DOI: 10.HH/j.1468-2397.2007.00486.x IntJSoc Welfare 2007: 16: 326-338 INTERNATIONAL journal of SOCIAL WELFARE ISSN 1369-6866 Defamilisation and welfare state regimes: a cluster analysis Bambra C. Defamilisation and welfare state regimes: a cluster analysis Int. J Soc Welfare 2007: 16: 326-338 © 2007 The Author(s), Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare. The role of gender as a source of social stratification within and between welfare states is increasingly being paid attention to in the welfare state regimes debate. Defamilisation has emerged as a potentially important concept in this context, as it enables the comparison and classification of welfare states in terms of how they facilitate female autonomy and economic independence from the family. However, the methodology used, or the understanding of the concept, limits existing defamilisation typologies. These typologies have therefore been unable to provide an accurate examination of welfare state variation using this concept and, indeed, have in some ways undermined and devalued the usefulness of defamilisation. This article uses cluster analysis to build upon previous research and resurrect the concept of defamilisation. In contrast to existing work in this area, the analysis produces a five-fold typology of welfare state regimes. This typology differs in many ways from existing models of welfare state regimes, although some core countries emerge as regime ideal types. The article concludes by reflecting on the broader implications of cluster analysis, and defamilisation, for welfare state modelling and future research in this area. Clare Bambra Durham University, Stockton-on-Tees Key words: typologies, cluster analysis, gender, welfare state regimes, defamilisation Clare Bambra, Wolfson Building, Durham University, Queens Campus, Stockton on Tees TS17 6BH, UK E-mail: clare.bambra@durham.ac.uk Accepted for publication September 29, 2006 Introduction Welfare state modelling is a longstanding strand within the comparative social policy literature (Outright, 1965; Esping-Andersen, 1990; Therborn, 1987; Titmus, 1974; Wilensky, 1975; Wilensky & Lebraux, 1958), and there is now a burgeoning amount of work (Pierson, 1998) about welfare state regimes and welfare state typologies (see Table 1). One of the most extensive critiques of this modelling has been made on the basis of gender (Bambra, 2004; Borchost, 1994; Bussemaker & Kersbergen, 1994; Daly, 1994; Gornick & Jacobs, 1998; Hobson, 1994; Langan & Ostner, 1991; Lewis, 1992; Lewis & Ostner, 1995; O'Connor, 1996; Orloff, 1993; Sainsbury, 1994, 1999; Siaroff, 1994; Trifiletti, 1999) as welfare typologies have assumed an overtly genderless, but covertly androcentric, approach (Sainsbury, 1994) and have thereby ignored gender as a form of social stratification (Arts & Gelissen, 2002). The few attempts at modelling that have acknowledged that relationships with, and experiences of, welfare states are mitigated by gender (Siaroff, 1994) have traditionally fallen into one of three types: 'gendering' existing welfare state typologies, such as the work of Orloff (1993) or O'Connor (1996) on Esping-Andersen's three worlds typology (Esping-Andersen, 1990, 1999); highlighting aspects of welfare state models that also work for women (Gornick & Jacobs, 1998; Trifiletti, 1999); or creating alternative more gender-focused systems of welfare state classification (Lewis, 1992; Lewis & Ostner, 1995; Pascall & Lewis, 2004; Sainsbury, 1999; Siaroff, 1994). Prominent within this latter approach are Lewis' breadwinner model (Lewis, 1992; Lewis & Ostner, 1995; Pascall & Lewis, 2004) and Sainsbury's (1999) work on public childcare provision. However, these approaches have tended to be limited by the focus on only one indicator, a small number of countries or a static, qualitative concept (Esping-Andersen, 1999). This has led to the more recent development of 'defamilisation' typologies (Bambra, 2004; Esping-Andersen, 1999; Korpi, 2000). The concept of defamilisation is often defined as 'the degree to which individual adults can uphold a socially acceptable standard of living, independently of family relationships, either through paid work or through social security provisions' (Lister, 1997: 173). This concept acknowledges that often 'the functional equivalent of market dependency for many women is family dependency' (Esping-Andersen, 1999: 45). Defamilisation can therefore 326 © 2007 The Author(s) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare. Published by Blackwell Publishing, 9600 Garsmgton Road, Oxford OX4 2DQ, UK and 350 Mam Street, Maiden, MA 02148, USA Defamilisation and welfare state regimes be utilised as a way of testing the extent to which welfare states, and welfare state regimes, facilitate female autonomy and economic independence from the family The concept has been used intermittently within welfare state modelling (Bambra, 2004; Esping-Andersen, 1999; Korpi, 2000; Pascall & Manning, 2000; Taylor-Gooby, 1996). However, it has often been inaccurately defined and operationalised by commentators as the extent to which welfare states decommodify the family. For example, both Esping-Andersen (1999) and Korpi (2000) utilise this conceptualisation of defamilisation, and their typologies subsequently rely either on factors that assess the extent to which welfare states support the family (Esping-Andersen, 1999) or on different family models (Korpi, 2000). Korpi's family support typology and Esping-Andersen's attempt to operationalise defamilisation are problematic because, instead of examining the extent to which welfare states facilitate women's autonomy and independence, their work actually assesses the extent to which welfare states support the family. For example, Korpi (2000) measures the type of family support provided