Introduction
The path-breaking work of Esping-Andersen (Reference Esping-Andersen1990, Reference Esping-Andersen1999) has faced criticism on conceptual (for example, Borchost, Reference Borchost and Sainsbury1994; Kasza, Reference Kasza2002) and methodological (for example, Pitruzello, Reference Pitruzello1999; Bambra, Reference Bambra2006) grounds. However, his case selection of eighteen welfare capitalist states has been the subject of relatively few comments (Shalev, Reference Shalev2007; Ebbinghaus, Reference Ebbinghaus2012). Similarly, while the ‘welfare modelling business’ (Abrahamson, Reference Abrahamson1999) has flourished and expanded its territory beyond the original eighteen countries to Southern Europe, Central or Eastern Europe, East Asia and Latin America, the issue of case selection has rarely been considered. However, the issue is regarded as crucial for validity and reliability in cross-national comparative studies (Geddes, Reference Geddes1990; Ebbinghaus, Reference Ebbinghaus2005).
This article reviews case selection practices in welfare modelling. It poses the following questions: what approaches to case selection have been adopted by Esping-Andersen (Reference Esping-Andersen1990) and other architects of welfare regime typologies; and what are the stated justifications for their case selections? In order to answer these questions, this article critically analyses thirty-three studies that have attempted to produce their own welfare regime typologies based on Esping-Andersen's (Reference Esping-Andersen1990) welfare regime concept or methodology. It consists of five parts. First, it briefly discusses the theoretical background of case selection in social policy. Second, it briefly explains the selection process. Third, it critically reviews five different approaches to case selection. The fourth and fifth parts consist of a discussion and conclusion. It is concluded that the key issue of case selection in welfare modelling has been neglected, and some methodological and theoretical suggestions for case selection are provided.
Case selection in cross-national studies
Neither standard statistical textbooks in comparative social science nor basic social research design texts (for example, Robson, Reference Robson2011; Bryman, Reference Bryman2012) tend to discuss the case selection issue in qualitative or quantitative cross-national studies (Ebbinghaus, Reference Ebbinghaus2005). Even comparative case studies and broader cross-national studies tend to pay limited attention to this issue (Hug, Reference Hug2003; but see Blatter and Haverland, Reference Blatter and Haverland2012: 64, 217). Some comparative political researchers have emphasised the importance of case selection for both internal and external validity, and warned against possible selection bias in either quantitative or qualitative studies (Geddes, Reference Geddes1990; Hug, Reference Hug2003). Geddes (Reference Geddes1990: 132–3) analyses how inappropriate case selections can lead to two types of erroneous inference. First, any characteristic that the selected cases share can be misinterpreted as a cause affecting dependent variables. It is erroneous if all other cases have the same characteristic too. Second, there is the problem of assuming that a (lack of) relationship between variables within the selected cases reflects relationships in the entire population of cases, especially when the selected cannot represent the whole. She implies that the first mistake is relatively more common in qualitative studies, and the second in quantitative work. In either case, ‘how the cases you choose affect the answers you get’ (Geddes, Reference Geddes1990). Ebbinghaus (Reference Ebbinghaus2005, Reference Ebbinghaus2012) goes on to illustrate some complex problems, particularly involving macro-level quantitative comparative analysis. The first problem is the ‘illusion of random sampling’, since a set of countries is rarely drawn from a random sampling, and in practice is mostly preselected by researchers. The second is the stratified sample problem, which results from largely heterogeneous characteristics of nations with, for example, variant populations and GDPs, which can violate the homogeneity assumption in inferential statistics. The third is the ‘outlier problem’, indicating the common practice of excluding cases for reasons of data availability or applicability. This practice can lead to inconsistent case selection, for example, from fourteen to twenty-two member states from the same OECD membership. Mills et al. (Reference Mills, van de Bunt and de Bruijn2006) point at the choice given to researchers between a small and a relatively large sample size which can pose problems. A large group of cases with a small number of variables may lead to ‘superficial, though potentially statistically sound results’ (Reference Mills, van de Bunt and de Bruijn2006: 622), while a few cases with relatively numerous variables may hinder effective testing of causal models.
