How does economic globalisation affect the welfare state? Focusing on the mediating effect of welfare regimes

Transcription

1 DOI: /j x Int J Soc Welfare 2009: 18: How does economic globalisation Blackwell Oxford, IJSW International XXX Original Kim 2008 & does Zurlo The UK Articles economic Publishing Author(s), Journal globalisation of Ltd Journal Social Welfare compilation affect the welfare Blackwell state? Publishing Ltd and the International Journal of Social Welfare affect the welfare state? Focusing on the mediating effect of welfare regimes INTERNATIONAL JOURNAL OF SOCIAL WELFARE ISSN Kim TK, Zurlo K. How does economic globalisation affect the welfare state? Focusing on the mediating effect of welfare regimes Int J Soc Welfare 2009: 18: The Authors, Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare. In recent years the impact of globalisation on the welfare state has become a major issue in comparative policy studies. Some empirical studies demonstrate a negative relationship between globalisation and the welfare state, while others show adverse findings or a non-significant relationship. The impact of globalisation, however, can be neither uniform nor unidirectional because of the differences in the political economies of individual welfare states. Welfare regimes reflect qualitative differences in arrangements of welfare institutions and the associated enduring configuration of the welfare nexus, suggesting that welfare regimes may influence the impact of globalisation on the welfare state. We scrutinise the relationship between globalisation and the welfare state by sampling 18 affluent countries from 1980 to 2001 and concentrating on the mediating effect of three welfare regime types. Our study provides a comprehensive examination of the relationship between globalisation and the welfare state using a state-ofthe-art analytical technique the mixed-effect model. Findings suggest that welfare regimes respond differently to the impact of globalisation and therefore mediate the relationship between globalisation and the welfare state. Globalisation negatively affects the welfare state in a social democratic regime, while it marginally affects the welfare state in liberal and conservative regimes. Tae Kuen Kim, Karen Zurlo School of Social Policy and Practice, University of Pennsylvania, USA Key words: globalisation, welfare state, welfare regime, mediating effect, mixed-effect model Tae Kuen Kim, School of Social Policy and Practice, University of Pennsylvania, 3815 Walnut Street, Philadelphia, PA 19104, USA kimtk@sp2.upenn.edu Accepted for publication December 14, 2007 Introduction In the last quarter of the 20th century, welfare states, facing the new global economic environment, have undergone a profound transformation. One of the most controversial issues of this restructuring process concerns the impact of economic globalisation on the welfare state. How is the welfare state influenced by the wave of economic globalisation (Glatzer & Rueschemeyer, 2005)? Although a consensus is emerging that economic globalisation has some type of relationship with the welfare state, the exact nature of the relationship remains unclear (Brady, Beckfield & Seeleib-Kaiser, 2005: 921). Prior research (Hicks & Zorn, 2005; Kim, 2007; Wibbels, 2006) that empirically examined the dynamic between economic globalisation and the welfare state focused on the holistic relationship and ignored the divergent effects of economic globalisation. The impact of economic globalisation on the welfare state, however, can be neither uniform nor unidirectional because of the differences in the political economies of individual welfare states (Yang, 2000). For instance, Rudra (2002) demonstrated that economic globalisation is positively associated with welfare spending in developed countries, whereas it is negatively related to welfare spending in less-developed countries. Furthermore, analysing 57 less-developed countries, Rudra and Haggard (2005) showed that welfare spending in politically authoritarian countries is more sensitive to the pressures of economic globalisation than it is in Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare. 130 Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

2 How does economic globalisation affect the welfare state? democratic countries. These results imply that responses to the impact of economic globalisation can vary according to the different welfare schemes and socioeconomic situations among developed countries. In the present study, we scrutinise the relationship between economic globalisation and the welfare state among developed countries, concentrating on the mediating effect of welfare regimes. Welfare regimes reflect qualitative differences in the arrangements of welfare functions between major social institutions including market, family and state (Esping-Andersen, 1990). As a result, welfare regimes prescribe the functional mechanism of the welfare state and therefore may significantly influence the relationship between economic globalisation and the welfare state. While there are several typologies of welfare regimes, the three-fold typology of welfare regimes suggested by Esping- Andersen has dominated the debate on comparative social policy over the last two decades, especially among developed countries (Scruggs & Allan, 2006). Despite the critique of the model, this model provides a systematic empirical validation for a set of ideal types (Scruggs & Allan, 2006). Moreover, as Esping-Andersen s welfare-regime model is based on the broad approach to the welfare state reflected in political economy (Powell & Barrientos, 2004), it provides a comprehensive framework for identifying the unique features of the welfare nexus in given countries. Thus, by employing Esping-Andersen s welfare regime model, we test two main hypotheses: Hypothesis 1: The degree of economic globalisation will affect the size of the welfare state. Hypothesis 2: The impact of globalisation on the welfare state will be mediated by welfare regime type. Globalisation and the welfare state Globalisation encompasses various aspects including global communication networks, an international flow of knowledge and information and global