¹ Esping-Andersen, Gøsta, The Three Worlds of Welfare Capitalism (Cambridge: Polity Press, 1990)Google Scholar

² Arts, Wil and Gelissen, John, ‘Three Worlds of Welfare Capitalism or More?’ Journal of European Social Policy, 12(2), 137–158CrossRefGoogle Scholar

³ Burgoon, Brian and Baxandall, Phineas, ‘Three Worlds of Working Time: The Partisan and Welfare Origins of Working Time in OECD Countries’, Politics and Society, 32 (2004), 439–473CrossRefGoogle Scholar

Arts, Wil and Gelissen, John, ‘Welfare States, Solidarity and Justice Principles: Does the Type Really Matter?’, Acta Sociologica, 44 (2001), 283–300CrossRefGoogle Scholar

Hicks, Alexander and Kenworthy, Lane, ‘Varieties of Welfare Capitalism’, Socio-Economic Review, 1 (2003), 29–61CrossRefGoogle Scholar

Noël, Alain and Thérien, Jean-Philippe, ‘From Domestic to International Justice: The Welfare State and Foreign Aid’, International Organization, 49 (1995), 523–553CrossRefGoogle Scholar

Thérien, Jean-Philippe and Noël, Alain, ‘Political Parties and Foreign Aid’, American Political Science Review, 94 (2000), 151–162CrossRefGoogle Scholar

⁴ Scruggs, Lyle A. and Allan, James P., ‘Social Stratification and Welfare Regimes for the 21st Century: Revisiting The Three Worlds of Welfare Capitalism’, World Politics, 60 (2008), 642–664CrossRefGoogle Scholar

⁵ Hicks and Kenworthy, ‘Varieties of Welfare Capitalism’.

⁶ Esping-Andersen, Gøsta, ‘Why No Socialism Anywhere? A reply to Alex Hicks and Lane Kenworthy’, Socio-Economic Review, 1 (2003), 63–72CrossRefGoogle Scholar

⁷ Noël and Thérien, ‘From Domestic to International Justice’.

⁸ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 70Google Scholar

⁹ Section 4 explains how these index scores were constructed.

¹⁰ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 78Google Scholar

¹¹ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 70Google Scholar

Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 75Google Scholar

¹² Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 72Google Scholar

¹³ It should be noted that Esping-Andersen does not employ the terminology of measurement models in The Three Worlds of Welfare Capitalism.

¹⁴ Esping-Andersen, The Three Worlds of Welfare Capitalism, Table 3.2, p. 72.

¹⁵ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 75Google Scholar

¹⁶ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 76Google Scholar

¹⁷ In the 1980 and 1996–2002 replication datasets of Scruggs and Allan, the correlations of universalism and benefit differentials are 0.19 and 0.06, respectively.

¹⁸ SEM analyses are to be preferred over the most widely used alternative approach to assessing the construct validity of measurement models: principal component analysis (PCA). PCA is a form of exploratory factor analysis, which does not test for goodness of fit. Therefore it provides no statistical criteria for judging how well a measurement model fits the data.

¹⁹ All variables in the data are measured at interval level, except for corporatism, which is a positive count variable. This is treated as a semi-interval level measure in the analysis. In small samples the results may sometimes be strongly affected by outliers. Yet, there are no variables in the available data with values beyond 3 standard deviations of the mean. These data thus meet the most important requirements to conduct SEM analyses.

²⁰ The RMSEA evaluates the discrepancy between the sample model to be estimated and the notional population model (under the null hypothesis of perfect fit) relative to the degrees of freedom specified for the test model. It is zero for a perfect fit.

²¹ Bollen, Kenneth A. and Long, J. Scott, eds, Testing Structural Equation Models (Newbury Park: Sage Publications, 1993), pp. 136–162Google Scholar

²² First, the regression weights that are necessary to fix the scale of the unobserved error terms of the seven attributes were constrained to one. To fix the scale of the three latent regime dimensions, the regression weights of liberalism on poor relief, of conservatism on etatism and of socialism on universalism were set at one. Estimation then showed that the model was not identified. Therefore, as suggested by the program output notes of AMOS, the three regime variances were constrained to one. As a result of these constraints, the model was identified. However, once estimated, it turned out to be inadmissible, because it included (slightly) negative estimates of the error variances of etatism and universalism. To remedy this defect, these error variances were constrained to zero. Re-estimation then produced an admissible model: the EA test model.