According to Ebbinghaus (Reference Ebbinghaus2005: 135), ‘it is astonishing how often quantitative researchers use macro-level data . . . without any explicit reflection on the (pre-)selection’. Ebbinghaus (Reference Ebbinghaus2012: 12) urges researchers to ‘be more explicit about the rationale for selecting cases and specify the scope conditions when making generalisations beyond the observed cases’.
Methods
The author searched the published literature for studies of welfare typologies in two steps. The first step involved synthesising lists compiled in previously published reviews of Arts and Gelissen (Reference Arts, Gelissen, Castles, Leibfried and Lewis2010) and Ferragina and Seeleib-Kaiser (Reference Ferragina and Seeleib-Kaiser2011), with eleven and twenty studies respectively. Of these thirty-one studies, ten overlap and two articles fall short of this article's criteria − that a study should produce its own typology. In the end, nineteen studies remain. The second step finds an additional fourteen articles through ‘backward and forward reference list checking’ (Gough et al., Reference Gough, Oliver and Thomas2012), including scanning the titles of the reference lists of the selected nineteen studies, and searching Google Scholar by finding studies published in English that have referred to the already selected studies. This resulted in a total of thirty-three studies. Key information in relation to case selection was extracted from each included literature item, thereby enabling synthesis and comparison related to the primary research question.
Five approaches of case selection
The thirty-three studies can be divided into five groups by their approaches: Esping-Andersen's (Reference Esping-Andersen1990) original eighteen nations; OECD members, non-OECD regions, non-OECD nations, and ‘all’ nations. Each approach has its (lack of) selection criteria and respective (lack of) justification.
Approach 1: Esping-Andersen's eighteen nations
The first group consists of Esping-Andersen's (Reference Esping-Andersen1990) original eighteen nations of ‘the worlds of welfare capitalism’ (Castles and Mitchell, Reference Castles and Mitchell1992; Kangas, Reference Kangas, Janoski and Hicks1994; Ragin, Reference Ragin, Janoksi and Hicks1994; Shalev, Reference Shalev1996; Korpi and Palme, Reference Korpi and Palme1998; Bambra, Reference Bambra2004; Bambra, Reference Bambra2005, Reference Bambra2006; Scruggs and Allan, Reference Scruggs and Allan2006, Reference Scruggs and Allan2011; Ferragina et al., Reference Ferragina, Seeleib-Kaiser and Tomlinson2012; Talme, Reference Talme2013). They choose the same cases because some focus on re-testing Esping-Andersen's welfare regimes in terms of validity (Bambra, Reference Bambra2006; Scruggs and Allan, Reference Scruggs and Allan2006; Reference Scruggs and Allan2011), while others re-analyse Esping-Andersen's data with new methods of factor analysis (Shalev, Reference Shalev1996), cluster analysis (Kangas, Reference Kangas, Janoski and Hicks1994) and multiple correspondence analysis (Ferragina et al., Reference Ferragina, Seeleib-Kaiser and Tomlinson2012). However, it is argued that Esping-Andersen (Reference Esping-Andersen1990, Reference Esping-Andersen1999) has never been clear on his selection and its implications (Shalev, Reference Shalev2007; Ebbinghaus, Reference Ebbinghaus2012). Esping-Andersen (Reference Esping-Andersen1990: 111) simply notes that his study is ‘limited to the eighteen major industrialized capitalist democracies . . . Ours is therefore not a sample, but a universe of comparable nations. Hence, no conclusions can be generalised beyond this rather unique group of political economies’.
However, he has been ‘very reluctant’ to add more regime-clusters to his original three for the controversial cases of East Asian and Southern European nations (Arts and Gelissen, Reference Arts and Gelissen2002: 153). Esping-Andersen (Reference Esping-Andersen1999: 88) himself emphasises that ‘the desired explanatory parsimony would be sacrificed’ if more welfare regime types are added to his three archetypes in applying the concept beyond the original eighteen nations.
Given this, it is unclear if his core eighteen nations (1) represent the worlds of welfare capitalism despite his above-mentioned denial (i.e. the nature of the sample), (2) equal the worlds of welfare capitalism (i.e. population), (3) just include some welfare states with available data (i.e. nations without comparable data excluded) or (4) are simply a result of his own particular method of selection. Without clear justification, his choice seems ‘theoretically arbitrary and should always be open to challenge’ (Shalev, Reference Shalev2007: 297). It is also not clear which interpretation each subsequent researcher has chosen in their individual studies.