civil societies. The most influential aspect of globalisation is the interdependence of market systems around the world (Stiglitz, 2006). The intertwining of national economies and the dependency of government finance on global markets has shifted the structure of political economic institutions of the nation-state (Castells, 2004). Thus, in the present study we restrict our focus to economic globalisation, which entails the economic integration of the countries of the world through the increased flow of goods, services and capital (Stiglitz, 2006). We conceptualise economic globalisation (hereafter globalisation) as the intensification of international economic exchange and international economic integration, following the operational definition of Brady et al. (2005). Previous arguments regarding the impact of globalisation on the welfare state generally have come to one of three conclusions. A negative perspective contends that globalisation undermines existing welfare states and impedes new developments in social welfare policy, resulting in the era of welfare state crisis (Hardt & Negri, 2001; Huber & Stephens, 2001; Schwartz, 2001). In contrast, a positive perspective claims that globalisation and the welfare state can be compatible and mutually reinforcing. Thus, a positive perspective suggests that globalisation eventually expands the welfare state (Garrett, 1998; Rieger & Leibfried, 2003; Rodrik, 1997). Criticising the current debate on the impact of globalisation as an exaggerated one, a sceptical perspective argues that globalisation per se is an insignificant or marginal influence on the welfare state (Atkinson, 2002; Kittel & Winner, 2005; Wade, 1996). A negative perspective, called the efficiency hypothesis (Blackmon, 2006), argues that globalisation is the cause of welfare state decline. The basic premise of the efficiency hypothesis is that the instrumental capacity of the nation-state is decisively undermined by the globalisation of core economic activities (Castells, 2004). According to Brady et al. (2005), the globalisation of production and investment necessitates welfare state retrenchment, since states lose autonomy over social welfare policies in the face of the overwhelming global economy. Moreover, under the neo-liberal economic paradigm of globalisation, big government is regarded as inefficient in the globalised economic environment, since high levels of taxes and social spending impede developing an internationally competitive economy (Kim, 2007). Therefore, as national economic schemes open to the international market, governments are forced to adapt to the imperatives of global competition with an austere fiscal policy, which induces the cutting of cost-intensive social welfare programmes (Genschel, 2004). Several studies link globalisation and welfare state reduction. For example, Garrett and Mitchell (2001) and Burgoon (2001) demonstrated that openness in economic exchange is associated with less governmental social spending. On the other hand, a positive perspective, called the compensation hypothesis (Blackmon, 2006), claims that the emerging internationalisation of economies is related to a high demand for social security, which in turn facilitates an upward shift of taxation and social spending levels (Kittel & Winner, 2005). According to Kim (2007: 182), Economic integration exposes national economies to the turbulences in the world economy, generating more uncertainty and volatility in the domestic economy. Faced with greater economic insecurity, the state would be pressed to provide more social protection and insurance. In short, economic globalisation creates more demand for compensatory Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare 131

3 Kim & Zurlo government spending that mitigates economic and social dislocations due to increasing openness. As a result, the more countries experience volatility and uncertainty from fluctuations in international finance and trade, the more governments develop social welfare policies that stabilise the economic security and politically appeal to their citizens (Brady et al., 2005). Consistent with this perspective, several empirical studies have found positive relationships between globalisation and the welfare state. Hicks (1999) and Rodrik (1997) demonstrated that economic openness significantly increases social expenditures. The sceptical perspective contends that there is neither theoretical reason nor empirical evidence to believe that domestic social policy autonomy is influenced by increasing globalisation (Genschel, 2004). Scholars supporting this perspective emphasise that domestic factors, such as societal and political institutions, as well as historical legacy, minimise the globalisation effect in the social policy formation procedure (Berger, 2000). Advocates of the sceptical perspective argue that the politics of a given country, rather than external forces such as globalisation, is the determinant that governs a social policy (Pierson, 2001). Furthermore, it is argued that in a mature welfare state social welfare policies form strong developmental trajectories that are inherently difficult to reverse through constitutional structure and the popularity of welfare programmes (Hacker, 2002). Thus, the sceptical perspective argues that welfare states are able to exert resilience on the impact of globalisation and therefore bolster welfare state stability regardless of globalisation (Brady et al., 2005). With this perspective, several empirical studies (Atkinson, 2002; Brady et al., 2005) have found that globalisation does not significantly affect the welfare state nor does it marginally influence the welfare state. Limitations of prior research In the present study, we analyse 22-year panel data of 18 affluent countries from 1980 to Our analysis is distinct from previous empirical research in three important ways. First, while most previous research has examined the period of welfare state development from 1960 to the early to mid 1990s, there are only a few studies investigating the later 1990s (Allan & Scruggs, 2004; Brady et al., 2005). Since the period from the mid to late 1990s marks the mature era of economic globalisation (Castells, 