¹ Esping-Andersen, Gosta, The Three Worlds of Welfare Capitalism (Princeton:Princeton University Press, 1990Google Scholar).

² There are several useful reviews of the “worlds of welfare” literature, including Arts, Wil and Gelissen, John, “Three Worlds of Welfare or More? A State of the Art Report,” Journal of European Social Policy 12, no. 2 (2002CrossRefGoogle Scholar); and Andersen, Esping, Social Foundations of Post-Industrial Economies (New York:Oxford University Press, 1999CrossRefGoogle Scholar), especially chap. 5. The number of “worlds” has been expanded to include other areas. For the Antipodes, see, for example, Castles, Frank and Mitchell, Deborah, “Identifying Welfare State Regimes: The Links between Politics, Instruments and Outcomes,” Governance 5, no. 1 (1992CrossRefGoogle Scholar). For Southern Europe, see, for example, Ferrara, Maurizio, “The ‘Southern Model’ of Welfare in Social Europe,” Journal of European Social Policy 7, no. 1 (1996Google Scholar); and Bonoli, Guiliano, “Classifying Welfare States: A Two Dimension Approach,” Journal of Social Policy 26, no. 3 (1997CrossRefGoogle Scholar). Other lines of criticism suggesting different regime arrangements concern the limited aspects of welfare state protection addressed in Three Worlds. The almost exclusive focus on a male breadwinner household model is criticized in, for example, Orloff, Ann, “Gender and the Social Rights of Citizenship,” American Sociological Review 58, no. 3 (1993Google Scholar); Sainsbury, Diane, ed., Gendering Welfare States (London: 1996CrossRefGoogle Scholar); and Alan Siaroff, “Work Welfare and Gender Equality: A New Typology,” also in Sainsbury. Other authors argue that the focus on major social insurance/assistance programs (health care, unemployment, sick pay, and public pensions) is too narrow for other reasons. See, for example, Room, Graham, “Commodification and Decommodification: A Developmental Critique,” Policy and Politics 28, no. 3 (2000CrossRefGoogle Scholar); Gal, J., “Decommodification and Beyond: A Comparative Analysis of Work-Injury Programmes,” Journal of European Social Policy 14, no. 1 (2004CrossRefGoogle Scholar); and Powell, Martin and Barrientos, Armando, “Welfare Regimes and the Welfare Mix,” European Journal of Political Research 43, no. 1 (2004CrossRefGoogle Scholar). Finally, several have criticized the technical approach to uncovering the three regime types. See Shalev, Michael, “Limits and Alternatives to Multiple Regression in Comparative Research,” Comparative Social Research (Greenwich, Conn.:JAI Press, 2007Google Scholar); and Hicks, Alexander and Kenworthy, Lane, “Varieties of Welfare Capitalism,” Socio-economic Review 1, no. 1 (2003CrossRefGoogle Scholar).

³ For a small selection of research that takes the Three Worlds typology as its starting point, see Mehrtens, F. John III, “Three Worlds of Public Opinion? Values, Variation, and the Effect on Social Policy,” International Journal of Public Opinion Research 16, no. 2 (2004CrossRefGoogle Scholar); Cathie Martin, Jo, “Reinventing Welfare Regimes: Employers and the Implementation of Active Social Policy,” World Politics 57 (October 2004CrossRefGoogle Scholar); Svallfors, Stefan, “Worlds of Welfare and Attitudes to Redistribution: A Comparison of Eight Western Nations,” European Sociological Review 13, no. 3 (1997CrossRefGoogle Scholar); Pierson, Paul, “The New Politics of the Welfare State,” World Politics 48 (January 1996CrossRefGoogle Scholar); Evelyne Huber and John D. Stephens,

⁴ Development and Crisis of the Welfare State: Parties and Policies in Global Markets (Chicago:University Chicago Press, 2001Google Scholar); Swank, Duane, Global Capital, Political Institutions, andPolicy Change in Developed Welfare States (Cambridge:Cambridge University Press, 2002CrossRefGoogle Scholar); Goodin, Robert E., Headey, Bruce, Muffels, Ruud, and Dirven, Henk-Jan, The Real Worlds of Welfare Capitalism (Cambridge:Cambridge University Press, 1999CrossRefGoogle Scholar); and Beller, Emily and Hout, Michael, “Welfare States and Social Mobility: How Educational and Social Policy May Affect Cross-national Differences in the Association between Occupational Origins and Destinations,” Research in Social Stratification and Mobility 24, no. 4 (2006CrossRefGoogle Scholar). “In addition to regarding welfare states as systems of social stratification, Esping-Andersen (in fn. 1 and in fn. 2, 74) and others such as Arts and Gelissen (fn. 2) have also emphasized how the degree to which national welfare regimes decommodify labor is essential to the characterization of welfare states.