Nine years after the publication of Three Worlds of Welfare Capitalism, Esping-Andersen's (Reference Esping-Andersen1999) selection appears again to be arbitrary, or even mysterious, as he adds Spain and Portugal to the core nations, perhaps in response to his critics over the omission of Mediterranean nations. However, this time he omits Switzerland, one of the original eighteen nations, without explanation except for its single appearance as a ‘liberal’ welfare regime in one footnote (1990: 77).
Many subsequent studies uncritically follow the pool of nations, with the exception of Korpi and Palme (Reference Korpi and Palme1998), who provide two justifications for the selection of the same eighteen nations. First, they explain that these eighteen countries each have ‘a history of uninterrupted political democracy during the post-World War II period with more than 1 million inhabitants’ (Reference Korpi and Palme1998: 665). The second reason is that their data are limited and contain ‘information on the development of social insurance programs in 18 OECD countries’ (ibid.). Whether it is coincidence or not, Korpi and Palme (Reference Korpi and Palme1998) provide some justification for Espng-Andersen's selection. However, it is questionable whether current comparative studies should maintain the set of ‘worlds of welfare capitalism’ provided twenty-five years ago, with, for example, larger OECD membership and more refined international datasets (see below).
Approach 2: more than eighteen OECD nations
The second group of studies looks beyond the original eighteen nations but within OECD membership. The first candidates were Southern European nations (Leibfried, Reference Leibfried, Ferge and Kolberg1992; Siaroff, Reference Siaroff and Sainsbury1994; Ferrera, Reference Ferrera1996; Bonoli, Reference Bonoli1997; Obinger and Wagschal, Reference Obinger and Wagschal2001; Powell and Barrientos, Reference Powell and Barrientos2004; Castles and Obinger, Reference Castles and Obinger2008; Schröder, Reference Schröder2008). The discussions on the nature of a Mediterranean welfare regime were especially facilitated by these states’ entrance into the then European Community (EC) in the 1980s. The three Southern European nations of Greece, Spain and Portugal, ignored by Esping-Andersen (Reference Esping-Andersen1990), had been founding members of the OECD since 1961, but became new members of the then nine-nation group of the EC in 1981 (Greece) and 1986 (Spain and Portugal). Leibfried (Reference Leibfried, Ferge and Kolberg1992) was one of the earliest to suggest the addition of three Southern European nations to add the fourth ‘Latin Rim’ category to Esping-Andersen (Reference Esping-Andersen1990)s three worlds. As his title of Towards a European Welfare State? indicates, the focal point at that moment was the integrated European social policy (so that the non-European nations of Japan and Canada were excluded from his selection). Ferrera also contends that the Southern Europeans were increasingly visible with ‘the entry into the European Community’ (Reference Ferrera1996: 18). Since then, Southern Europeans almost became regulars in the welfare regime studies. For example, since 2000, out of twenty-five studies providing new typologies, Spain appears twelve times, compared to fifteen times for Japan, one of the original members (see appendix). Small countries such as Luxembourg (five times – Siaroff, Reference Siaroff and Sainsbury1994; Bonoli, Reference Bonoli1997; Soede, Reference Soede2004, and others) and Iceland (four times – Siaroff, Reference Siaroff and Sainsbury1994; Saint-Arnaud and Bernard, Reference Saint-Arnaud and Bernard2003, and others) are rarely included. Their relatively tiny populations can violate the homogeneity assumption of inferential statistics (Ebbinghaus, Reference Ebbinghaus2012).
East Asian nations are the next additions to the welfare modelling business. Japan, the only non-Western nation among the original eighteen nations, was arguably the most controversial case among the core group. Esping-Andersen (Reference Esping-Andersen1997) devotes an article to convincing sceptics that his welfare regime typology could be adapted to this ‘hybrid or unique’ nation. However, Ferragina et al. (Reference Ferragina, Seeleib-Kaiser and Tomlinson2012: 594) choose to exclude Japan from the original group, because ‘Japan's political and societal system is rooted in the tradition of neither liberalism, nor Catholicism nor social democracy’. South Korea, the new entrant, and the only other Asian nation in the OECD, is called ‘intriguing’ (Esping-Andersen, Reference Esping-Andersen1999: 90) or ‘theoretically important’ (Hudson and Kühner, Reference Hudson and Kühner2009: 39), but has been largely ignored, with a few exceptions (for example, Hudson and Kühner, Reference Hudson and Kühner2009).