2004; Stiglitz, 2006), excluding this period may erroneously estimate the impact of economic globalisation. By including the late 1990s, our analysis demonstrates the more exact relationship between economic globalisation and the welfare state. Second, prior studies have operationalised economic globalisation in a narrow sense (Brady et al., 2005). The concept of economic globalisation can be sufficiently assessed with variables that include economic performance as well as economic regulation (Stiglitz, 2006). We measured the multidimensional features of economic globalisation with five variables: foreign direct investment (FDI), trade share, the liberalisation of capital transaction, the liberalisation of currency transaction and the openness of economic regulation. Thus, our analysis comprehensively contains the multiple facets of economic globalisation. Lastly, the most important limitation of past studies involves statistical techniques, i.e. the application of panel data to comparative analyses. Cross-country panel data require the use of sophisticated statistical tools. While conventional statistical methods assume that observations are independent, cross-country panel data clearly violate the independency assumption (Luke, 2003; Raudenbush & Bryk, 2002). To overcome this serial correlation, most longitudinal comparative analyses on the relationship between economic globalisation and the welfare state (Brady et al., 2005; Garrett & Mitchell, 2001; Rudra & Haggard, 2005) depend on pooled time series analyses or autoregressive models with panelcorrected standard errors. However, these methods are less suited when manipulating time-constant covariates and analysing long-term effects (for a detailed explanation, see Kittel & Winner, 2005: ). Therefore, we employed a mixed-effect model to adjust for cross-sectional and time-specific idiosyncrasies in cross-country panel data. In this sense, our study provides a robust statistical methodology that is applicable to other types of comparative analyses conducted by social welfare policy researchers. Data and measures To investigate the impact of globalisation on the welfare state, we used data from 18 developed countries collected between 1980 and 2001, inclusive of Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, The Netherlands, New Zealand, Norway, Sweden, Switzerland, the UK, and the USA. These 18 affluent countries were selected based on Esping-Andersen s (1990) original welfare regime model. We chose this period for analysis because the most relevant variables, especially social expenditure and globalisation indicators, are not available for the selected countries before In addition, since this period overlaps in an analytical time-frame with Esping-Andersen s model, the period is theoretically suitable to examine the mediating effect of the welfare regimes. Considering that contemporary globalisation germinated in the mid 1970s and grew from the early 1980s (Castells, 2004; Guillen, 2001; Stiglitz, 2006), our analytical period may provide a broad framework for understanding the relationship between globalisation and the welfare state. 132 Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare

4 How does economic globalisation affect the welfare state? The dataset used in this study was created through the integration of two data sources: OECD Factbook (Organisation for Economic Cooperation and Development, 2007b), and the Comparative Welfare States Data Set of Luxembourg Income Study (Huber et al., 2004). For six countries, Australia, Canada, Japan, New Zealand, Norway, and the USA, we have complete 22-year panel data for all covariates. For 12 countries, however, we excluded 52 cases from the analysis due to missing data for several of the variables. As a result, our final sample includes 344 individual cases stemming from 18 countries. We used the ratio of social expenditure to the total gross domestic product (GDP) as the proxy of the welfare state. The concept of the welfare state generally refers to the use of governmental power to protect people from income losses inherent in an industrial society and to provide for a minimum standard of economic wellbeing for all citizens (Peterson, 1985: 602). Social expenditure as a percentage of GDP is a measure of the extent to which governments assume responsibility for supporting the standard of living of disadvantaged or vulnerable groups (Organisation for Economic Cooperation and Development, 2007b). In our dataset, social expenditure includes not only cash benefits and direct in-kind provision of goods and services, but also tax breaks for social welfare purposes. Thus, this indicator comprehensively assesses the welfare effort of a given country and represents the quantitative aspect of the welfare state. In addition, as the social expenditure variable of this study includes tax expenditure, or indirect social spending, this indicator estimates a more exact accounting of the welfare states, avoiding the hidden welfare state controversy (see Howard, 1997). While a welfare state analysis based on social expenditure change has contextual limitations (see Korpi, 2003: 593), as Swank (2001) argued, we have counteracted these limitations with proper control variables in our analysis. Our social expenditure variable does not include private spending. Nevertheless, this limitation is unlikely to render invalid results since private social spending is small in comparison with public spending among developed welfare states (Adema, 2001). Moreover, in order to examine multi-dimensional aspects of the welfare state, we disaggregated social expenditure into three sub-categories active and passive labour market spending and social service spending and used these categories as dependent variables, respectively. Active labour market spending includes social expenditure aimed at the improvement of the beneficiaries prospect of finding gainful employment or to increase their earnings capacity. This category contains spending on public employment services and administration, job training and labour market programmes to promote employment for the unemployed (Organisation for Economic Cooperation and Development, 2007a: 14). Passive labour market spending