²³ First the model estimated regression weights of liberalism on poor relief, of conservatism on etatism and of socialism on universalism set at one. That model was inadmissible because the covariance matrix of latent variables was not positive definite. The three regime variances were then constrained to one, and the error variance of universalism to zero. This produced an admissible model: the EA test model (SA 1980). The choice of constraints for the EA test model (SA 1996–2001) is the same.

²⁴ One might suspect that the EA measurement model does not fit the three datasets, due to the large disagreement between the measurements of the means-tested poor relief indicator (see note twenty-five below). This possibility was checked by deleting the indicator from the measurement model (thus making liberalism depend positively only on private pensions and private health). However, none of the test models specified for this exercise fit the data at all well.

²⁵ Scruggs and Allan, ‘Social Stratification and Welfare Regimes for the 21st Century: Revisiting The Three Worlds of Welfare Capitalism’, p. 653–654Google Scholar

²⁶ See Esping-Andersen, The Three Worlds of Welfare Capitalism, pp. 27 and 69.

²⁷ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 60Google Scholar

Esping-Andersen, The Three Worlds of Welfare Capitalism, pp. 85–86Google Scholar

Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 75Google Scholar

²⁸ Esping-Andersen, The Three Worlds of Welfare Capitalism, pp. 68–9 and 75.

²⁹ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 68Google Scholar

³⁰ Esping-Andersen, The Three Worlds of Welfare Capitalism, pp. 26 and 68–9.

³¹ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 71CrossRefGoogle Scholar

³² Scruggs and Allan, ‘Social Stratification and Welfare Regimes for the 21st Century: Revisiting The Three Worlds of Welfare Capitalism’, p. 648Google Scholar

³³ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 76Google Scholar

³⁴ A problem with this criterion of typicality is that it allows countries to score strong on one regime and have medium scores on both others. For example, the Netherlands, with EA scores of 6 on socialism (strong), 6 on liberalism (medium) and 4 or conservatism (medium) has been repeatedly qualified as a hybrid case in the comparative welfare literature, yet it is regarded as a typical case of socialism based on Esping-Andersen's criterion. It should also be noted that Esping-Andersen qualified Finland as ‘typically’ socialist since he placed that country's conservative EA score of 6 in the medium rather than in the strong category of the conservative regime. This is inconsistent with having Finland in the strong category of socialism, however, since its EA score on this regime is also 6. In Table 1, this inconsistency has been corrected by qualifying Finland as a hybrid case.

³⁵ Sharma, Subhash, Applied Multivariate Techniques (Danvers/Hoboken: John Wiley & Sons, Inc., 1996), p. 118Google Scholar

³⁶ Esping-Andersen, The Three Worlds of Welfare Capitalism, p. 71CrossRefGoogle Scholar

³⁷ Austria's comparatively low ranking of 4 on liberalism in Table 1 is explained by the fact that the EA score of liberalism incorporates a low value of 2.8 on the indicator of means-tested poor-relief, well below the mean of 5.9. Since this indicator is not included in the liberal social insurance dimension, Austria's corresponding factor regression score ends up rather higher in Table 2 at 0.64, just above the borderline of strong and medium.

³⁸ Our social insurance analysis of countries based on Scruggs and Allan's data differs somewhat from their analysis, which follows Esping-Andersen's ranking methods and is based on recalculations of the EA regime scores. Scruggs and Allan's 1980 replication typology diverges far more radically from the original than this study's social insurance typology does (see Scruggs and Allan, ‘Social Stratification and Welfare Regimes for the 21st Century: Revisiting The Three Worlds of Welfare Capitalism’, Table 5 on p. 652). However these results are only of limited interest, because the EA scores are invalid.

³⁹ Noël and Thérien, ‘From Domestic to International Justice’.

⁴⁰ Noël and Thérien, ‘From Domestic to International Justice’, p. 540Google Scholar

⁴¹ Noël and Thérien, ‘From Domestic to International Justice’, p. 539Google Scholar

⁴² Noël and Thérien, ‘From Domestic to International Justice’, p. 549Google Scholar

⁴³ Noël and Thérien, ‘From Domestic to International Justice’, p. 548Google Scholar

⁴⁴ Noël, Alain and Thérien, Jean-Philippe, ‘Public Opinion and Global Justice’, Comparative Political Studies, 35 (2002), 631–656CrossRefGoogle Scholar

⁴⁵ Scruggs and Allan, ‘Social Stratification and Welfare Regimes for the 21st Century: Revisiting The Three Worlds of Welfare Capitalism’.

⁴⁶ Shalev, Michael, ‘Limits and Alternatives to Multiple Regression in Comparative Research’, Comparative Social Research, 24 (2007), 261–308CrossRefGoogle Scholar