Some Central or Eastern European nations have been rarely analysed. The Czech Republic, Hungary and Poland, which joined the OECD in 1995, 1996 and 1996, respectively, have largely been overlooked with only a few exceptions (for example, Castles and Obinger, Reference Castles and Obinger2008). Turkey is the most curious case. Even though it was one of the twenty OECD founding members in 1961, it has been virtually invisible. Even some studies explicitly expressing OECD member nations as a target group ignore Turkey without any explanation (Siaroff, Reference Siaroff and Sainsbury1994; Castles and Obinger, Reference Castles and Obinger2008; Hudson and Kühner, Reference Hudson and Kühner2009). It is arguably due to its cultural and religious uniqueness as the only Muslim OECD member nation, its low GDP per capita of US$1,567 in 1980, which was less than one tenth of that of Sweden (World Bank, 2014) and its low social spending of 3.2 per cent of GDP in 1980, which was one fifth of the OECD average (OECD, 2014). Overall, in short, ‘it seems odd that . . . much of the welfare modelling research still fails to encompass the far larger group of nations that can be presumed to be ‘welfare capitalist’ states than was case in the 1980s’ (Hudson and Kühner, Reference Hudson and Kühner2012: 53).
Approach 3: non-OECD regions
The third group concentrates on a particular region outside the traditional OECD boundary. The aim is in part to identify ‘a geographically based cluster of states that seem to share some distinctive features’ (Hay and Wincott, Reference Hay and Wincott2012: 62). The key question is whether the welfare regime typology can be applied to emerging welfare states with different socioeconomic, political and cultural backgrounds, or even developing nations which can hardly be defined as welfare states. Attention has been fixed relatively more on three regions (for example, Esping-Andersen, Reference Esping-Andersen1996; Arts and Gelissen, Reference Arts, Gelissen, Castles, Leibfried and Lewis2010) including East Asia (see Kwon, Reference Kwon1997; Wilding, Reference Wilding2008), Central or Eastern Europe (CEE, see Fenger, Reference Fenger2007; Aidukaite, Reference Aidukaite2009), and Latin America (see Franzoni, Reference Franzoni2008; Barrientos, Reference Barrientos2009). Each region has at least two OECD members, such as Japan and South Korea; the Czech Republic, Estonia, Hungary, Slovakia, Slovenia and Poland; and Chile and Mexico. In the studies of these regions, cases were selected at roughly three different levels: ‘individual nation’, ‘sub-region(s)’ and ‘the whole region’. In the case of CEE, these consist of single case studies of individual nations such as Poland (Keryk, Reference Keryk2010), Hungary (Lelkes, Reference Lelkes2000) and the Czech Republic (Saxonberg and Sirovátka, Reference Saxonberg and Sirovátka2009); sub-regions such as the Central European ‘Visegrad Four’ (Cerami, Reference Cerami2008), three Baltic states (Aidukaite, Reference Aidukaite2004), Balkan nations (Sotiropoulos et al., Reference Sotiropoulos, Neamtu and Stoyanova2003) and the entire CCE region (Deacon, Reference Deacon and Jones1993; Bohle, Reference Bohle2007). The varying number of cases in the regional studies might seem to be a curious phenomenon. However, these can also be regarded as experimental attempts to check the adaptability of welfare regime concepts by testing them at various regional levels. Consequently, each case selection should be reviewed and assessed based on individual theoretical and methodological approaches.
