A Service of

Leibniz-Informationszentrum econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible. zbw for

Biegert, Thomas

Article — Published Version Welfare Benefits and in Affluent Democracies: The Moderating Role of the Institutional Insider/Outsider Divide

American Sociological Review

Provided in Cooperation with: WZB Berlin Social Science Center

Suggested Citation: Biegert, Thomas (2017) : Welfare Benefits and Unemployment in Affluent Democracies: The Moderating Role of the Institutional Insider/Outsider Divide, American Sociological Review, ISSN 1939-8271, Sage Publications, Thousand Oaks, CA, Vol. 82, Iss. 5, pp. 1037-1064, http://dx.doi.org/10.1177/0003122417727095

This Version is available at: http://hdl.handle.net/10419/181677

Standard-Nutzungsbedingungen: Terms of use:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you genannten Lizenz gewährten Nutzungsrechte. may exercise further usage as specified in the indicated licence. www.econstor.eu ASRXXX10.1177/0003122417727095American Sociological ReviewBiegert 7270952017

American Sociological Review 2017, Vol. 82(5) 1037 –1064 Welfare Benefits and © American Sociological Association 2017 https://doi.org/10.1177/0003122417727095DOI: 10.1177/0003122417727095 Unemployment in Affluent journals.sagepub.com/home/asr Democracies: The Moderating Role of the Institutional Insider/Outsider Divide

Thomas Biegerta

Abstract The effect of generous welfare benefits on unemployment is highly contested. The dominant perspective contends that benefits provide disincentive to , whereas others portray benefits as job-search subsidies that facilitate better job matches. Despite many studies of welfare benefits and unemployment, the literature has neglected how this relationship might vary across institutional contexts. This article investigates how and minimum income benefits affect unemployment across levels of the institutional insider/ outsider divide. I analyze the moderating role of the disparity in protection for holders of permanent and temporary contracts and of the configuration of bargaining. The analysis combines data from 20 European countries and the United States using the European Union Labour Force Survey and the Current Survey 1992–2009. I use a pseudo-panel approach, including fixed effects for sociodemographic groups within countries and interactions between benefits and institutions. The results indicate that unemployment benefits and minimum income benefits successfully subsidize job search and reduce unemployment in labor markets with a moderate institutional insider/outsider divide. However, when there is greater disparity in employment protection and when bargaining either combines low unionization with high centralization or high unionization with low centralization, generous benefits create a disincentive to work, plausibly because attractive job opportunities are scarce.

Keywords unemployment, , labor institutions, institutional interactions, quantitative methods

In 2008, U.S. policymakers extended the at the end of 2013. The basic argument for maximum duration of unemployment benefits not further extending benefits was, as U.S. from a 26-week limit to 73 weeks (with some variation across states) to ameliorate the aWZB Berlin Social Science Center effects of the dramatic increase in unemploy- ment during the economic crisis. This more Corresponding Author: generous approach to benefits was short- Thomas Biegert, WZB Berlin Social Science Center, Reichpietschufer 50, 10785 Berlin, lived, however, as the temporary Emergency Germany Unemployment Compensation program expired E-mail: [email protected]

Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-geförderten) Allianz- bzw. Nationallizenz frei zugänglich / This publication is with permission of the rights owner freely accessible due to an Alliance licence and a national licence (funded by the DFG, German Research Foundation) respectively. 1038 American Sociological Review 82(5)