by different welfare states, not the support given specifically to women. Similarly Esping-Andersen is essentially concerned with the decom-modification of family life, the extent to which the welfare state enables the family to survive independently of the market. It is perhaps, therefore, not very surprising that the results of these two approaches to defamilisation closely match Esping-Andersen's (1990) original 'androgynous' three worlds of welfare typology (see Table 1). An alternative definition of defamilisation, one that is much more related to the broader gender-based critique of welfare state modelling outlined earlier, has been operationalised in a previous article by this author (Bambra, 2004). Drawing on Taylor-Gooby (1996), this definition of defamilisation refers to the extent to which the welfare state enables women to survive as independent workers and decreases the economic importance of the family in women's lives. Defamilisation in this conceptualisation is therefore concerned with women's freedom from the family, rather than the freedom of the family (Bambra, 2004). Subsequently and unlike the defamilisation of Korpi or Esping-Andersen, it was not just about family policy, but about the extent to which the broader welfare state environment facilitates the participation of women in society. That is not to say that policies aimed at the family itself (such as public provision of childcare) will not also have an indirect effect on women's independence from the family, but that this particular conceptualisation of defamilisation focuses purely on the context in which the position of women is or is not promoted. In the previous article (Bambra, 2004), this author used four factors to create a defamilisation index that measured the relationship between women, the state and the family: (i) relative female labour participation rate; (ii) maternity leave compensation; (iii) compensated maternity leave duration; and (iv) average female wage. However, the resulting defamilisation typology again broadly confirmed Esping-Andersen's 'three worlds' typology with the production of three groupings of welfare states with broadly similar country compositions (Table 1). This earlier work thereby suggested that taking a more overtly gendered approach to welfare state modelling had little impact on the resulting typology and was perhaps, therefore, an unnecessary endeavour (Bambra, 2004). However, this operationalisation of defamilisation was extremely problematic as it copied the Esping-Andersen methodology of index-based regime construction (Esping-Andersen, 1990). Esping-Andersen's method has been subject to extensive methodological criticism, particularly around the additive nature of the indexes and the reliance upon averaging (Castles & Mitchell, 1993; Kangas, 1994; Pitruzzello, 1999; Ragin, 1994), the use of weighting (Bambra, 2006; Fawcett & Papadopoulos, 1997) , and also for the use of one standard deviation around the mean to classify the countries into regimes (Bambra, 2006; Fawcett & Papadopoulos, 1997), which meant that the resulting typology could only ever be three-fold: regime classification is either above (high/ social democratic), below (low/liberal), or between (medium/conservative) one standard deviation around the mean (Esping-Andersen, 1990: 54). In this way, a four- or five-fold classification of welfare states is methodologically impossible (Bambra, 2006; Papadopoulos, 1998) . In replicating this method the results of the earlier (Bambra, 2004) defamilisation typology were of limited value, not least because it was impossible to produce anything other than three regimes. However, it is important that the woman-focused definition of defamilisation used within this work is not also undermined by the methodological problems of index construction or how the resulting typology was interpreted. The concept remains as a useful means of comparing both between and within welfare states in terms of women's independence and it also reflects the broader, and now largely unacknowledged gender criticism of welfare state modelling. The concept of defamilisation retains much value for welfare state modelling and comparative social policy. It is the purpose of this article, therefore, to resurrect the defamilisation concept for welfare state modelling by using a more robust method of classification: cluster analysis (Gough, 2001; Kangas, 1994; Pitruzzello, 1999). The resulting welfare state typology(s) will be then compared with the original defamilisation typology (Bambra, 2004), other defamilisation typologies (Esping-Andersen, 1999; Korpi, 2000), and mainstream welfare state classifications (Table 1). The article will conclude by reflecting on the broader implications of the analysis for welfare state modelling and future research in this area. © 2007 The Authors) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare 327 Bambra Table 1. Welfare state classifications. Welfare state regimes Liberal Conservative Social democratic Latin rim Radical Bismarckian Esping-Andersen (1990, 1999) Australia Finland Austria 18 countries Canada France Belgium Ireland Germany Metherlands Mew Zealand Japan Denmark UK Italy Morway USA Switzerland Sweden Leibfreid (1992) Australia Austria Denmark France 15 countries Mew Zealand Germany Finland Greece UK Morway Italy USA Sweden Portugal Spain Castles & Mitchell (1993) Ireland Germany Belgium Australia 18 countries Japan Italy Denmark Mew Zealand Switzerland Metherlands Morway UK USA Sweden Ferrera (1996) Ireland Austria Denmark Greece 15 countries UK Belgium Finland Italy France Morway Portugal Germany Sweden Spain Luxembourg Metherlands Switzerland Bonoli (1997) Ireland Belgium Denmark Greece 16 countries UK France Finland Italy Germany Morway Portugal Luxembourg Sweden Spain Metherlands Switzerland