. (), Investing in services in West European welfare states, Journal ofEuropean Social Policy, , : –.

Korpi, W. (), Faces of inequality: gender, class and patterns of inequalities indifferent types of welfare states, Social Politics, , : –.

Korpi, W. and Palme, J. (), The paradox of redistribution and the strategy ofequality: welfare state institutions, inequality and poverty in the Western countries,American Sociological Review, : –.

Leibfried, S. (), Towards a European welfare state. In Z. Ferge and J. E. Kolberg(eds), Social Policy in a Changing Europe, Frankfurt: Campus-Verlag.

Lewis, J. (), Gender and the development of welfare regimes, Journal of EuropeanSocial Policy, , : –.

Lewis, J. and Ostner, I. (), Gender and the evolution of European social policy.In S. Leibfried and P. Pierson (eds), European Social Policy, Washington, DC: BrookingsInstitution.

Obinger, H. and Wagschal, U. (), Das Stratifizierungskonzept in der Clustera-nalytischen Uberprüfung. In S. Lessenich and I. Ostner (eds), Welten des Wohlfahrts-kapitalismus: Der Sozialstaat in vergleichender Perspektive, Frankfurt: Campus Verlag,pp. –.

OECD (Organization for Economic Cooperation and Development) (), SocialExpenditure Database. Available at: www.oecd.org/els/social/expenditure (accessed March ).

Orloff, A. (), Gender and the social rights of citizenship: the comparative analysisof gender relations and welfare states, American Sociological Review, : –.

Pallant, J. (), SPSS Survival Manual, Maidenhead: McGraw-Hill.Pascall, G. and Lewis, J. (), Emerging gender regimes and policies for gender

equality in a wider Europe, Journal of Social Policy, , : –.Pierson, C. (), Beyond the Welfare State, London: Polity.Pitruzzello, S. (), Decommodification and the Worlds of Welfare Capitalism: a Cluster Analysis,

Florence: European University Institute.Ragin, C. (), A qualitative comparative analysis of pension systems. In T. Janoski

and A. Hicks (eds), The Comparative Political Economy of the Welfare State, Cambridge:Cambridge University Press.

Sainsbury, D. (), Gender, policy regimes and politics. In D. Sainsbury (ed.),Gender and Welfare State Regimes, Oxford: Oxford University Press.

Scruggs, L. and Allan, J. (), Welfare state decommodification in eighteen OECDcountries: a replication and revision. Available at: http://www.polisci.uconn.edu/Pages/People/FacultypercentPages/Scruggs/scruggspapers.htm (accessed January ).

Shalev, M. (), The Privatisation of Social Policy? Occupational Welfare and the WelfareState in America, Scandinavia and Japan, London: Macmillan.

Siaroff, A. (), Work, welfare and gender equality: a new typology. In D. Sainsbury(ed.), Gendering Welfare States, London: Sage.

Silva, A. and Stam, A. (), Discriminant analysis. In L. Grimm and P. Yarnold(eds), Reading and Understanding Multivariate Statistics, Washington, DC: AmericanPsychological Association.

Tabachnick, B. and Fidell, L. (), Using Multivariate Statistics, London: Allyn and Bacon.Titmuss, R. (), Social Policy, London: Allen and Unwin.Weinfurt, K. (), Multivariate analysis of variance. In L. Grimm and P. Yarnold

(eds), Reading and Understanding Multivariate Statistics, Washington, DC: AmericanPsychological Association.

http://www.polisci.uconn.edu/Pages/


S P & A, V. , N. , F

Wildeboer Schut, J., Vrooman, J. and de Beer, P. (), On Worlds of Welfare: Institu-tions and their Effects in Welfare States, The Hague: Social Office of the Netherlands.

Wilensky, H. (), The Welfare State and Equality, Berkeley: University of CaliforniaPress.

Wilensky, H. and Lebaux, C. (), Industrial Society and Social Welfare, New York:Russell Sage.

Documents

& ‘Sifting the Wheat from the Chaff’: A Two …...Corporatist ustria Belgium Finland France Italy Social Democratic Denmark Sweden Norway Undeﬁned Germany Ireland Japan Netherlands