Senator Jon Kyl put it in 2010, that generous comparative analyses. Some of this work unemployment “doesn’t create new offers empirical support for the disincentive jobs. In fact, if anything, continuing to pay argument (e.g., Card et al. 2015; Lalive 2007; people unemployment compensation is a dis- Meyer 1990; Nickell 1997); other work sup- incentive for them to seek new work” (156 ports the job-search-subsidy argument (e.g., Congressional Record:845). Regarding Sena- Gangl 2004, 2006; Nelson and Stephens tor Kyl’s point, Card and colleagues (2015) 2012). Moreover, critical assessments of the argued that although the duration of unem- literature have convincingly called into ques- ployment spells generally increases with ben- tion the existing results, particularly in regard efit levels, this relationship was particularly to how micro-level mechanisms out at strong during the crisis. Indeed, many schol- the aggregate level (Atkinson and Mickle- ars argue that generous and long-running wright 1991; Avdagic and Salardi 2013; Bac- welfare benefits act as a disincentive to work caro and Rei 2007; Baker et al. 2005; Howell and thus increase unemployment (e.g., Lalive and Rehm 2009; Sjöberg 2000). Ultimately, 2007; Layard, Nickell, and Jackman 2005; the literature offers quite mixed empirical evi- Meyer 1990; Nickell 1997; for a recent over- dence on the relationship between generous view see Schmieder and von Waechter 2016). welfare benefits and unemployment levels. The notion of a disincentive effect has found This article proposes that the effect of ben- its way into influential policy recommenda- efits on unemployment differs by institutional tions (OECD 1994, 2006) and has been used context. The argument rests on the assumption to justify welfare state retrenchment in a that benefits are part of the interplay of institu- number of countries in recent decades (Blyth tions in a complex system (see Amable 2003; 2002). Hall and Soskice 2001; Iversen 2005). Draw- The disincentive perspective is highly con- ing on recent research on the growing dispari- tested. Other researchers have presented funda- ties between labor market insiders (individuals mentally different arguments about the effect of with permanent employment) and outsiders welfare benefits on unemployment. Howell and (the unemployed and individuals with tempo- Azizoglu (2011), for instance, argue that if ben- rary employment) (Barbieri and Cutuli 2016; efit generosity increased U.S. unemployment Biegert 2014; Emmenegger et al. 2012; Palier during the crisis, it did so by keeping workers and Thelen 2010; Schwander and Häuser- closely attached to the labor market rather than mann 2013), I highlight the interaction of by encouraging them to drop out of the labor welfare benefits with the institutional divide force. The most prominent theoretical argu- between labor market insiders and outsiders. I ment is that instead of functioning as work argue that the institutional insider/outsider disincentives, benefits financially subsidize the divide is determined by the disparity between job-search process (Gangl 2004, 2006; Pollmann- employment protection legislation (EPL) for Schult and Büchel 2005). According to this individuals on permanent contracts and tem- perspective, welfare benefits for the unem- porary contracts and by the configuration of ployed relieve the pressure to take on bad jobs unionization and centralization in the wage and enable further education and training bargaining process. By inadvertently influenc- (Estevez-Abe, Iversen, and Soskice 2001; ing the availability of quality jobs for jobseek- Morel, Palier, and Palme 2012). Eventually, ers, the degree to which institutions create a this leads to better job matches and fewer job division between labor market insiders and separations, which lowers unemployment in outsiders moderates how generous welfare the long run (Nelson and Stephens 2012). benefits affect unemployment. The disincentive and job-search-subsidy The article makes three unique contribu- perspectives differ greatly in their understand- tions to the literature. First, by focusing on the ing of the functions of welfare benefits interaction with the institutional insider/out- for jobseekers. Both views have been tested in sider divide, it develops theoretical explana- a multitude of case studies and macro tions of the relationship between welfare Biegert 1039 benefits and labor market outcomes. To test disincentive perspective is based on the theo- these propositions empirically, I combine retical argument that individuals’ reservation institutional indicators with data on individual increase when high living standards unemployment from 20 European countries in can be achieved without having to work. the European Union Labour Force Survey Given that the reservation wage marks the (EULFS) and the United States Current Popu- line below which individuals will reject job lation Survey (CPS; Flood et al. 2015) for offers, higher benefits will result in a larger 1992–2009. I transform the cross-sectional number of jobs not being taken. Therefore, time series data into pseudo-panels at the level the basic expectation is that countries with of sociodemographic groups (Deaton 1985; more generous welfare benefits will have Verbeek and Vella 2005). Applying the higher unemployment. pseudo-panel technique enables the use of Many quantitative studies have tested the fixed effects. It thus helps rule out bias due to disincentive perspective. Micro-level studies stable differences between sociodemographic usually look at policy shifts and explore groups and country-specific factors. As a sec- whether they lead to changes in the duration ond contribution, the analysis illustrates an of individual unemployment spells (e.g., Card underutilized method that provides compara- et al. 2015; Lalive 2007; van Ours and Vodop- tive researchers with a novel approach to ivec 2006). Meyer (1990), for instance, cross-sectional time series data (for other showed that higher benefits lead to fewer exits recent applications, see Barbieri and Cutuli from unemployment in the United States. 2016; Jæger 2013; Neugebauer 2015). Finally, Strengthening the case for the disincentive the article extends the scope of previous stud- perspective, he found that exits from unem- ies by analyzing both unemployment benefits ployment became more frequent just when and minimum income benefits. Most existing benefits were about to expire. More recently, research focuses on unemployment benefits as Card and colleagues (2015) used data from the sole welfare benefit for the unemployed. Missouri and a regression kink design to show This neglects the fact that many jobseekers are that the increase in the unemployment dura- not eligible to receive unemployment insur- tion due to benefits is markedly larger in ance (Atkinson and Micklewright 1991; Howell adverse macro-economic conditions. and Rehm 2009). Minimum income benefits To test whether the micro-level mecha- include social assistance, benefits, nism plays out on the aggregate level, macro- child support, and other benefits the state pro- comparative studies use the variation of vides to needy people when benefit systems, unemployment benefits across countries and such as unemployment insurance, fail (Nelson years to investigate the relationship with 2010). By analyzing the relationship between unemployment rates. In a prominent study, unemployment and both unemployment ben- Nickell (1997) regressed unemployment rates efits and minimum income benefits, the pre- on macro indicators of institutional arrange- sent analysis thus delivers more robust ments. He found a positive association evidence for the relationship between welfare between generous benefits and unemploy- benefits for the unemployed and unemploy- ment in OECD countries. In comparison to ment levels. findings regarding other labor market institu- tions, such as EPL, unionization, and wage bargaining centralization, the detrimental Welfare Benefits as a effect of unemployment benefits is one of the Work Disincentive more consistent results in this literature (e.g., A large body of research argues that generous Layard et al. 2005; OECD 2006). benefits for jobseekers create a disincentive At first sight, there seems to be empirical to work, which raises unemployment levels support for the disincentive perspective. How- (e.g., Card et al. 2015, Lalive 2007; Layard ever, several critical summaries of the litera- et al. 2005; Meyer 1990; Nickell 1997). The ture argue that the existing evidence is far less 1040 American Sociological Review 82(5) compelling than is widely believed (see Atkin- job-search subsidy (Gangl 2004, 2006; Nelson son and Micklewright 1991; Avdagic and Sala- and Stephens 2012). The basis for this argu- rdi 2013; Baccaro and Rei 2007; Baker et al. ment is that employment and unemployment 2005; Howell and Rehm 2009; Sjöberg 2000). are the result of a matching process conducted Although the evidence from micro-level stud- by jobseekers and employers (see Sørensen ies is fairly robust, Howell and Rehm (2009) and Kalleberg 1981). In contrast to the ortho- point out that the magnitude of the effects dox economic perspective, the job match per- found in these studies is typically quite modest spective highlights that unemployment is not (see also Atkinson and Micklewright 1991). primarily caused by poor labor supply. Rather, Single-country case studies, moreover, cannot it is the result of the interaction between labor account for the labor market and macro-eco- supply and labor demand. nomic context, which raises the question of In this framework, generous benefits can confounding macro factors and generalizabil- promote employment through two mecha- ity. Regarding macro-comparative studies, nisms. First, benefits allow jobseekers to be Howell and Rehm (2009) problematize the more selective about job offers by providing comparability of the reported unemployment a buffer in times of joblessness. This rates used in these studies because national increases the quality of the ultimate match definitions of unemployment differ. Further- between jobseeker and job, which in turn more, they criticize the use of gross replace- decreases separations and thus boosts ment rates as an indicator for unemployment employment levels (Gangl 2004, 