includes all cash transfers to people that compensate for unemployment. This category contains redundancy payments out of public resources as well as pensions to beneficiaries before they reach the standard pensionable age if these payments are made because they are out of work (Organisation for Economic Cooperation and Development, 2007a: 15). Finally, social service spending includes all other social expenditure (in-cash and in-kind benefits), except for spending regarding healthcare, labour market programmes, disability-relative benefit, housing policy and old age pension (Organisation for Economic Cooperation and Development, 2007a: 15). While the measures of globalisation have not yet converged on a single indicator (Held et al., 1999), globalisation can be measured by trade, direct investment and the characteristics of legal institutions regarding international exchange (Quinn, 1997). Conceptualising globalisation as the intensification of international economic exchange and integration, we focused on two dimensions of globalisation: investment and trade openness (ITO) and economic system openness (ESO). While the former represents the degree of international economic exchange and integration in terms of actual economic performance, the latter represents it in terms of economic regulation. To calculate investment and trade openness (ITO), we summed inward and outward FDI as a percentage of GDP and the share of trade in GDP. FDI is a key element in the rapidly evolving process of international economic exchange and integration. FDI not only creates direct, stable and long-lasting connections between economies, but also encourages the transfer of technology and know-how between countries (Organisation for Economic Cooperation and Development, 2007b). In addition to FDI, international trade in goods and services is a principal channel of economic exchange and integration (Organisation for Economic Cooperation and Development, 2007b). Hence, these two indicators comprehensively represent the intensification of economic exchange and integration in terms of actual economic performance. We measured FDI with investment as a percentage of GDP by a resident entity in one economy with the objective of obtaining a lasting interest in an enterprise resident in another economy. For trade share, we used the average of imports and exports (of both goods and services) at current prices as a percentage of GDP. To estimate economic system openness (ESO), we used three indicators: the liberalisation of capital, the liberalisation of currency account transactions, and the accession to international legal agreements. An important facet of globalisation is the level of exchange restriction regulating international financial transaction (Castells, 2004). We used openness scores developed by Quinn (1997). The liberalisation of capital and currency Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare 133

5 Kim & Zurlo Table 1. Summary of the main variables. Liberal Conservative Social-democratic Dependent variable Social expenditure as percentage of GDP 17.6 (3.1) 21.9 (5.7) 27.3 (3.2) Globalisation indicators Investment trade openness (ITO) 32.5 (18.4) 26.3 (9.9) 48.0 (16.3) Economic system openness (ESO) 12.3 (1.9) 12.2 (1.4) 13.1 (1.7) Control variables GDP per capita 16,362 (6,102) 17,940 (5,704) 17,875 (5,634) Unemployment rate 8.4 (3.1) 7.1 (3.6) 5.8 (2.5) Elderly population rate 11.9 (1.8) 13.9 (1.7) 15.1 (1.5) Left cabinet 8.1 (7.3) 9.5 (6.1) 24.7 (10.2) Right cabinet 22.7 (9.1) 14.7 (15.9) 4.6 (3.1) Note: Numbers in parentheses represent the standard deviation. GDP, gross domestic product. account transactions refers to the degree of openness in capital and currency account transactions, respectively. Capital liberalisation scores range from zero to four and currency liberalisation scores range from zero to eight (for detailed information, see Quinn, 1997: 535). The accession to international legal agreements refers to the degree of a nation s ability to restrict financial flow. This indicator ranges from zero to two (for detailed information see Quinn, 1997). By integrating these three indicators, we were able to measure the intensification of economic system openness in a given country. Hence, our economic system openness index is scored on a scale of For the typology of welfare regimes, we employed Esping-Andersen s (1990) model which identifies three welfare regime types: liberal, conservative and social democratic. The concept of regime refers to an enduring configuration of institution and policies that are closely interconnected, evolving over long periods of time (Hacker, 2002: 11). Hence, we assumed that the welfare regime of a given country to be the product of path dependence (for a detailed argument, see Hacker, 2002) and treated it as a time-constant variable. We controlled other covariates such as the unemployment rate, GDP per capita, the rate of the aging population (over 65) and cumulative left and right cabinet scores. 1 We included these variables, following the contemporary analytical model of welfare states found in most influential studies (Brady et al., 2005; Hicks, 1999; Huber & Stephens, 2001). Prior studies contain other variables such as military spending and inflation rates. These variables, however, do not significantly affect the level of social spending. In order to construct a parsimonious model, we selected five variables that significantly influence social expenditures. Table 1 summarises the main variables. 