Korpi & Palme (1998) Canada Austria Finland Australia 18 countries Denmark Belgium Morway Ireland France Sweden Metherlands Germany Mew Zealand Italy Switzerland Japan UK USA Pitruzzello (1999) Canada Germany Belgium Australia Austria 18 countries Ireland Metherlands Denmark Mew Zealand Finland UK Switzerland Morway France USA Sweden Italy Japan Korpi (2000) Australia Austria Denmark 18 countries Canada Belgium Finland Japan France Morway Mew Zealand Germany Sweden Switzerland Ireland UK Italy USA Metherlands Bambra (2004) Australia Austria Denmark 18 countries Japan Belgium Finland Mew Zealand Canada Morway USA France Sweden Germany Ireland Italy Metherlands Morway Switzerland UK Methods Cluster analysis, despite having many obvious benefits for welfare state classification, is a surprisingly under- used approach in comparative social policy (Gough, 2001). In cluster analysis, countries are classified on the basis of the combination of predetermined selection criterion (defamilisation factors) so that each country in 328 © 2007 The Author(s) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare Defamilisation and welfare state regimes Table 2. Defamilisation data 2003/2004. Relative female economic Maternity leave compensation Compensated maternity Average female wage activity rate for persons for duration covered leave duration (per cent of male aged 15-64^ (per cent of normal wages) (number of weeks) average wage)h 2003 2004 2004 2003 Australia 15 0 0 891 Austria 171 100 16 60k Belgium 16 75b 15 81 Canada 121 55<= 18 Denmark 9 90d 18 871 Finland 6 70 21 831 France 14 1001 16 781 Germany 16 100 14 74 Greece 231 100 17 82m Ireland 211 70 18 69 Italy 25 80 22 Japan 26 60 14 60 Metherlands 17 100 16 78" Mew Zealand 15 1001 14 80 Morway S 100e 52 88 Portugal 15 100 17 641 Spain 24 100 16 Sweden 4 801 69 91 Switzerland 18 80 14 69i UK 16 50a 26 79 USA 14 0 0 a Calculated as the difference between the female and male labour participation rate. For example, if the male participation rate was 78.9 per cent and the female participation rate was 76.4 per cent, then the relative female labour participation rate would be (-)2.5 per cent. b 82 per cent for the first 30 days, 755 for the rest (up to a ceiling), c 17-18 weeks depending on province (up to a ceiling), i Up to a ceiling. e 42-52 weeks parental leave (9 weeks reserved for mother) at 80-100 per cent, f Parental leave, 80 per cent for 390 days, 90 days at flat rate, a 90 per cent for first 6 weeks and flat rate thereafter, h Wages in manufacturing. No sex segregated data available, i Data from 2002; k Data from 2001; i Data from 1999; m Data from 1998; " Data from 2000. (Based on: Bambra, 2004; UN, 2005) a cluster is similar to the others in that cluster and different from countries in the other clusters (Gough, 2001; Grimm & Yarnold, 2000). In this way, the clusters represent different regime types. There are two more commonly used forms of cluster analysis: hierarchical and K-means (Gough, 2001; Pitruzzello, 1999). Hierarchical cluster analysis locates the closest pair of countries and combines them to form a pair; this (joining cases into pairs or joining two pairs) continues until all cases are in one cluster. Once countries are joined in a cluster, they remain joined throughout the rest of the analysis (Cramer, 2003; Gough, 2001). In this way, the clusters emerge from the data, facilitating the emergence of welfare state taxonomies. However, hierarchical cluster analysis is rather atheoretical and so it is often conducted alongside K-means cluster analysis (Cramer, 2003; Gough, 2001; Grimm & Yarnold, 2000; Pitruzzello, 1999). K-means enables the a priori specification of the number of clusters to be formed (Gough, 2001). This has a secondary benefit for the classification of welfare states on the basis of defamilisation, as it enables the testing of the number of different types of welfare state regimes (3, 4 or 5) suggested by the welfare state modelling literature (see Table 1). All calculations in this article use squared Euclidean distance and scale-standardised versions of the defamilisation factors. Analysis was carried out using SPSS version 11.0. The analysis uses three of the original four measures of defamilisation (Bambra, 2004): (i) relative female economic activity rate (expressed as the difference between the female and male economic activity rate and thereby controlling for variations in national economic activity rates); (ii) maternity leave compensation; and (iii) compensated maternity leave duration. The original fourth factor - average female wage expressed as a percentage of male average wage (Bambra, 2004) - has been excluded from the cluster analysis for two reasons: firstly, there was no data for four (Canada, Italy, Spain and the USA) of the 21 countries used in this new analysis (see Table 2); and, secondly, it was significantly correlated (r =-0.659, p = 0.01) with measure (i), relative female economic activity rate, and the results © 2007 The Authors) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare 329 Bambra of cluster analysis are undermined by multicolinearity (Grimm & Yarnold, 2000). It should be noted that the first measure, relative economic activity rate, differs only slightly from the relative labour market participation rate used in the earlier analysis (Bambra, 2004). These quantitative indicators were 'specifically selected to highlight a certain aspect of how the welfare state undermines female dependency on the family and facilitates their economic independence' (Bambra, 2004: 204). Relative female labour participation rate has been chosen because it indicates the extent to which the economy of the welfare state facilitates female employment. It is 'relative' because it is measured in relation to male employment levels, thus