2006; Poll- benefit generosity (i.e., the percentage of pre- mann-Schult and Büchel 2005). Second, vious earnings an average production worker generous benefits allow workers to invest in receives from insurance before and social specific skills, because they provide insur- security contributions). ance at times when these workers are seek- Studies using net replacement rates (i.e., ing jobs. The result is a workforce with the insurance payments net of taxes and generally higher and more specific skills, contribution) have found again creating higher quality matches for job only a weak correlation between unemploy- vacancies (Estevez-Abe et al. 2001). Hence, ment benefits and unemployment rates proponents of the job-search-subsidy per- (Howell and Rehm 2009; Sjöberg 2000). spective do not question the potentially These limitations are increasingly recognized longer duration of individual unemployment in the literature on the disincentive perspec- spells in a context of high benefits. How- tive (for an overview, see Schmieder and von ever, they reason that generous benefits will Waechter 2016), but most of the discussion is lead to lower unemployment in the long run limited to the size of the disincentive effect. and on the aggregate level because of better In summary, the empirical support for the job matches and the improved employability disincentive argument is mixed, particularly of jobseekers (see also Morel et al. 2012; in regard to aggregate unemployment. Addi- Wulfgramm and Fervers 2015). tional mechanisms such as job-search-subsidy These arguments have been tested in effects could explain why micro-level disin- micro-level designs and macro-comparative centives do not necessarily translate into research, although not as extensively as the aggregate unemployment. disincentive argument. On the micro level, studies have focused on the quality of job matches after unemployment spells and how Welfare Benefits as a they vary across different benefit environ- Job-Search Subsidy ments. Pollmann-Schult and Büchel (2005), In contrast to the disincentive perspective, for instance, showed that in Germany, not scholars have argued that generous benefits receiving unemployment benefits is associ- could reduce unemployment by serving as a ated with shorter search periods but also with Biegert 1041 over-education in the next job. Comparing benefits. Minimum income benefits—that is, Germany and the United States, Gangl (2004) means-tested publicly provided benefits— found that generous unemployment benefits have rarely been considered in studies on lead to less dramatic losses in post-unemployment benefit effects. Theoretically, they should income. This finding was confirmed in a operate via the same mechanisms as unem- wider comparison involving the United States ployment benefits—either by acting as work and 12 European countries (Gangl 2006). disincentives or by subsidizing job search. Nelson and Stephens (2012) investigated the Because minimum income benefits are more employment effects of social investment poli- widely available to the unemployed popula- cies, such as generous initial unemployment tion, they need to be included in studies of benefits,1 active labor market policies social security regimes for unemployed indi- (ALMP), and childcare provision. They con- viduals (Pfeifer 2012). ducted an analysis similar to the macro-level studies on the disincentive perspective and found that unemployment benefits are posi- The Institutional tively associated with employment levels and Insider/Outsider job quality. Divide and Welfare Beyond the methodological questions dis- Benefit Effects on cussed in the previous section, both literatures Unemployment have unfortunately neglected two issues. First, neither account has investigated institu- The argument in this section rests on the tional interactions. The literature on national assumption that welfare benefits are part of a production systems argues strongly for the complex institutional system (Amable 2003; existence of interdependencies in the institu- Hall and Soskice 2001; Iversen 2005): “insti- tional settings of (see Amable tutions matter and . . . institutions interact” 2003; Hall and Soskice 2001; Iversen 2005). (Belot and van Ours 2004:640). Hence, the The function of an institution such as welfare relationship between benefits and unemploy- benefits likely depends on the institutional ment plausibly depends on the distribution of context. Institutional interactions have started job opportunities for jobseekers. Job-search to play a more prominent role in quantitative subsidies will likely be more successful when labor market research (e.g., Bassanini and there is a larger pool of quality job opportuni- Duval 2009; Belot and van Ours 2004; Gangl ties. Conversely, generous benefits should act 2006; see also Schmieder and von Waechter as a stronger disincentive to jobseekers in 2016). However, to the best of my knowl- contexts with fewer good job opportunities. edge, no studies on unemployment have Recent research that describes a growing gap explicitly focused on the interaction between between labor market insiders and outsiders benefits and their labor market context. indicates that the institutional configuration Second, both literatures focus overwhelm- of the labor market affects the distribution of ingly on unemployment benefits. In most job opportunities. countries, social rights to unemployment ben- efits have to be earned through previous Institutions and the Divide between employment. As a consequence, only a par- Insiders and Outsiders ticular segment of the unemployed can receive benefits through unemployment With the growing attention to rising economic insurance. This selectivity might result in inequality in rich democracies, there has been biased outcomes because the dependent vari- a renewed interest in the notion of labor mar- able in macro studies is usually the aggregate ket insiders and outsiders. Building from clas- unemployment rate, which includes many sic theories of insiders and outsiders and dual who are not eligible for unemployment or segmented labor markets (see Doeringer 1042 American Sociological Review 82(5) and Piore 1971; Kalleberg, Wallace, and regulating institutions have effects on overall Althauser 1981; Lindbeck and Snower 1988), unemployment. Instead, it argues that the European scholars have described a process institutionalized advantages enjoyed by insid- of labor market dualization between individu- ers may affect the distribution of jobs. als with permanent employment (insiders) and those without it (outsiders), that is, the The Institutional Insider/Outsider unemployed and individuals in temporary Divide, Welfare Benefits, and jobs, who have a higher propensity of becom- ing unemployed (Emmenegger et al. 2012; Unemployment Palier and Thelen 2010; Rueda 2005, 2014).2 The divide between labor market insiders and This recent literature rests on the assumption outsiders is relevant to the study of how ben- that insiders are in a powerful bargaining efits affect unemployment for two reasons. position because the replacement of workers First, it is the unemployed, and thus outsiders, has transaction costs for employers. There- who receive unemployment benefits or mini- fore, insiders can achieve their goals in a mum income benefits. Second, the insider/ variety of ways, be it through cross-class outsider divide affects the availability of coalitions with employers, union representa- higher quality jobs. In a context of a strong tion, or social democratic parties (see Davidsson insider/outsider divide, employed insiders and Emmenegger 2013; Goldthorpe 1984; remain in their positions and employers are Rueda 2005). This, the dualization literature more selective, which leads to fewer and argues, disadvantages outsiders. worse job offers for outsiders. Here, generous Although insiders have an advantageous benefits create a disincentive to job search bargaining position in every , the because quality job offers are scarce, whereas dualization literature argues that the gap low benefits may force jobseekers to take on between insiders and outsiders can be wid- jobs they would otherwise decline. By con- ened by labor market institutions. The setup trast, a labor market with a moderate insider/ of institutions such as EPL and the wage bar- outsider divide will yield better job opportu- gaining process determine the positional nities for jobseekers. In this instance, gener- advantage of individuals in jobs at the core of ous benefits might lower unemployment the labor market (Emmenegger et al. 2012; because improved job-search and job-matching Palier and Thelen 2010; Thelen 2014). Tradi- processes can help jobseekers avoid bad jobs. tionally, the literature has emphasized the Two sets of labor market institutions are benefits of regulation and coordination, for most likely to determine the insider/outsider instance, in regard to wage inequality (e.g., divide: (1) EPL for permanent and temporary Brady, Baker, and Finnigan 2013; Jacobs and contracts, and (2) the configuration of the Myers 2014). By contrast, the dualization lit- wage bargaining process in terms of unioni- erature points to configurations in which reg- zation and centralization (see Emmenegger ulating institutions have unintended effects et al. 2012; Palier and Thelen 2010; Rueda on inequality (Emmenegger et al. 2012; Palier 2005; Thelen 2014). First, EPL determines and Thelen 2010; Rueda 2005, 2014). Thelen how difficult it is to hire and fire employees, (2014) cautions researchers not to conflate which reduces the flow in and out of the labor coordinated with egalitarian capi- market. EPL thus stabilizes the positions of talism. The unintended consequence of insti- insiders (Barbieri 2009; Gangl 2003; Gebel tutionalized advantages for labor market and Giesecke 2011). If letting people go in insiders is that it increases the barriers that economic downturns is impeded by strict jobseekers have to overcome to become insid- EPL, employers are less likely to offer insider ers (Biegert 2014; Fervers and Schwander positions to jobseekers. This may be exacer- 2015). Unlike orthodox economics, however, bated if employers can easily offer temporary the dualization literature does not claim that instead of permanent contracts (Eichhorst and Biegert 1043