1 Cumulative left and right cabinet scores tabulate left and right seats as a proportion of seats held by all government parties in each individual year and then sums these percentages since 1946 (Brady et al., 2005: 927). Methodology and analytical frame Most cross-country panel data have multi-level structures where repeated measures (e.g. each year s social expenditure) are nested within a given country. Although our dataset has a total of 344 individual cases (individuallevel), these cases are clustered into 18 countries (countrylevel). Individual cases from the same country tend to be more similar due to their closeness in space and/or time. For example, this year s social spending of a given country may be more closely related to last year s social spending of the country rather than to other countries. This interdependency among individual cases is called intra-class correlation, or group homogeneity, and clearly violates independent assumption (Kreft & de Leeuw, 2004). The multi-level structure of cross-country panel data can be efficiently analysed with the mixed-effect model. In our dataset, each country has a maximum of 22 cases, representing 22 years of annual social spending. The change in annual rates of social spending for a given country can be represented through a two-level structure. At level 1, the annual rate of social spending of each country is represented by a 22-year trajectory. At level 2, the trajectories of each country may depend on country-level characteristics (Raudenbush & Bryk, 2002). An analysis of panel data using the mixed-effect model has several advantages compared with pooled-time estimation or other multivariate statistics. First, the time points at which measurements are obtained need not be constant for all subjects. Second, the subject with incomplete observations can be included in the analysis without biased estimation. There are important considerations in many cross-country panel datasets, since some countries have missing variables for several years. 2 Third, researchers can efficiently specify the interactions of 2 Our dataset contains complete information for six countries, but 52 cases from 12 countries are missing. For example, while we have 22-year observations for the USA, we have only 10-year observations for Switzerland. These unbalanced observations may produce biased estimations when using conventional time-corrected methods. 134 Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare

6 How does economic globalisation affect the welfare state? interest (Norusis, 2004: 250). Equation (1) shows the basic concept of the mixed-effect model of our analysis. Level-1: SE tj = β 0j + β 1j ITO t 1 + β 2j ESO t 1 + β ij CONVAR t 1 + ε tj Level-2: β 0j = γ 00 + (γ 01 CON + γ 02 SOC) + μ 0j β 1j = γ 10 + (γ 11 CON + γ 12 SOC) + μ 1j β 2j = γ 20 + (γ 21 CON + γ 22 SOC) + μ 2j (1) Whereas SE tj refers to social expenditure at year t of country j. β 0j is the mean social expenditure of country j. ITO t 1 and ESO t 1 refer to investment and trade openness and economic system openness, respectively, at year t 1. ΣCONVAR t 1 refers to the vector of control variables at year t 1. ε tj is the variance of social expenditure at year t of country j, resulting from the unique effect of a given year. Following the convention of longitudinal comparative studies (Brady et al., 2005; Rudra, 2002; Rudra & Haggard, 2005), we lag all covariates by one year. Intercept (β 0j ) and slopes for globalisation indicators (β 1j and β 2j ) of country j become the outcome variables at level 2, where they may depend on country-level characteristic and type of welfare regime. Since we use the liberal regime-type as a reference category, we include two dummy variables (CON and SOC) to represent conservative and social democratic regimes. Instead of treating the intercept and slopes as fixed constants, the mixed-effect model assumes that they are random variables with a specified probability distribution (Allison, 2003; Norusis, 2004). By doing so, the mixed-effect model adjusts for intraclass correlation. Equation (2) shows the final model specification of our analysis. SE tj = γ 00 + γ 01 CON + γ 02 SOC + γ 10 ITO t 1 + γ 11 CON ITO t 1 + γ 12 SOC ITO t 1 + γ 20 ESO t 1 + γ 21 CON ESO t 1 + γ 22 SOC ESO t 1 + β ij CONVAR t 1 + (ε tj + μ 0j + μ 1j ITO t 1 + μ 2j ESO t 1 ) (2) In order to merge two levels into a single equation, we substituted β 0j, β 1j, and β 2j of level 1 with level 2. Equation (2) specifies that social expenditure at year t of country j is represented by the combination of the grand mean social expenditure among all the countries (γ 00 ), fixed-effect (from γ 01 CON to Σβ ij CONVAR t 1 ), and random-effect between and within countries (in parentheses). Furthermore, this equation contains interaction terms defined by globalisation indicators and welfare-regime type (γ 11, γ 12, γ 21, and γ 22 ), called cross-level interactions. By testing statistical significance of the cross-level interaction terms, we examined the mediating effect of welfare regimes on the relationship between globalisation and the welfare state. Figure 1. Net trends in social expenditure as a per cent of GDP by welfare regime type. Results Following the recommendation of Beck and Katz (1995), we initially assessed the degree of correlation of the error terms before applying the correction, by checking intra-class correlation (ICC). We tested ICC of the dependent variable, social expenditure, with the unconditional random-effect model, called the null model, where all independent variables are excluded from the equation. Results reveal that ICC of this dataset equals 0.83 (within-country variance = 4.5 and between-country variance = 27.6), 3 indicating high serial correlation. Generally, ICC greater than 0.25 implies that researchers should adjust for the correlation for the unbiased estimation (Kreft & de Leeuw, 2004). Hence, we apply the mixed-effect model to the analysis. Before discussing the relationship between globalisation and the welfare state, we present the net trends in social expenditure and globalisation indicators by welfare regime type. Figure 1 shows the average social expenditure among three welfare regimes from 1980 to During the entire period, social democratic regimes had greater levels of social expenditure than did conservative and liberal regimes. We found, however, different patterns before and after Before 1992, the social expenditures of conservative regimes were closer to those of the liberal regimes. While the social expenditures of the conservative regimes continued to increase, social expenditures of the liberal regimes declined after a peak 19.6 per cent in In this period, social expenditures of the social democratic regimes also decreased after a peak of 29.7 per cent in As a result, conservative regimes converged on 3 ICC is calculated by the proportion of the between-group variance to the sum of the between-group and the withingroup variance in the dependent variable (Raudenbush & Bryk, 2002). Hence, in our analysis, ICC is equal to 27.6/ ( ). Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare 135