Approach 4: non-OECD nations
The fourth group also looks beyond the OECD nations but with broader approaches. They categorise a longer list of nations across the continents. Apart from Rudra (Reference Rudra2007), the majority of works have been undertaken by researchers associated with the University of Bath (Gough and Wood, Reference Gough and Wood2004; Abu Sharkh, Reference Abu Sharkh2006; Wood and Gough, Reference Wood and Gough2006; Abu Sharkh and Gough, Reference Abu Sharkh and Gough2010; Gough and Abu Sharkh, Reference Gough and Abu Sharkh2011; Gough Reference Gough and Kennett2013). They aim to extend the analysis of welfare regime theory from the developed worlds ‘to the developing worlds’ (Abu Sharkh and Gough, Reference Abu Sharkh and Gough2010: 27), or from the North to the ‘Global South’ (Gough, 2007: 2). Based on detailed studies of Latin America, East Asia and Africa, Gough and Wood (Reference Gough and Wood2004) claim to radically redefine basic concepts of the welfare regime. First, they extend the components of the welfare mix to cover the role of the community-based relationship and international organisations, such as the EU and OECD. Second, they replace the decommodification index with more general measures of welfare outcomes, as Esping-Andersen's key index is not applicable to developing nations where the labour market is immature. They point to two ‘meta-welfare regimes’ of ‘informal security regimes’ and ‘insecurity regimes’ (Gough and Wood, Reference Gough and Wood2004), in addition to the three conventional welfare regimes.
They face another obstacle in terms of which nations to choose from all of the developing nations. Wood and Gough select sixty-one nations from ‘countries which experience problematic states as well as imperfect markets’ (2006: 1696), without any clear criteria provided, apart from the clue that they ‘rely on commonly available data covering all the transitional, underdeveloped and developing countries outside the OECD’ (ibid.: 1703). However, they include five newer OECD members, the Czech Republic, Poland, Hungary, the Slovak Republic and Chile, but other OECD entrants such as Turkey, South Korea and Mexico are excluded. Abu Sharkh and Gough (Reference Abu Sharkh and Gough2010) are much clearer in their choice of sixty-five nations out of ‘all countries’ which are (1) non-OECD nations; (2) have a population of more than three million; (3) have available comparable data. Gough (Reference Gough and Kennett2013: 205) stresses that their wider focus could ‘provide a bridge between thinking about social policy in the North and the South, without imposing Northern framework and solutions on the rest of the world’. Their wider selection of nations contributes to understanding welfare or ‘illfare’ in developing nations but cannot be free from criticism. For example, Gough places scores of nations in a rather singular order ‘according to the distances of their final cluster centres from the OECD welfare states’ (Gough, Reference Gough and Kennett2013: 216). Paradoxically, the broad-brush depiction and categorisation of geographically remote, culturally different, socioeconomically distinctive nations summoned from all over the world and analysed with several variables, could be considered another case of ‘Northern framework imposition’.
Approach Five: all nations
The last group is certainly the most ambitious as it encompasses all (OECD and non-OECD) nations (Abu Sharkh, Reference Abu Sharkh2009; Hudson and Kühner, Reference Hudson and Kühner2012). Hudson and Kühner (Reference Hudson and Kühner2012) collect fifty-fvie nations from the OECD's ‘fringe’ in addition to its core. Their selection criteria are relatively clear. Out of eighty-one nations classified by the World Bank as upper-middle and high-income and with a total population over 500,000, fifty-five nations could provide comparable datasets. They include thirty OECD members and twenty-five ‘on the cusp of the OECD's level of wealth’ (Reference Hudson and Kühner2012: 43). They note that this sample is not without problems: ‘We accept that these strategies may be seen as contentious by some readers and therefore would argue that our findings should only be seen as tentative at this point’ (Reference Hudson and Kühner2012: 47). They also attempt to justify the widening sample by proposing fuzzy set ideal type analysis as methodologically having ‘considerable advantages in allowing us to broaden the sample of countries analyzed’ (ibid.: 53), and go as far as to contend that ‘it seems foolish to make the first step in such an approach the abandonment of the best datasets available to social policy analysts’ (ibid.: 53). Their argument seems to be that the fuzzy set ideal type analysis with a wider sample of nations is the second worst option, with doing nothing as the worst. Their stance sounds appealing because they at least acknowledge their apparent limitations and seek arguably the best possible way to encompass the emerging welfare models rather than keep fixating on the core OECD group.