Marx 2011; Palier and Thelen 2010). If tem- with the disparity in EPL for permanent and porary contracts are less protected than per- temporary contracts, unionization, and wage manent contracts, employers are more likely bargaining centralization. Benefits increase to offer temporary jobs to jobseekers, and less unemployment in contexts with a larger EPL likely to convert temporary jobs into perma- disparity, higher unionization, and higher centralization. Benefits reduce unemploy- nent positions. This increases the number of ment in contexts with a smaller EPL dispar- job separations and reduces the number of ity, lower unionization, and lower central- attractive job opportunities (Barbieri and ization. Cutuli 2016; Bentolila et al. 2012; Gebel and Giesecke 2016; Noelke 2016). The disparity Hypothesis 1 is called the regulation between EPL for permanent contracts and hypothesis because it posits that the insider/ temporary contracts will thus increase the outsider divide is determined on a single insider/outsider divide and reduce job oppor- dimension between a flexible and a regulated tunities for jobseekers. labor market. Yet, there is reason to doubt that Second, unionization and centralization labor market institutions have a uniform may shape the wage bargaining process to the impact on the insider/outsider divide. When detriment of outsiders. Higher levels of describing institutional regimes and their unionization increase unions’ bargaining dualization tendencies, researchers have power, which results in less wage inequality, found large insider/outsider divides in Conti- better working conditions, and stronger wage nental European and Mediterranean coun- growth for the employed (Brady et al. 2013; tries. Liberal and show Freeman and Medoff 1984; Rueda and Pon- lower levels of dualization (Häusermann and tusson 2000; Western and Rosenfeld 2011). Schwander 2012; Thelen 2014). The assump- Because their membership consists primarily tion that higher levels of regulation necessar- of permanent workers, unions are often ily lead to a greater insider/outsider divide is thought to represent insider interests (Lind- thus at odds with the high levels of unioniza- beck and Snower 1988). In addition, increas- tion and wage bargaining centralization in ing wages and secure positions for the Nordic countries. Hence, the complementar- employed might make employers reluctant to ity of institutional arrangements is essential. hire. If so, jobseekers will be more likely to When centralized wage bargaining is in place, receive offers for atypical jobs, if any. In strong unions have an incentive to pursue addition to unionization, wage bargaining moderate wage growth that benefits the whole centralization is important in the bargaining economy, because detrimental outcomes can- process (Palier and Thelen 2010; Rueda not be externalized (Calmfors and Driffill 2005). Centralization refers to the level at 1988). Thus, strong unions cooperate with which bargaining takes place and the degree employers and the government to achieve a to which unions are able to coordinate their beneficial bargaining outcome (Thelen 2014; goals. Centralized bargaining might help Wright 2000). This results in moderate wage unions achieve better outcomes for their increases and better job opportunities for job- members, which, in line with the previous seekers (Hicks and Kenworthy 1998; Streeck argument, could come at the detriment of 1992; Western 1998).3 By contrast, some outsiders. Unionization and centralization Continental European countries, such as Ger- could thus increase the insider/outsider many, are less unionized. In such cases, divide, which would negatively affect job unions seek bargaining outcomes tailored to opportunities for jobseekers. their specific clientele (Eichhorst and Marx 2011; Palier and Thelen 2010).4 They pursue Hypothesis 1 (regulation): The effect of increas- increases, which are less ing unemployment benefits and minimum considerate of the greater economy but still income benefits on unemployment varies have widespread influence because of 1044 American Sociological Review 82(5) relatively high levels of centralization and The pseudo-panel technique is relatively coverage. The outcome is beneficial for insid- uncommon in sociological research (but see ers but leaves outsiders with lower chances of Barbieri and Cutuli 2016; Jæger 2013; Neu- quality employment. Accordingly, a regulated gebauer 2015). Pseudo-panels allow research- labor market regime may lead to either seg- ers to estimate panel data models on the basis mented or solidaristic coordination depending of repeated cross-sections. In a seminal arti- on the configuration of its institutional com- cle, Deaton (1985) proposed following ponents (Thelen 2004). cohorts and estimating cohort fixed effects from repeated cross-sectional data. The Hypothesis 2 (configuration): The effect of in- researcher can define cohorts by any number creasing unemployment benefits and mini- of time-constant individual characteristics. mum income benefits on unemployment varies The idea is that after grouping all individuals with the configuration of the wage bargaining who share the same individual characteristics process. Benefits increase unemployment in into cohorts, researchers can treat the group contexts of lower unionization and higher cen- tralization and vice versa. Benefits reduce un- means within these cohorts as panel observa- employment in contexts of high unionization tions. When following the cohorts, scholars and high centralization and contexts of low observe new samples of individuals every unionization and low centralization. year. At the group level, however, individuals can be considered comparable over the years as long as the repeated cross-sections are rep- ANALYTIC APPROACH resentative. This study uses country-specific The analysis aims to estimate how the relation- birth cohort, sex, and education groups as the ship between benefits and unemployment var- units of analysis. The dependent variable is ies by the level of the institutional insider/ the mean of unemployment within these outsider divide. To this end, I combine the time groups, that is, their unemployment rate. The series of country-level institutional indicators analysis thus operates on the meso level of with data on individual unemployment from 20 the groups. Instead of correlating institutional European countries and the United States from indicators with country-level unemployment 1992 to 2009.5 The use of cross-national data rates, I examine the relationship between ben- over a long period of time maximizes variation efits and the employment performance of in institutional arrangements. The multilevel sociodemographic groups, using individual- structure of the data overcomes limitations of level information to model the group mean. prior studies, which often operated exclusively The analysis uses the panel data structure at the macro level (e.g., Layard et al. 2005; in fixed-effects panel regression models (Alli- Nelson and Stephens 2012; Nickell 1997; son 2009). Fixed-effects models use repeated OECD 2006). I transform the repeated cross- observations of the unit of analysis to decom- sectional data from the EULFS and CPS into pose the error term of a linear regression pseudo-panels to estimate fixed-effects models model into a time-constant error term ϑi and (Deaton 1985). The term pseudo-panels refers a time-varying error term εit. Formally, the to a technique to transform repeated cross-sec- models can be expressed as follows: tional data on the individual level into synthetic panel observations on the level of social groups; Ycit =+ixββXPit ++Pitxβ PiXPtit this enables the use of panel regression tech- ++βϑZiZ ti+ εit niques in the absence of real individual-level panel data. Because the analytic approach is where Yit is the unemployment rate of socio- intertwined with the construction of the dataset, demographic group i in year t. On the right- I first explain pseudo-panels in more detail and hand side ci is the time-constant cohort-specific how I use them in fixed-effects panel regres- intercept. βxpXit refers to the welfare benefits, sion models. that is, unemployment benefits and minimum Biegert 1045