7 Kim & Zurlo Table 2. Effect of welfare regime type on social expenditure, investment and trade openness (ITO) and economic system openness (ESO). Dependent variable Social expenditure ITO ESO β(t) β(t) β(t) Fixed-effects Conservative 5.33 (2.6)** 6.01 ( 0.7) 0.17 (0.2) Social democratic 9.58 (4.6)*** (2.0)* 0.14 (0.2) Intercept (13.3) (4.3) (16.4) Random-effects Within-country 4.56 (12.5)*** (12.6)*** 1.77 (12.6)*** Between-country (2.9)*** (2.9)*** 0.99 (2.7)** 2 log likelihood 1, , , * p < 0.05; ** p < 0.01; *** p < Figure 2. Trends in investment and trade openness (ITO) and economic system openness (ESO) by welfare-regime type. Note: liberal = liberal regime, con = conservative regime, soc = social-democratic regime. social democratic regimes at the end of the period (24.7 versus 25.8 per cent), while liberal regimes lagged behind (17.4 per cent). Displaying the trends in the globalisation indicators is also instructive. Figure 2 shows trends of ITO and ESO by welfare regime type. ITO dramatically increased after 1994 for all three regimes. Also, the patterns of upward trends in ITO were similar for all three welfare regimes. These increases were most salient, however, for the social democratic regimes after Despite fluctuations in the liberal and conservative regimes, the long-term stability of ESO was secular, compared with the trends in ITO. During the entire period, the three welfare regimes maintained a relatively high level of ESO, ranging from 10.3 to Considering that the maximum score of ESO is 14, the figure exhibits a considerably flexible economic exchange regulation among all three regime types. To sum up, all three regimes displayed different patterns of social spending from 1980 to In addition, while all three regimes experienced similar levels of ESO, the social democratic regimes experienced greater ITO than did the other two. Based on these descriptive results, we examined the linear effect of welfare regime types on the main variables: social expenditure, ITO and ESO. Table 2 shows the results of the mixed-effect model. Adjusting for within-country and between-country variances, there were significant differences in social spending among the three regimes. The social democratic and conservative regimes had greater social expenditures by an average of 9.6 per cent and 5.3 per cent, respectively, than did the liberal regimes. Moreover, compared with the null-model, between-country variance shrunk from 27.7 to 12.3, indicating that welfare regime types solely account for about 55 per cent of the variance in social expenditure between countries. 4 The mixed-effect model analysis also demonstrates that ITO in the social democratic regimes was significantly higher than in the liberal and conservative regimes. On the other hand, there were no significant differences in ESO. This result reveals that the social democratic regimes experienced relatively higher levels of globalisation than did the liberal and conservative regimes, particularly with respect to ITO, or actual economic performance. While this baseline analysis shows that welfare regime types are associated with social expenditure, it does not 4 Between-country variance in the null model refers to the total variance in the dependent variable (in our analysis, social expenditure) among countries. Thus, the proportion of reduced between-country variance by adding covariates means that the added covariates account for the proportion of the total variance in the dependent variable. In our analysis, between-country variance of the null model is 27.7 and it decreases to 12.3 by adding welfare regimes in the model. In other words, welfare regimes reduce the variance in social expenditure by 15.4 (= ) among countries. Therefore, welfare regimes explain about 55 per cent (= 15.4/27.7) of the total variance in social expenditure among the countries analysed (for more detailed information, see Luke, 2003). 136 Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare

8 How does economic globalisation affect the welfare state? Table 3. Effect of globalisation on social expenditure: main-effect model and interaction-effect model. Main-effect model β(t) Interaction-effect model β(t) Fixed-effects Covariates GDP per capita ( 10 2 ) 0.02 (4.57)*** 0.02 (4.53)*** Unemployment rate 0.40 (8.37)*** 0.44 (8.89)*** Elderly population rate 0.18 (1.21) 0.08 (0.52) Left cabinet 0.29 (0.55) 0.07 (1.15) Right cabinet 0.04 ( 0.93) 0.03 ( 0.69) Globalisation indicators ITO 0.12 ( 7.20)*** 0.06 ( 1.39) ESO 0.02 (0.27) 0.01 ( 0.15) Welfare regimes Conservative 3.96 (1.83)* 4.26 (1.63) Social-democratic (4.15)*** (5.35)*** Cross-level interaction Conservative ITO 0.01 (0.02) Social-democratic ITO 0.14 ( 2.51)** Intercept (7.41) (7.83) Random-effects Within-country 2.60 (12.47)*** 2.27 (11.79)*** Between-country (2.74)** (2.32)** 2 log likelihood 1, , GDP, gross domestic product; ITO, investment and trade openness; ESO, economic system openness. * p < 0.05; ** p < 0.01; *** p < provide the mediating effect of welfare regimes on the relationship between globalisation indicators and social expenditure. Why did each welfare regime type show different patterns and trends in social expenditure when they experienced similar or different globalisation phenomena? To answer this research question, we first analysed the main-effect model where globalisation indicators, welfare regime type and other covariates are included as independent variables. The second column in Table 3 shows the result of the main-effect model. 