Abu Sharkh (Reference Abu Sharkh2009) has probably been the most ambitious and encompassing study when it comes to its scale. As she emphasises, ‘In the strictest sense, it is not even a sample since almost all the nation-states of the world are included – provided they report data or let the UN or World Bank “guestimate” data in negotiations with the country’ (Reference Abu Sharkh2009: 13). She ends up with seventy-nine nations, after excluding those with populations fewer than three million and without a comparable dataset. However, the author later accepts that the mix of developing and developed nations with only a few variables would complicate the analysis ‘by increasing the range of variation and obscuring important differences within the rest of the world’ (Abu Shark and Gough, Reference Abu Sharkh and Gough2010: 32).
Discussion
Of the thirty-three studies, seventeen still focus around Esping-Andersen's eighteen core group. If we count the studies including Southern European nations, over-two thirds of welfare modelling studies are contained roughly within the ‘18 + 3’ group, except for only seven studies (Soede, Reference Soede2004; Rudra, Reference Rudra2007; Castles and Obinger, Reference Castles and Obinger2008; Hudson and Kühner, Reference Hudson and Kühner2009, Reference Hudson and Kühner2012; Abu Sharkh, Reference Abu Sharkh2009; Abu Sharkh and Gough, Reference Abu Sharkh and Gough2010). This lopsided trend inevitably invites criticism for an ‘ethnocentric’ (Walker and Wong, Reference Walker, Wong and Kennett2013) approach.
This customary narrow focus seems to have two reasons. First, the theoretical reason sees a tension between ‘isolationism’ and ‘expansionism’. The ‘isolationists’ in this context can be defined as a group of researchers who have been sceptical or cautious of the broader application of the welfare modelling business as they view welfare regime concepts as basically rooted in the Western tradition and consequently as having certain limitations for exportation out of the West (for example, Ferragina and Seeleib-Kaiser, Reference Ferragina and Seeleib-Kaiser2011; Vrooman, Reference Vrooman2012). This theoretical prudence contrasts with some ambitious visions by the ‘expansionists’ group, who are more open or willing to apply the West-based theoretical frames away from the traditional home turf (for example, the University of Bath group, Hudson and Kühner, Reference Hudson and Kühner2009, Reference Hudson and Kühner2012). This theoretical tension has significant implications for the contemporary welfare modelling business as it also directly involves the key question: ‘are welfare regime concepts applicable outside of the original eighteen nations (or traditional Western welfare states)?’ To put it very crudely, the isolationists would say ‘no’ to the question, while the answer from the other side would be more positive. For example, the ‘radical isolationist’ position of Ferragina and Seeleib-Kaiser goes as far as dropping Japan out of the eighteen core group in their analysis as ‘any such analysis should include functional equivalents of public policy to make the comparison meaningful’ (2011: 594). Their viewpoints also correspond to some non-Western researchers who cast doubt on the applicability of the Western frameworks to politically and historically different regions (Kwon, Reference Kwon1997; Takegawa, Reference Takegawa2009). On the other side, as an example of typical expansionists, Abu Sharkh and Gough (2010: 28) argue that the welfare regime approach is ‘a fruitful paradigm’ for thinking about social policy across the developing world as the concepts place modern welfare states within a wider welfare mix and also incorporate welfare outcomes. The group's ambitions are directed at as many OECD countries as possible (Hudson and Kühner, Reference Hudson and Kühner2012) or ‘all the nation-states of the world’ (Abu Sharkh, Reference Abu Sharkh2009: 13).
It should also be noted that the two contrasting stances are not necessarily contradictory. While the isolationists are cautious of undue exportation of the concepts, the latter groups are more challenging in recasting the concepts and devising new variables to overcome the ‘export barriers’. The latter groups generally admit their studies are meaningful largely as a ‘first step’ (Hudson and Kühner, Reference Hudson and Kühner2012: 53), and their ‘validity, reliability and comparability are open to question’ (Wood and Gough, Reference Wood and Gough2006: 1703). Therefore, the key question is rephrased as not whether the welfare regime is exportable or not, but how to make the exportation justifiable and meaningful in the welfare modelling business.
The second reason behind the narrow focus is that a lack of available datasets has hindered expansion of the focuses. Esping-Andersen's (1990) immense influence is not only due to his introduction of three welfare regime concepts, but also to refined comparable datasets set up at the Swedish Institute for Social Research (www.sofi.su.se/english), with several studies using the same dataset (Ragin, Reference Ragin, Janoksi and Hicks1994; Shalev, Reference Shalev1996).