income benefits. βpPit signifies the institu- specific advantages of education, age, or sex tional insider/outsider divide. βxpXitPit repre- groups. In addition, the models use panel robust sents the interaction term between benefits standard errors, which are consistent in case of and the institutional insider/outsider divide. I heteroscedasticity and autocorrelation.7 test the regulation hypothesis by using two- In addition to enabling fixed-effects way interactions between the benefits and the regression, a pseudo-panel approach with EPL ratio, unionization, and wage bargaining cross-national comparative multilevel data centralization. I test the configuration hypoth- provides a number of advantages over pre- esis with three-way interactions, in which the vious studies. First, unlike country case interaction between benefits and unionization studies, this analysis models the macro con- is further interacted with centralization. βZZit text of benefits and the heterogeneous denotes various time-varying control variables effects of benefits across different institu-

(see below). ϑi and εit are the time-constant tional settings. Second, the individual-level and time-varying component of the error term. information on labor market status is cross- The fixed-effects transformation subtracts nationally comparable. Previous macro the unit-specific mean of each variable from comparative studies relied on nationally its value in each time period. The equation reported unemployment rates. Because estimated via the fixed-effects models is the national definitions of unemployment differ, following: comparing reported unemployment rates is problematic (Howell and Rehm 2009). This ()YYit −=ixββ()XXit −+iP()PPit − i is not an issue for the present study because the unemployment rates for the cohorts are +−ββXP XP +−()ZZ++ ε xP ()it it ii Ziti it based on micro-level information.8 Third, the multilevel structure of the data allows The transformation eliminates the time-con- for adjustment for compositional heteroge- stant group-level difference in unemployment neity. Previous research shows that macro- ci and the time-constant error term ϑi . As a level institutions can have idiosyncratic result, coefficients are estimated using within- effects on unemployment in different age, subject variation alone, which rules out bias sex, and education groups (Bertola, Blau, due to time-invariant unobservables.6 The and Kahn 2007). In contrast to macro-level cohort panels in this study are determined by studies, the meso-level data make it possible the country of residence, birth year, sex, and to adjust for compositional heterogeneity in education. Thus, the fixed effects account for terms of country-specific birth cohort, sex, all time-constant unobserved confounders on and education groups.9 the level of these groups. The inclusion of Of course, pseudo-panels are not prefera- fixed effects on the meso level of synthetic ble over individual-level panel data. One cen- cohorts is the key advantage over the more tral assumption of the present models is that frequently applied two-way fixed-effects the effect of benefits is not systematically models, which use individual-level data in a biased within the country-specific birth cross-national setting and include fixed cohort, sex, and education groups. However, effects for countries and time (e.g., Brady et al. in the absence of long-running cross-national 2013). The pseudo-panel fixed effects account panel data, constructing pseudo-panels from for more fine-grained unobserved time-con- the repeated cross-sections is arguably a next- stant heterogeneity than do country-level best solution. The representative nature of the fixed effects. The technique thus offers an cross-sections could even lead to advantages invaluable advantage in light of the number compared to individual panel data, because of potential confounders (Allison 2009). For these data are not subject to panel attrition instance, this differences out labor market and nonresponse as sources of bias (Deaton structures, culture, work ethic, and the 1985; Verbeek and Vella 2005). 1046 American Sociological Review 82(5)