reducing the influence of different national unemployment rates. This factor provides a measure of one way in which women gain economic independence from the family, enter the public realm and gain access to certain social rights (Meyer, 1994: 67-68). Maternity leave compensation and compensated maternity leave duration are intended to show whether the welfare state provides economic support when women decide to have children or if it encourages reliance on the family. Maternity leave compensation shows the level of replacement income that women receive when they are absent from work due to pregnancy, whereas compensated maternity leave duration indicates the length of time for which women can take paid maternity leave (Bambra, 2004). The data for the defamilisation factors relate to 21 countries: Australia, Austria, Belgium, Canada, Denmark Finland France, Germany, Greece, Ireland Italy, Japan, Norway, The Netherlands, New Zealand Portugal, Spain, Sweden, Switzerland UK and the USA. These countries represent the original 18 OECD countries used by Esping-Andersen (1990) and later by Castles and Mitchell (1993), plus the Mediterranean countries added by Leibfreid (1992), Ferrera (1996) and Bonoli (1997). This analysis therefore offers a broader range of countries than previous defamilisation (Bambra, 2004; Esping-Andersen, 1999; Korpi, 2000) or gender-focused typologies (Lewis, 1992; Lewis & Ostner, 1995; Pascall & Lewis, 2004; Sainsbury, 1999; Siaroff, 1994). Data were obtained from the United Nations (UN, 2005) for the years 2003 and 2004. The data are therefore much more recent than that used in other typologies [e.g. Bambra (2004) uses data from 1996 and 1997]. Results The unstandardised data for each of the defamilisation measures are outlined in Table 2 and Figures 1, 2 and 3 show the spread of country scores for each of the three measures. Patterns are evident even in these univariate graphics, e.g. the Nordic countries have the lowest relative female economic activity rate showing that the gap between male and female economic activity 30 Belgium ■0-X-0-- « Switzerland ■Ok? O Ireland ■0-0-Oe? O O Greece Spain Italy Japan JJ-O-^-Oe? O Denmark Finland ■o-o-Ot? Canada ■ö"ö-«X*fl-fl-J)-J>J>J>^^^J>J>J>*J>**J>*J>J>***J>-ö'-ö'-ö'*fl'fl'-ö'*c? O USA ■0-e? Norway Sweden JJ-U-^-O-O-U-e? Figure 4. Hierarchical cluster analysis dendrogram (21 countries). 332 © 2007 The Author(s) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare Defamilisation and welfare state regimes Country 0 5 10 15 20 25 Austria Netherlands «■= France Portugal Germany New Zealand Greece Spain Belgium Switzerland Ireland Japan «•«•«•««■0--0--0--0-**«-0-0-0-0-0-(HH>««-0-= Italy .a-o-i* Denmark Finland Canada Figure 5. Hierarchical cluster analysis dendrogram - omitting Australia, Norway, Sweden and the USA (17 countries). that they are not overly influencing the nature of the other clusters. The K-means cluster analysis is used to examine further the five-fold typology produced by the hierarchical cluster analysis, and also to test the different number of welfare state regime types suggested by the mainstream welfare state modelling literature (as outlined in Table 1). K-means was conducted for K = 2, K = 3, K = 4 and K = 5. The typologies produced by the various K-means analysis are presented in Table 4. Overall, the K-means confirms a large degree of the results of the hierarchical analysis, as certain countries are clustered together throughout the analysis regardless of the changing number of clusters: Sweden and Norway, Greece, Spain, Italy and Japan, and Austria, The Netherlands, Germany, New Zealand, France and Portugal. Sweden and Norway remain isolated together throughout the K-means, even in K = 2. However, the USA and Australia are initially joined with other cases, most notably Canada, until they too are separated out at K = 4. Furthermore, in K — 4 and K — 5 the clusters very closely resemble those produced by the hierarchical analysis. For example, in K = 4, Greece, Spain, Italy and Japan, which formed cluster two in the hierarchical analysis, are also together in cluster three. Although in K — 5 they separate, with Spain and Greece joining cluster two, it should be noted that some countries are closer to the cluster centre than others and that Greece is 1.01 away from the centre and Spain 1.18. This suggests that they are less central to that cluster than other cases that exhibit smaller distances such as The Netherlands, Austria or Germany. Similarly, the largest cluster identified by the hierarchical analysis remains relatively intact across the various K-means analyses, and at K = 5 only Ireland is absent from the comparable cluster two. The only other country to fall into a different cluster than expected from the hierarchical analysis is Denmark, which is in cluster two in K = 5 rather than alongside Finland, Canada and the UK in cluster four. It should be noted that Denmark is the country furthest away from the centre of cluster two (1.4). This means that there are actually only four countries with different cluster memberships under the K — 5 analysis than that found in the hierarchical analysis: Denmark, Ireland, Greece and Spain. The K-means analysis therefore provides further support for the existence of a five-fold defamilisation typology. The K-means also provides other useful information about the clusters, as an examination of the distances between countries and the five cluster centres shows that some cases form the basis of a particular cluster, whilst others are less closely identified with it. Countries central to a cluster are: the USA in cluster one (distance from centre, 0), Austria and The Netherlands in cluster two (distance from centre, 0), Japan in cluster three (distance, 0), the UK in cluster four (distance, 0), and Sweden in cluster