⁵ See Scruggs, Lyle and Allan, James P., “Welfare State Decommodification in Eighteen OECD Countries: A Replication and Revision,” Journal of European Social Policy 16, no. 1 (2006CrossRefGoogle Scholar); and Bambra, Clare, “Decommodification and the Worlds of Welfare Revisited,” Journal of European Social Policy 16, no. 1 (2006CrossRefGoogle Scholar).

⁶ Esping-Andersen (fn. 1), 58.

⁷ Polanyi, Karl, The Great Transformation (1944; Boston:Beacon Press, 2001Google Scholar).

⁸ Esping-Andersen (fn. 1), 65.

⁹ Quoted in Heclo, Hugh and Madsen, Henrik, Policy and Politics in Sweden: Principled Pragmatism (Philadelphia:Temple University Press 1987), 157Google Scholar.

¹⁰ The original variables and discussion of the index are in Esping-Andersen (fn. 1), 69–78.

¹¹ CWED data are available from the authors and at http://sp.uconn.edu/~scruggs/wp.htm.

¹² Social Security Administration, Social Security Programs throughout the World (Washington, D.CGovernment Printing Office, 1981Google Scholar, 2004).

¹³ A score of 0 is given if there are two or fewer distinct schemes; a score of 2 is given if the number of schemes is greater than two and less than or equal to five; and finally, a score of 4 is given to countries with more than five distinctive pension schemes.

¹⁴ For the early 1980s, we used the country series in OECD, Social Expenditure Database (SOCX2004) (Paris:OECD, 2004Google Scholar) and the original source, International Labour Organization; for data we used Cost of Social Security (Geneva:ILO, 1983Google Scholar). Values for most countries are very similar across the two sources. We chose the OECD source for the United States in 1980 because the ILO series appears to underestimate state and local government employee pension spending. We used the ILO'S estimate for Italy in 1980, because the OECD figure excluded public sector pensions managed by the Treasury. We used the OECDs estimate for the Netherlands, because it excluded disability pensions, and these do not appear to be treated as retirement pensions elsewhere.

¹⁵ The largest difference occurs for Sweden. The ILO data show a large, sustained jump in the public service pension expenditures between the late 1970s and the early 1980s in Sweden, from about 1.2 percent to 2.4 perent of GDP. We were inclined to believe the higher figure, because, among other reasons, Sweden has a relatively large public sector workforce.

¹⁶ Where share of expenditures is 1 percent or less of GDP, the score is zero. Between 1 and 2.1 percent (inclusive), the score is 2, and the maximum score of 4 is assigned to expenditures greater than 2.1 percent.

¹⁷ The ILO survey changed in the early 1990s and no longer collected separate data for public employee pensions. For comparable recent information for circa 2000, we rely primarily on data in Robert Palacios and Edward Whitehouse, “Civil Service Pension Schemes around the World,” World Bank Social Protection Discussion Paper no. 602 (2006). We used the most recent ILO data (generally for 1989) for the countries not covered in their paper.

¹⁸ We count three programs based on the Social Security Administration's distinction between special programsforrailroad and federal employees and the many separate schemesforindividual state and local government employees. (This situation creates considerable fragmentation by subnational units if not distinct occupations, since most states have at least one public pension fund for their employees, with its own set of rules.)

¹⁹ An important general limitation of using public sector spending to gauge civil servants' privileged status is that the shares increase via a larger public sector workforce.

²⁰ In assigning stratification index scores, we used the original scoring metric in Three Worlds. Private spending shares up to 10 percent are scored 0; shares between 10 and 20 percent are scored 2; and private spending shares greater than 20 percent are scored 4. (NB: the cut point for obtaining a high score in Three Worlds is actually below the mean, with or without the United States included!)

²¹ Index values for means-tested programs are calculated following the general procedure adopted in Three Worlds. First, we dropped the extremely high values for New Zealand and Australia before computing the mean and standard deviation. Next, we used the standardized value cut points used in Three Worlds, −.6 and +.4, to create lower and upper bounds of 6.7 and 15.6, respectively.

²² The original results in Three Worlds strongly imply that non-earnings-related public pension supplements were not counted as means-tested spending in the Nordic countries and the United Kingdom but were so counted in the United States and Canada. It also appears that certain other near-cash benefits (for example, housing assistance and food programs) were excluded from the calculation in the former and included in the latter.

²³ See, for example, Castles and Mitchell (fn. 2).

²⁴ Esping-Andersen provides an extensive list of sources used in computing private pension payments in each country, but these are not easily available; Andersen, Esping “The Comparison of Policy Regimes: An Introduction,” in Rein, Martin, Andersen, Gosta Esping, and Rainwater, Lee, eds., Stagnation and Renewal in Social Policy (Armonk, N.Y.:M. E. Sharpe, 1987Google Scholar)). Pension income categories in the Luxembourg Income Study, in theory, allow us to compute the private sector share. However, our analysis of the LIS data reveal that income sources are not reliably divided among these categories, producing scores that are not credible in a number of cases.