Data high education represents persons with a ter- tiary education (ISCED5-6). Hence, I sort The analysis uses data on all working-age indi- individuals from 21 countries, 2 sexes, 13 viduals (age 15 to 64) from the EULFS for 20 birth cohorts, and 3 educational levels into European countries and the CPS for the United 1,638 units of observation. Because some Sates covering 1992 to 2009. Both the EULFS countries entered the EULFS at a later time and the CPS provide large-scaled, standard- point or missed years, some birth cohorts are ized, and representative repeated cross- not part of the sample, which leaves the actual sectional information on individuals in private number of units at 1,579.11 The units of obser- with a special focus on their work- vation are followed up to 18 years, yielding a ing life. Because I only use basic variables on total case number of 18,266. labor market status, birth year, sex, education, The dependent variable is labor market sta- and marital status, the two datasets are compa- tus (1 = unemployed, 0 = employed). The sur- rable (for other work combing the two data­ veys follow the definition of the ILO (1982), sets, see Hipp and Leuze 2015). The which considers people unemployed if they do combination of the datasets offers the singular not have a job, if they have actively looked for opportunity to analyze the labor markets of work in the past four weeks, and if they are cur- and the United States over a long time rently available for work. Individuals are period using annual micro-level data. The deemed employed if they had worked at least analysis starts in 1992 because the EULFS did one hour during the previous week.12 Members not collect information on education prior to of the country-specific sociodemographic that year. Macro-level institutional indicators groups are collapsed into one observation. are merged with the micro data. The full set of Therefore, individual-level variables are aggre- macro-level variables is available until 2009, gated at the group mean. Consequently, the which is the upper limit for the observation dependent variable is the sociodemographic period. Before transforming the data into groups’ unemployment rate.13 When construct- pseudo-panels, the full dataset comprises ing the cohorts, yearly cohort-cells have to be almost 20 million cases, with yearly case num- of reasonable size so that the unemployment bers ranging from about 8,500 in Denmark in rate can be estimated robustly.14 In the present 2000 to about 270,000 for Italy in 2005. dataset, the cohort-cells have 16,500 members on average, which should provide for reliable measurement. However, cohort-cell case num- Panel Construction and Dependent bers vary widely. To tackle possible measure- Variable ment error in the dependent variable, the models The cohort panels are based on information on weight the cohort-cells by the square root of the country of residence, sex, birth cohort, and number of observations (Neugebauer 2015). education. To construct the sociodemographic This gives more weight to observations whose groups, the individuals are first sorted accord- measurement of the unemployment rate is more ing to their 21 different countries of residence. robust.15 Then they are grouped according to their sex. The sample is split into 13 five-year birth Explanatory Variables cohorts spanning birth years from 1928 to 1994.10 Finally, individuals are grouped The main explanatory variables are country- according to three educational levels follow- level institutions. I use net replacement rates ing the 1997 version of the International Stan- to indicate the generosity of unemployment dard Classification of Education (ISCED): benefits. Net replacement rates are the per- low education includes respondents with centage of one’s former income received by lower secondary education or less (ISCED0- an average production worker from unem- 2), medium education contains upper- and ployment insurance net of taxes and social post-secondary education (ISCED3-4), and security contributions. Van Vliet and Biegert 1047