five (distance, 0). Other countries are less closely associated with the cluster in which they end up, e.g. Denmark (1.4), Greece (1.01) and Spain (1.18) in cluster two, Finland (1.84) in cluster four and Norway (1.47) in cluster five. This suggests that some countries are harder to classify than others and that their cluster membership is more fluid. It also suggests that certain core countries are fundamentally different from one another in terms of the combined defamilisation variate and therefore form the basis of distinct clusters. Table 5 shows the final cluster centres and how each of the three defamilisation measures contributes to discrimination between the clusters. The F-statistic (calculated using analysis of variance) shows which of these measures contributes the most to discriminating between the clusters (the larger the F-statistic the larger the discriminating power of the variable). For example, in the five-cluster typology (K — 5) the economic activity variable has a negative influence on clusters one, three and four but a positive influence on clusters two and five, and the maternity leave compensation variable has a negative influence on clusters one, four and five but a positive influence on clusters two and three. The maternity leave duration variable contributed the most to discriminating between the © 2007 The Authors) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare 333 Table 4. K-means clusters. Two clusters Three clusters Four clusters Five clusters K Country Distances K Country Distances K Country Distances K Country Distance^ 1 Australia 2.7 Australia 0.2 1 Australia 0.2 1 Australia 0.17 Austria 0.8 Canada 2.2 USA 0.0 USA 0.00 Belgium 0.1 Finland 3.0 Canada 1.1 USA 0.0 2 Austria 1.4 2 Austria 0.00 Denmark 1.4 Belgium 1.3 Belgium 0.84 Finland 1.9 2 Austria 1.2 Canada 1.3 Denmark 1.40 France 0.9 Belgium 1.6 Denmark 0.0 France 0.50 Germany 0.8 Denmark 2.5 Finland 0.9 Germany 0.21 Greece 1.3 France 1.7 France 0.9 Greece 1.01 Ireland 0.8 Germany 1.4 Germany 1.3 Metherlands 0.00 Italy 1.5 Greece 0.2 Metherlands 1.4 Mew Zealand 0.36 Japan 1.6 Ireland 1.1 Mew Zealand 1.1 Portugal 0.34 Metherlands 0.8 Italy 0.8 Portugal 1.1 Spain 1.18 Mew Zealand 0.9 Japan 1.4 UK 1.8 Switzerland 0.69 Portugal 0.9 Metherlands 1.2 Spain 1.5 Mew Zealand 1.5 3 Greece 1.4 3 Ireland 0.94 Switzerland 0.3 Portugal 1.5 Ireland 0.9 Italy 0.86 UK 1.1 Spain 0.0 Italy 0.9 Japan 0.00 USA 2.7 Switzerland 1.2 Japan 0.0 UK 2.2 Spain 1.4 4 Canada 0.87 2 Norway 0.7 Switzerland 1.5 Finland 1.84 Sweden 0.7 3 Norway 1.5 UK 0.00 Sweden 0.0 4 Norway 1.5 Sweden 0.0 5 Norway 1.47 Sweden 0.00 a Figures rounded to 1 decimal place; b Figures rounded to 2 decimal places. Table 5. K-means final cluster centres. Z-score Activity Compensation Duration Two Clusters 1 0.17 -0.05 -0.29 2 -1.64 0.44 2.71 F 8.05 0.41 83.28 Three clusters 1 -0.67 -1.50 -0.66 2 0.40 0.34 -0.19 3 -1.64 0.44 2.71 F 8.29 11.19 50.83 Four clusters 1 -0.21 -2.53 -1.31 2 -1.64 0.44 2.71 3 -0.31 0.29 -0.15 4 1.19 0.16 -0.19 F 17.09 14.00 76.58 Five clusters 1 -0.21 -2.53 -1.31 2 0.16 0.60 -0.26 3 -1.64 0.44 2.71 4 -0.74 -0.60 0.13 5 1.38 -0.22 -0.11 F 7.73 46.23 75.64 Figures rounded to 2 decimal places. clusters in all of the typologies: (respectively F = 83.28, F = 50.83, F = 76.58, F = 75.64), although in K = 5, the maternity compensation variable began to exert more influence (F = 46.23). Table 6 shows the distances between the final cluster centres and thereby indicates the extent to which they are similar or different from one another. In the five-fold typology, clusters two (Austria, The Netherlands etc.) and five (Norway and Sweden) are most similar to one another (1.48). Interestingly, the two most distant clusters are not one (USA and Australia) and five (Sweden and Norway), but one and three (Ireland Italy and Japan), at a distance of 5.20. Discussion The results of the hierarchical and K-means cluster analysis were very similar and therefore fairly conclusive: when analysed using cluster analysis, defamilisation produces a five-fold typology of welfare states with three broad clusters and two pairs (see Table 7). This undermines the three-fold typologies of Esping-Andersen (1990, 1999), Korpi (2000) and Bambra (2004), and lends support to those that assert the existence of more than three regime types (Bonoli, 1997; Castles & Mitchell, 1993; Ferrera, 1996; Korpi & Palme, 1998; Leibfreid 1992; Pitruzzello, 1999). This analysis particularly reinforces the decommodification cluster analysis of Pitruzzello (1999), which also found in favour of a five-fold classification of welfare states. However, the country clusters produced by the defamilisation analysis and the regimes within the literature typologies differ substantially in content (Table 7 compared with Table 1): generally within the Defamilisation and welfare state regimes Table 6. Distances between final cluster centres. Clusters Two clusters 1 2 1 3.54 2 3.54 Three clusters 1 2 3 1 2.18 4.00 2 2.18 3.54 3 4.01 3.54 4.01 Four clusters 1 2 3 4 1 5.20 3.05 3.23 2 5.20 3.16 4.06 3 3.05 3.16 1.51 4 3.23 4.06 1.51 Five clusters 1 2 3 4 5 1 3.32 5.20 2.46 3.05 2 3.32 3.48 1.56 1.48 3 5.20 3.48 2.93 4.19 4 2.46 1.56 2.93 2.18 5 3.05 1.48 4.19 2.18 Figures rounded to 2 decimal places. Table 7. Final defamilisation clusters (hierarchical and K = 5 combined). 