²⁵ See, for example, Pierson, Paul, Dismantling the Welfare State? Reagan, Thatcher, and the Politics of Retrenchment (New York:Cambridge University Press, 1994CrossRefGoogle Scholar); Garrett, Geoffrey, Partisan Politics in the Global Economy (Cambridge:Cambridge University Press, 1998CrossRefGoogle Scholar); Swank, Duane, Global Capital, Political Institutions, and Policy Change in Developed Welfare States (Cambridge:Cambridge University Press, 2002CrossRefGoogle Scholar); Castles, Francis, The Future of the Welfare State: Crisis Myths and Crisis Realities (New York:Oxford University Press, 2004CrossRefGoogle Scholar); and Lindert, Peter, Growing Public: Social Spending and Economic Growth since the Eighteenth Century (New York:Cambridge University Press, 2004Google Scholar).

²⁶ This data set is not publicly available.

²⁷ It appears that estimates for these categories in Three Worlds are from Flora, Peter, State, Economy, and Society in Western Europe, 1815–1975: A Data Handbook (Frankfurt/Chicago:Campus Verlag/St. James Press, 1983CrossRefGoogle Scholar). Results for the United Kingdom in Three Worlds do not match coverage based on National Insurance contributions, our main source for coverage. Coverage figures for Italy, France, and Belgium are prone to some error, as they are not always directly reported. CWED uses national sources for Belgium. For France, CWED uses national sources for unemployment and the number of salarie and non-salane covered by assurance maladie in official social security statistics. This also produces coverage estimates that are higher than those found in Flora's study. CWED estimates may also be higher due to attempts to include those receiving pensions in another's name, such as survivors and those in receipt of couples supplements.

²⁸ In the former case the difference arises because we used the portion of those above retirement age (sixty-five) actually in receipt of a pension. In Canada CWED estimates for unemployment insurance are based on statistics from Canadian sources on employment insurance coverage.

²⁹ We constructed the stratification index scores for universalism using the standardized zxA points, as we did earlier for the means-tested benefits (fn. 19). The cut point for a low universalism score in the original is z=−.6, while the cut point for the high score is z=.7.

³⁰ A few states do, in fact, have sickness programs. These programs have a structure very similar to unemployment insurance, but since the total workforce in these states is a fraction of the national total, we treat the country as having no program.

³¹ The most dramatic changes between 1980 and 2002 are in Finland, where all three programs eliminated a maximum benefit ceiling, though its inegalitarian impact should not be overstated. The marginal replacement rate for higher earnings is small compared to the one for average earnings.

³² Index scores for equality were based on z-values used in Three Worlds. Scores below z=−.5 were scored 0, while those above 0.7 were scored 4.

³³ A factor analysis (nonrotated) with the published regime scores as variables suggests that of the two factors extracted, the liberal, socialist, and conservative indices load (respectively), +.55, −.62, .07 and −.34, −.23, +.60. That result is, in fact, consistent with Three Worlds, because the data cannot be reliably reduced to one or two dimensions, and the two extracted dimensions suggest a liberal-socialist distinction independent of conservatism and a conservative dimension that is (vaguely) negatively correlated with liberalism and socialism.

³⁴ We would go further to suggest that the high conservatism category include countries scoring a 6. This would be consistent with the treatment of scores on the socialist stratification index. Doing so would make Belgium a clear conservative welfare state in our replication. Japan would then score high on two dimensions: conservatism and liberalism. Such a change would make Esping-Andersen's original results appear slightly less clear, as Finland would then be high in conservatism and socialism. (The remarkable factor analysis scores are not affected by this decision.)

³⁵ The Eigenvalue for factor 1 is .96; for factor 2, it is .44. Eigenvalues for the original scores were .69 and .49, respectively. The nonrotated factor loadings for two dimensions are .36, .51, −.75 and .51, −.43, and .04, for socialism, liberalism, and conservatism, respectively.

³⁶ While not the expressed point of our analysis, there do appear to be some cases where the interpretation of measures was ill considered, for example, the possible inconsistent treatment of means-tested benefits in Nordic countries compared with the treatment in North America or the Antipodes. The implication is that the development of the indicators was biased toward finding (1) larger differences than otherwise existed and (2) bias that exaggerated the distinctiveness policies in the three regime types. In our earlier work on decommodification (Scruggs and Allan [fn. 5]), we also found evidence to indicate that some of the original coding decisions consistently favored certain countries.

³⁷ The impact on large-N cases may be obvious—for example, miscoding an explanatory variable biases regression results. But the same problem applies to small-N comparisons, at least insofar as their inferences about stratification effects are based on the misclassified cases.

³⁸ One alternative view of our results is that they are flawed and that the original results support the conventional categories. However, we follow procedures outlined in the creation of the original index. We also documented a number of errors that have exaggerated variation across conventional regimes and to minimize it within them. Our results may be an object lesson for the importance of independent replication.