Caminada (2012) provide net replacement et al. 2005; Nelson and Stephens 2012; Nick- rates for singles and one-earner families with ell 1997; OECD 2006). An important dimen- two children that focus on the replacement sion of the wage bargaining process is how rate in the initial phase of unemployment. To many workers are actually covered by its take into account different family situations, I outcome. I use Visser’s (2013) adjusted cov- use the average of the two indicators.16 erage indicator to model the percentage of Nelson (2010) collects information on the workers who are covered by wage bargaining absolute amounts of minimum income bene- agreements.19, 20, 21 Welfare states try to “acti- fits. The main component of the indicator is vate” unemployed citizens via active labor social assistance. Housing supplements, child market policies (ALMP), such as labor mar- support, and other benefits are added as long ket retraining, job-search assistance, direct as they are not deducted from social assis- job creation, and employment subsidies tance. I use the average absolute payments (Bonoli 2010). The extent to which a country from several constellations to con- invests in ALMP is measured as public expen- struct a ratio that divides payments by the ditures relative to the GDP. To consider busi- average wage. The indicator thus captures the ness cycles, this indicator is divided by the economic support that minimum income ben- unemployment rate. Labor taxes affect labor efit schemes provide as a percentage of the demand and supply because they increase average wage. Minimum income benefits do labor costs for employers yet also lower not just serve as an alternative measure of employees’ net earnings (Nickell 1997). The unemployment benefits. The correlation of OECD calculates the labor wedge for a the two benefit schemes was rather low (.31 single-earner couple with two children and an in the dataset used here). average income. The tax wedge is the sum of I use three indicators to measure the institu- personal and social security con- tional insider/outsider divide on the labor mar- tributions as a percentage of total income. ket. The OECD provides time series for the Childcare policies affect the job opportunities employment protection of individuals on per- of parents who want to work (Gornick, Mey- manent and temporary contracts, which quanti- ers, and Ross 1997). A country’s dedication to fies the costs and procedures involved with publicly provided childcare is measured via dismissal (range 0 to 6) (Venn 2009). To capture the total on childcare as a the disparity in protection for workers in differ- percentage of the GDP. Finally, I use the ent contract types, I calculate the EPL ratio OECD’s output gap as an indicator of national between the two indicators.17 Unionization, also business cycles (Bassanini and Duval 2009). provided by the OECD, captures the organiza- The output gap measures the distance between tional power of unions as the percentage of the trend-based prediction of a country’s GDP - and wage-earners who are union mem- and actual outcome. Table A1 in the online bers. To measure the centralization of the wage supplement summarizes the macro-level indi- bargaining process, I use an indicator developed cators by country. Because the multivariate by Iversen (1999) and extended by Visser analysis is based on within-country variation, (2013). The indicator measures the degree of the table provides the average value of the coordination and centralization by combining respective indicator by country (x–), the the level of bargaining and union concentration within-country standard deviations (w-sd), at the respective levels (range 0 to 1).18 and the number of years in which the indica- tor changed compared to the previous year (N Δ). Because the multivariate analysis used Controls standardized indicators to facilitate the com- Following previous studies, the analysis parison of effect sizes, the table also shows includes a set of macro-level controls that the overall within standard deviation to pro- might vary with time and were thus not vide substantive meaning for the interpreta- accounted for by the fixed effects (see Layard tion of the coefficients. 1048 American Sociological Review 82(5)