1 2 3 4 5 Unclear cases Australia Austria Italy Canada Morway Denmark USA Belgium Japan Finland Sweden Ireland France UK Greece Germany Spain Netherlands Mew Zealand Portugal Switzerland welfare state modelling literature, the Nordic countries are clustered together (social democratic); the Anglo-Saxon countries are placed in either one (liberal) or two (liberal and radical) regimes; and the continental European countries and Japan are placed into one (conservative), two (conservative and latin), or three (conservative, latin and bismarckian) regime types. There are only minor similarities between these regimes and the defamilisation clusters, most notably the existence of a broad continental cluster (cluster two in Table 7). However, the overall picture is very different from that expected from the mainstream welfare state regime literature, and the clusters also differ significantly from previous analyses conducted using the concept of defamilisation. All three previous defamilisation typologies (Bambra, 2004; Esping-Andersen, 1999; Korpi, 2000) promoted a three-fold division of welfare states and even allowing for the fact they only used 18 countries compared with the 21 in the cluster analysis, there is very little overall similarity between their results and those presented here, not least because the 335 © 2007 The Authors) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare Bambra defamilisation cluster analysis resulted in a five-fold division of welfare states. However, on closer examination, it is clear that very few of the mainstream typologies actually agree with one another over the exact classifications of all countries: of the seven mainstream typologies outlined in Table 1 (Bonoli, 1997; Castles & Mitchell, 1993; Esping-Andersen, 1990; Ferrera, 1996; Korpi & Palme, 1998; Leibfreid 1992; Pitruzzello, 1999), only seven country classifications are really agreed upon by all seven (plus Greece, Spain and Portugal by those that include them): Canada and the USA; Germany (and to a lesser extent The Netherlands); Norway and Sweden (and to a lesser extent Denmark). Similarly, the defamilisation typologies of Esping-Andersen (1999), Korpi (2000) and Bambra (2004), whilst all concluding in favour of a three-fold typology of welfare states, only actually agree on the positioning of nine countries (50 per cent of the cases examined): Australia, New Zealand and the UK; and France, Germany and Italy, Denmark, Norway and Sweden. This suggests that some countries are considered to be more core to certain regime types than others, and that the regime membership of the majority of cases is disputed. This has important implications for the interpretation of the results of the defamilisation cluster analysis. Both the hierarchical and the K-means cluster analysis resulted in a five-fold typology, but within each of the clusters it is clear that some countries are more central than others. For example, cluster one consists of only the USA and Australia; in cluster two Germany Austria and The Netherlands are the core countries (closest in distance to the cluster centre - see Table 4); in cluster three it is Japan, for cluster four it is the UK; and cluster five again is just a pair - Norway and Sweden. This follows the broad pattern of the existence of certain core countries, particularly those towards the extremes, in all types of welfare state classification (Table 1). This means that the results of the defamilisation analysis, whilst perhaps very different in how certain less clear countries are classified or indeed on the number of regimes suggested share certain core characteristics with the rest of the welfare state classification literature. Overall then, this suggests that some countries are more homogeneous across different policy areas and are therefore always found in the same welfare state regime regardless of which particular area is being examined: the USA, Germany and The Netherlands, and Denmark Norway and Sweden emerge as the overall most distinctive and thereby core countries in all welfare state modelling. The other countries are more heterogeneous in their policy profile and are therefore much more vulnerable to exhibiting fluctuating regime membership depending on which policy is under examination (Bambra, 2005b). These variations across different policy areas have led to the questioning of the entire concept of welfare state regimes and the possibility of accurate welfare state typologies (Kasza, 2002). However, certain core countries, which are found to be distinctive regardless of what particular variables or policy provision is under examination, are the 'ideal type' regimes around which the others cluster and their proximity to one or another may vary across different policy areas. Clearly, more work on countries such as Japan or Italy, which are more difficult to classify, is required and the fluid nature of classification needs to be recognised; the welfare state regimes concept does not therefore need to be abandoned but requires an acknowledgement that some countries are more easily classified than others. Defamilisation, or other concepts that reflect gender stratification, may therefore be useful in future welfare state modelling by facilitating a more common classification of these disputed cases. In the previous analyses of defamilisation, the similarities between the defamilisation regimes and more mainstream typologies led to the authors concluding that separate gendered or defamilisation analysis was not necessary (Bambra, 2004; Esping-Andersen, 1999). In the cases of Esping-Andersen (1999) and Bambra (2004), the similarities were due largely to the method of analysis (Bambra, 2006), and for Korpi (2000) and Esping-Andersen (1999) the flawed conceptualisations of defamilisation may also have been influential. In contrast, the more methodologically robust cluster analysis in this article, whilst lending some support to the idea of core regime countries, has highlighted significant differences between mainstream and defamilisation-based analysis. However, whether these differences are sufficient enough to support a move towards separate gender-based welfare state classifications is a more difficult issue. Conducting separate gender analysis removes it from the mainstream and