5 Controlling for GDP per capita, unemployment rate, elderly population rate and political configuration, social expenditure is negatively affected by ITO. A 1 per cent increase in ITO reduced social expenditure by 0.12 per cent. On the other hand, ESO did not significantly affect social expenditure. GDP per capita and the unemployment rates were positively related to social expenditure. A US$100 increase in GDP per capita and 1 per cent increase in unemployment rate increased social expenditure by 0.02 per cent and 0.4 per cent, respectively. To examine how welfare regimes mediate the relationship between globalisation and social expenditure, we constructed an interactioneffect model that includes cross-level interaction terms between ITO and welfare regimes in the analysis. We 5 In order to check any possible multicorrelinearity among the entire set of variables, we test the variance inflation factor (VIF) score. The result confirms that there is no substantial correlation among them. excluded ESO and welfare regimes interaction terms because ESO did not significantly affect social expenditure in the main-effect model. The third column in Table 3 shows the results of the interaction-effect model. We initially tested whether the cross-level interaction term was statistically significant through the deviance chi-square test. The difference in 2 log likelihood between the main-effect model and the interaction-effect model was 22.7 with 2 degrees of freedom. Thus, we conclude that the cross-level interaction term is statistically significant. Controlling for other variables, all three regimes were negatively influenced by ITO. The effect sizes of ITO on social expenditure, however, varied across welfare regimes. In liberal and conservative regimes, a 1 per cent increase in ITO decreased social expenditure by 0.06 and 0.05 per cent, respectively. However, a 0.01 per cent difference in the effect size between the liberal and conservative regimes was not statistically significant, indicating that both regimes are similarly influenced by ITO on social expenditure. In social democratic regimes, a 1 per cent increase in ITO decreased social expenditure by 0.2 per cent. The effect size of ITO on social expenditure in the social democratic regimes was more than three times larger than in the liberal and conservative regimes. GDP per capita and unemployment rate were also positively associated with social expenditure in the interactioneffect model. Finally, we analysed three separate models that use three disaggregated measures of social expenditure. Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare 137

9 Kim & Zurlo Table 4. Effect of globalisation on disaggregated measures of social expenditure. Active labour Passive labour Social service β(t) β(t) β(t) Fixed-effects Covariates GDP per capita ( 10 2 ) (2.22)* ( 1.56) (0.31) Unemployment rate ( 0.21) (1.63) ( 0.37) Elderly population rate (3.28)*** (1.31) (1.04) Left cabinet (0.59) (0.70) ( 2.18)* Right cabinet ( 1.98)* ( 2.13)* ( 1.48) Globalisation indicators ITO ( 7.60)*** ( 4.34)*** (4.01)*** ESO (1.64) (1.21) ( 1.46) Welfare regimes Conservative (0.20) (0.44) (1.57) Social democratic (0.65) ( 0.63) ( 0.02) Intercept (0.11) (1.52) ( 0.33) Random-effects Within-country (12.2)*** (12.5)*** (12.6)*** Between-country (2.54)* (2.48)* (2.19)* 2 log likelihood GDP, gross domestic product; ITO, investment and trade openness; ESO, economic system openness. * p < 0.05; ** p < 0.01; *** p < Table 4 shows the effect of globalisation on active and passive labour market spending and social service spending. Since the statistical powers of each model were relatively low due to missing data, our statistical analysis was very limited. Nevertheless, the results provide instructive insight regarding the relationship between globalisation and disaggregated measures of social expenditure. Welfare regime type did not affect either active and passive labour market spending or social service spending, and therefore did not mediate the impact of globalisation on such spending. On the one hand, ITO negatively affected active and passive labour market spending. A 1 per cent increase in ITO brought about a and per cent decrease of active and passive labour market spending, respectively. On the other hand, ITO positively affected social service spending. A 1 per cent increase in ITO increased social service spending by per cent. The results imply that the impact of globalisation varies according to different types of social welfare policy. Specifically, while globalisation reduces labour marketrelated expenditures, it expands other social service-related expenditures. Summary and discussion In recent years the impact of globalisation on the welfare state has become a main argument in comparative policy studies, but with controversial results. We argue that the effect of globalisation on the welfare state can be influenced by welfare regime types. Our study provides a comprehensive examination of the relationship between globalisation and the welfare state with a state-ofthe-art analytical technique, concentrating on the mediating effect of welfare regimes. First, we find that 18 affluent countries experienced different trends and degrees of globalisation between 1980 and An analysis using the mixed-effect model demonstrates no significant differences in ESO among the three welfare regime types, while it shows significantly higher ITO in the social democratic regime type. During the same period, these countries also experienced different patterns of social expenditure. While the conservative regimes maintained increases during the entire period, the social expenditures of the social democratic and liberal regimes decreased after peaks in 1993 and 1992, respectively. The mixed-effect model analysis also confirms that there are statistically significant differences in social expenditures among the three welfare regimes. Second, we find that some globalisation indicators cause a reduction in the welfare state when controlling for other influential variables and adjusting for serial correlations within given countries. ITO is negatively related to social expenditure regardless of welfare regime type. In contrast, ESO is not associated with social expenditures. These results imply that globalisation, in terms of economic performance and not economic regulation, directly influences the welfare state. This finding supports, to a certain extent, the efficiency hypothesis, which emphasises the adverse effects of globalisation on the welfare state. This finding also empirically confirms Korpi s (2003) argument that the pronounced welfare state crisis results from globalisation in developed welfare states. 138 Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare

10 How does economic globalisation affect the welfare state? The more important finding pertains to the different responses to the impact of globalisation among the three welfare regime types. The mixed-effect model analysis reveals a significant cross-level interaction effect between the globalisation indicator and welfare regime type. Particularly with respect to ITO, social expenditures in social democratic regimes are more sensitive to the pressures of globalisation than they are in liberal and conservative regimes. Although ITO decreases social expenditure in all three regimes, the effect size of ITO in social democratic regimes is more than three times larger than it is in the liberal and conservative regimes. Despite statistically significant ITO coefficients in the liberal and conservative regimes, their effect size is marginal compared with the social democratic regimes. This result suggests that welfare regimes react differently to the impact of globalisation, and therefore mediate the relationship between globalisation and the welfare state. Globalisation is negatively exertive on the welfare state in social democratic regimes, while it also negatively, but marginally, affects liberal and conservative regimes. This finding contradicts the conventional wisdom that globalisation does not threaten the generous social democratic and conservative welfare states, while it undermines only the uncoordinated liberal welfare states (Brady et al., 2005: 925). Rather, our finding is partially consistent with a new line of thinking about globalisation and the welfare state, called the convergence hypothesis. Some scholars argue that for the highly globalised countries with generous welfare states, such as the social democratic regimes in our analysis, greater levels of globalisation would cause welfare state retrenchment toward the mean level of the welfare state. Thus, globalisation may force the most generous welfare states to reduce their scope to a moderate or normal level 6 (Brady et al., 2005: 924). Another instructive finding is that different social welfare policies respond differently to the impact of globalisation. Our findings show that, regardless of welfare regime type, developed welfare states reduce labour market-related spending while they expand social service-related spending. These results imply that nation-states simultaneously carry out two functions to cope with the impact of globalisation. On the one hand, as the efficiency hypothesis argues, nation-states decrease social expenditure to adapt to the imperatives 6 The convergence hypothesis emphasises not only welfare retrenchment in generous welfare states, but also welfare expansion in minimal welfare states. Hence, one can expect a convergence procedure in which big spenders reduce while low spenders expand the welfare state (Brady et al., 2005). As our analysis excludes less developed countries, we cannot discuss welfare expansion in minimal welfare states. Even the less generous welfare states, e.g. liberal regimes, in our analysis show relatively higher levels of social expenditure than less developed countries do. of global competition under globalisation. As Kim (2007) argues, the flexibility of labour market policy is regarded as a prerequisite for successful adaptation to the global economic system. Thus, governments intention to deregulate the labour market seems to be expressed by cutting labour market-related spending. On the other hand, as the compensation hypothesis contends, nation-states increase social expenditure to deal with the insecurity and uncertainty of the domestic economy caused by globalisation. It seems that, as Castells (2004) argues, nation-states choose the strategy in which governments compensate for the losses of their citizens from the structural adjustment of globalisation by enhancing social service spending. This finding implies that the impact of globalisation varies according not only to welfare regime types but also to specific social welfare policies. Based on our findings, we suggest that there is a need for further research in the following two directions. First, the concept and measures of globalisation and the welfare state should be defined more precisely. Considering the more recent globalisation phenomena, Quinn s openness score, which we employed to measure ESO, is out-dated. For example, entering into the Free Trade Agreement has been the most important criterion in recent years for deciding the level of economic regulation and exchange (Doumbia-Henry & Gravel, 2006). Quinn s indicator does not reflect this, however. Thus, redefined measurements of globalisation are in need of development. In addition, many criticisms have been levelled at social spending as a measure of the welfare state (Scruggs, 2004). When analysing the characteristics of the welfare state from a cross-national perspective, social spending alone may not present the redistribution pattern or structure of a given country, while it does provide a quantitative aspect of public welfare effort (Esping-Andersen, 1990). One alternative measure of the welfare state might be core programme entitlements and the conditions for receiving them developed by Allan and Scruggs (2004). Analysis of institutionally based measures of the welfare state is expected to show another dimension of the impact of globalisation on the welfare state. Second, while our study illustrates the mediating effect of the welfare regime on the impact of globalisation, the results do not provide an internal mechanism or dynamic that explains why alternative welfare regimes respond differentially to globalisation. We assume that the convergence hypothesis is one plausible explanation for the welfare state reduction in social democratic regimes. However, it is uncertain whether high levels of globalisation, coupled with a generous welfare state, provide the main explanation for welfare state retrenchment in social democratic regimes. If high levels of globalisation matter, the effect of welfare regime type is merely a compounding effect Journal compilation 2009 Blackwell Publishing Ltd and the International Journal of Social Welfare 139