The more fundamental issue is the negligence of many studies in justifying their case selection. Just as Esping-Andersen (Reference Esping-Andersen1990) sets an undesirable precedent of unclear explanation for his choice of eighteen nations, some studies simply exclude without justification some key nations such as Japan and Canada (Leibfried, Reference Leibfried, Ferge and Kolberg1992), Japan and Switzerland (Saint-Arnaud and Bernard, Reference Saint-Arnaud and Bernard2003), and New Zealand (Castles and Obinger, Reference Castles and Obinger2008). Similarly, many articles state that their target countries are OECD members, but this can result in a different number of cases. For example, target groups include ‘eighteen OECD nations’ (Castles and Mitchell, Reference Castles and Mitchell1992), ‘23 OECD democracies’ (Siaroff, Reference Siaroff and Sainsbury1994), eighteen ‘old’ OECD nations’ (Goodin, Reference Goodin2001), ‘21 highly industrialized and democratic OECD nations’ (Obinger and Wagshall, Reference Obinger and Wagschal2001), to name but a few. By the year 2000, OECD membership was already thirty, but non-European members such as South Korea, Mexico, Turkey are constantly ignored (Ebbinghaus, Reference Ebbinghaus2012). Given this varying selection, the OECD title seems to serve only as a cover-up for a poor case selection process.
In the end, three suggestions can be made based on the findings. First, in methodological terms, relatively new datasets with a wider coverage of nations and indicators need to be utilised. For example, the Comparative Welfare State Data Set (Huber et al., Reference Huber, Ragin, Stephens, Brady and Beckfield2004), Comparative Family Policy Database (Gauthier, Reference Gauthier2010) and Comparative Welfare Entitlements Dataset (Scruggs et al., Reference Scruggs, Detlef and Kuitto2014) provide opportunities to explore the other regions. Relatively few studies have utilised the new statistics yet, with some exceptions (for example, Ferragina et al., Reference Ferragina, Seeleib-Kaiser and Tomlinson2012).
Second, in theoretical terms, if the international modelling business intends to go ‘global’, the trades need to be bilateral or multilateral, not unilateral. The business has been criticised for its unilateral Swedocentrism, Eurocentrism or ethnocentrism (Takegawa, Reference Takegawa2009; Walker and Wong, Reference Walker, Wong and Kennett2013) in applying its West-rooted concepts to ‘other worlds’. In other words, the business has been relatively active in ‘exporting’ its concepts out of the ‘core’ eighteen nations, but not as equally active in ‘importing’ concepts or variables from ‘other worlds’. To borrow one isolationist's comments, ‘a fair analysis would also have to include variables that are characteristic of these possibly distinct regimes’ (Vrooman, Reference Vrooman2012: 472). If we follow this logic, the current trade may have some components of ‘unfair’ trade. For example, the Anglophone debate tends to ignore studies in indigenous languages with only a few exceptions (for example, Powell and Kim, Reference Powell and Kim2014). Arguably, it might be a priority that new theories or concepts be developed to account for welfare of ‘other worlds’ in their own right, either free from the ‘Northern imposition’, or based on Western insights. It would facilitate genuine bilateral or multilateral, not unilateral, trades that in turn would pave ways for development of new theories and variables applicable to the wider cases.
Last and most importantly, to avoid the longstanding failure of justifying the case selection in the welfare modelling business, researchers need to explain their case selection process, justify their selection, clarify the characteristics and limitations of the selected cases, and specify how generalisable their conclusions may be beyond their cases.
Conclusion
Taking their lead from Esping-Andersen (Reference Esping-Andersen1990), few studies have discussed and justified their case selection. The subsequent debate has focused on debating the number of worlds while largely ignoring the number of cases, meaning that the crucial issue of case selection in the welfare modelling business has been long ignored. Both the original welfare modelling business within mainly ‘Western welfare states’, and its extension into ‘export markets’ of ‘other worlds’, need to address the important issue of case selection methodologically and theoretically. It remains an indispensable factor to the success of the business.
Appendix: Number of Cases in 33 Studies*