as has been largely the case to date, it is therefore treated as an appendix to the main debate and analysis. The results of this article, especially in terms of highlighting the different welfare state arrangements that emerge when a concept such as defamilisation is used suggest, above all, that analysis which acknowledges that gender is a form of social stratification within welfare states and welfare state regimes should be overtly, not implicitly or absentmindedly undertaken as part of mainstream analysis. Combining gender stratification concepts, such as defamilisation, with others like the decommodification of labour (Esping-Andersen, 1990) or welfare services (Bambra, 2005a,b), in a sound methodological manner is the only way in which to build a truly holistic classification of welfare states and one in which there are potentially less disputed cases. Conclusion This cluster analysis of defamilisation has produced a five-fold welfare state typology. This is in stark contrast 336 © 2007 The Author(s) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare Defamilisation and welfare state regimes to the three-fold typologies proposed by previous analyses of defamilisation (Bambra, 2004; Esping-Andersen, 1999; Korpi, 2000). However, in common with the mainstream literature and previous defamilisation research, certain countries have emerged as core components of welfare state regimes. Most notably, the defamilisation analysis has highlighted the centrality of the USA, of Germany and The Netherlands, and of Norway and Sweden, as the foundation blocks of three regimes and at least in terms of defamilisation, two more regimes, the UK and Japan. Much of this is in common with the main thrust of broader comparative welfare state analysis; however, certain clear differences have been highlighted and these suggest that defamilisation, or other concepts that are capable of encapsulating gender as one of the stratifications within welfare states and welfare state regimes, would be a useful conceptual addition to analyses of income protection and social expenditure (Bonoli, 1997; Castles & Mitchell, 1993; Esping-Andersen, 1990; Ferrera, 1996; Korpi & Palme, 1998; Leibfreid 1992; Pitruzzello, 1999) or welfare service provision (Bambra, 2005a,b). The usefulness of the concept of defamilisation has therefore been highlighted and rejuvenated by the use of a more robust method of analysis. Cluster analysis, and other more objective means of analysing welfare state differences and similarities, should be more apparent within future attempts at welfare state modelling. Previous attempts at welfare state classification, especially in terms of defamilisation (Bambra, 2006; Esping-Andersen, 1999), have been undermined by the limited methods used (Bambra, 2006) and the prominent reliance upon eyeballing the data (Gough, 2001) and fitting it to a priori assumptions (Bambra, 2006). Above all, then, this article has demonstrated the benefits of utilising a more methodologically robust approach to welfare state modelling. Future attempts at classification should therefore build upon this approach and develop a more holistic means of comparing welfare states, one which overtly acknowledges the role of gender stratification, and does so in a methodologically mature way. Limitations The analysis in this article is subject to a number of limitations. Firstly, defamilisation is measured using only three of the four original measures, as relative female wage had to be excluded on the grounds of multicollinearity Furthermore, the conceptualisation of defamilisation in this article, reflecting its original usage (Bambra, 2004), is centred on the labour market and subsequently, it does not examine other, more institutional, aspects of welfare state provision. If other indicators, or more indicators, had been used the conclusions may have been different. Similarly as cluster analysis requires quantifiable indicators, more 'qualitative' measures that may have been able to encapsulate aspects of institutional arrangements relevant to defamilisation (such as provision for income splitting within pension systems or pension contributions for time spent providing family care, e.g. Switzerland's Erziehungsgutschriften and Betreuungsgutschriften) could not be included in the analysis. The resulting defamilisation-based welfare state regimes are therefore unable to reflect such institutional matters. Subsequently caution should be applied to the results and their interpretation. Furthermore, as the data used in the analysis related only to the years 2003/2004, it is possible that the same analysis may produce different results if repeated for other years. Lastly, cluster analysis and other statistical techniques of regime construction could be viewed as overly quantifying to the detriment of the more qualitative and theoretical aspects of typology construction. References Arts W, Gelissen J (2002). Three worlds of welfare or more? Journal of European Social Policy 12: 137-158. Bambra C (2004). The worlds of welfare: Illusory and gender-blind? Social Policy and Society 3: 201-212. Bambra C (2005a). Worlds of welfare and the health care discrepancy. Social Policy and Society 4: 31-41. Bambra C (2005b). Cash versus services: worlds of welfare and the decommodification of cash benefits and health care services. Journal of Social Policy 34: 195-213. Bambra C (2006). Decommodification and the worlds of welfare: Revisited. Journal of European Social Policy 16: 73-80. Bonoli G (1997). Classifying welfare states: a two-dimension approach. Journal of Social Policy 26: 351-172. Borchost A (1994). Welfare State Regimes, Women's Interests and the EC. In: Sainsbury D, ed. Gendering Welfare States. London, Sage. Bussemaker J, Kersbergen K (1994). Gender and the Welfare State: Some Theoretical Reflections. In: Sainsbury D, ed. Gendering Welfare States. London, Sage. Castles FG, Mitchell D (1993). Worlds of Welfare and Families of Nations. In: Castles FG, ed. Families of Nations: Patterns of Public Policy in Western Democracies. Aldershot, Dartmouth. Cramer D (2003). Advanced Quantitative Data Analysis. Maidenhead, Mcgraw-Hill. Cutright P (1965). Political structure, economic development, and national social security programs. American Journal of Sociology 70: 537-550. Daly M (1994). Comparing Welfare States: Towards a Gender Friendly Approach. In: Sainsbury D, ed. Gendering Welfare States. London, Sage. Esping-Andersen G (1990). The Three Worlds of Welfare Capitalism. London, Polity. Esping-Andersen G (1999). Social Foundations of Post-Industrial Economies. Oxford, Oxford University Press. Fawcett H, Papadopoulos TN (1997). Social exclusion, social citizenship and decommodification: an evaluation of the adequacy of support for the unemployed in the European Union. West European Politics 20: 1-30. Ferrera M (1996). The Southern Model of Welfare in Social Europe. Journal of European Social Policy 6: 17-37. Gornick J, lacobs I (1998). Gender, the welfare state and public employment: a comparative study of seven industrialised countries. American Sociological Review 63: 688-710. © 2007 The Authors) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare 337 Bambra Gough I (2001). Social assistance regimes: A cluster analysis. Journal of European Social Policy 11: 165-170. Grimm L, Yarnold P (2000). Reading and understanding more multivariate statistics. Washington, American Psychological Association. Hobson B (1994). Solo Mothers, Social Policy Regimes, and the logics of Gender. In: Sainsbury D, ed. Gendering Welfare States. London, Sage. Kangas O (1994). The Politics of Social Security: on Regressions, Qualitative Comparisons and Cluster Analysis. In: Janoski T, Hicks A, eds. The Comparative Political Economy of the Welfare State. Cambridge, Cambridge University Press. Kasza G (2002). The Illusion of Welfare Regimes. Journal of Social Policy 31: 271-287. Korpi W (2000). Faces of inequality: Gender, class and patterns of inequalities in different types of welfare states. Social Politics 1: 127-191. Korpi W, Palme J (1998). The paradox of redistribution and the strategy of equality: welfare state institutions, inequality and poverty in the Western countries. American Sociological Review 63: 662-687. Langan M, Ostner I (1991). Gender and Welfare: Towards a Comparative Framework. In: Room G, ed. Towards a European Welfare State. Bristol, SAUS Publications. Leibfried S (1992). Towards a European Welfare State. In: Ferge Z, Kolberg JE, eds. Social Policy in a Changing Europe. Frankfurt, Campus-Verlag. Lewis J (1992). Gender and the Development of Welfare Regimes. Journal of European Social Policy 2: 195-211. Lewis J, Ostner I (1995). Gender and the Evolution of European Social Policy. In: Leibfreid S, Pierson P, eds. European Social Policy. Washington DC, Brookings Institute. Lister R (1997). Citizenship: Feminist Perspectives. London, Macmillan. Meyer T (1994). The German and British welfare state as employers: patriarchal or emancipatory. In: Sainsbury D, ed. Gendering Welfare States. London, Sage. O'Connor JS (1996). From women in the welfare state to gendering welfare state regimes. Current Sociology 44: 1- 130. Orloff A (1993). Gender and the Social Rights of Citizenship: The Comparative Analysis of Gender Relations and Welfare States. American Sociological Review 58: 303-328. Papadoupolos TN (1998). Welfare support for the unemployed: a comparison of social policy responses to unemployment in 12 EU member states. PhD thesis, University of York, UK. Pascall G, Lewis J (2004). Emerging Gender Regimes and Policies for Gender Equality in a Wider Europe. Journal of Social Policy 33:3 73-394. Pascall G, Manning N (2000). Gender and social policy: comparing welfare states in Central and East Europe and the former Soviet Union. Journal of European Social Policy 10: 240-266. Pierson C (1998). Beyond the Welfare State. London, Polity. Pitruzzello S (1999). Decommodification and the Worlds of Welfare Capitalism: a Cluster Analysis. Florence, European University Institute. Ragin C (1994). A qualitative comparative analysis of pension systems. In: Janoski T, Hicks A, eds. The Comparative Political Economy of the Welfare State. Cambridge, Cambridge University Press. Sainsbury D (1994). Gendering Welfare States. London, Sage. Sainsbury D (1999). Gender, Policy Regimes and Politics. In: Sainsbury D, ed. Gender and Welfare State Regimes. Oxford, Oxford University Press. Siaroff A (1994). Work, welfare and gender equality: a new typology. In: Sainsbury D, ed. Gendering Welfare States. London, Sage. Taylor-Gooby P (1996). The Response of Government: Fragile Convergence. In: George V, Taylor-Gooby P, eds. European Welfare Policy. London, Macmillan. Therborn G (1987). Welfare State and Capitalist Markets. Acta Sociologica 30: 237-254. Titmuss R (1974). Social Policy. London, Allen and Unwin. Trifiletti R (1999). Southern European welfare regimes and the worsening position of women. Journal of European Social Policy 9: 49-64. UN (2005). Statistics and indicators on men and women, downloadable from http://unstats.un.org/unsd/demographic/ products/indwm/ [date last accessed: April 10, 2005]. Wilensky H (1975). The Welfare State and Equality. Berkeley, University of California Press. Wilensky H, Lebaux C (1958). Industrial Society and Social Welfare. New York, Russell Sage. 338 © 2007 The Author(s) Journal compilation © 2007 Blackwell Publishing Ltd and the International Journal of Social Welfare Copyright of International Journal of Social Welfare is the property of Blackwell Publishing Limited and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.