Figure 1. Changes in Unemployment Rates and Unemployment Benefits from 1992 to 2009 in 21 Countries Note: Change scores for unemployment and unemployment benefits computed as differences between 1992 and 2009 (due to missing data, first observation is 1994 for SK and CZ, 1995 for HU, 1996 for PT, 2000 for EE, and 2002 for SI).

In addition to the institutional and macro- and 2009.22 Panel A in the two figures shows economic indicators, all models include dum- the overall bivariate association between mies for the survey waves to account for changes in benefits and unemployment in the economic shocks and trends that affect all 21 countries. To see whether there is descrip- countries. This makes it possible to assess the tive evidence for a moderating impact of the impact of institutional changes against a com- institutional insider/outsider divide, Panel B mon trend. Finally, the household context is groups countries into Continental/Mediterra- strongly associated with individual unem- nean, Nordic, Anglo-Saxon, and Eastern Euro- ployment (DiPrete and McManus 2000). In pean clusters. The existing literature sees a the absence of an ideal measure for the typical large insider/outsider divide in Continental/ household composition in the sociodemo- Mediterranean countries and a smaller divide graphic groups, the models include marital in Nordic and Anglo-Saxon countries (Häuser- status, operationalized as the average rate of mann and Schwander 2012; Thelen 2014). married individuals within the respective Eastern European countries have so far not groups. been included in such typologies. Figure 1 shows that, comparing 1992 and 2009, there is a negative bivariate association RESULTS between changes in unemployment benefits Descriptives and changes in the unemployment rate (Panel A). Some countries, such as Italy, , Figures 1 and 2 plot the change in unemploy- Denmark, and Ireland, managed to lower ment rates (provided by the OECD) against the unemployment by almost 4 percentage points. change in the level of unemployment benefits On the other hand, unemployment in Sweden, and minimum income benefits between 1992 Portugal, and the Czech Republic increased Biegert 1049

Figure 2. Changes in Unemployment Rates and Minimum Income Benefits from 1992 to 2009 in 21 Countries Note: Change scores for unemployment and unemployment benefits computed as differences between 1992 and 2009 (due to missing data, first observation is 1994 for SK and CZ, 1995 for HU, 1996 for PT, 2000 for EE, and 2002 for SI).

by more than 2 percentage points in the same increases of about 10 percentage points in period.23 Changes in the level of unemploy- Italy, Portugal, and Ireland.24 A strong nega- ment benefits range from reductions of around tive association between changes in minimum 20 percentage points in Sweden and Hungary income benefits and unemployment again to increases of 20 percentage points in Esto- emerges in the Anglo-Saxon and Nordic nia and almost 40 percentage points in Italy. countries (Panel B) when I cluster the coun- By clustering the countries, I can draw fitted tries and apply fitted lines. The bivariate lines that indicate potential institutional inter- association is still negative in the Continental actions (Panel B). The bivariate association and Mediterranean country cluster and the between unemployment benefits and unem- Eastern European country cluster, but it is ployment is negative in all four clusters. weaker in both cases. However, the association is strongest in the Overall, Figures 1 and 2 lend support to Anglo-Saxon and Nordic countries and rather the job-search-subsidy perspective. The anal- weak in Eastern European countries. The ysis also provides some initial indications that Continental and Mediterranean countries lie the relationship between welfare benefits and between these clusters. unemployment differs between regime con- Similarly, Figure 2 displays a negative texts. However, it also reveals that a simple association between changes in minimum grouping of countries according to typical income benefits and changes in unemploy- regime clusters is not sufficient (see also ment (Panel A). The changes in minimum Wulfgramm and Fervers 2015). There is income benefits that apply here range from a strong within-cluster variation, which indi- reduction of around 20 percentage points in cates that we need to go beyond the level of Sweden, the Czech Republic, and to regimes. Another reason for going beyond 1050 American Sociological Review 82(5) regime clusters is that the static regime typol- deviation). This implies that job-search-sub- ogies they are based on disregard institutional sidy effects might be stronger than disincen- change. Thelen (2014), for instance, points to tives. However, there is no significant diverging paths regarding dualization in two association between minimum income bene- Nordic countries, Sweden and Denmark, and fits and unemployment. two Continental European countries, Ger- Looking at institutions that determine the many and the Netherlands. The following insider/outsider divide, the EPL ratio shows a fixed-effects analyses model distinct institu- small but significant negative association tional interactions that account for changing with unemployment (.3 percentage points per contexts. standard deviation). The comparatively strong associations between unemployment and unionization and centralization point in oppo- Fixed-Effects Analyses site directions (around 3 percentage points Table 1 summarizes the results of five fixed- per standard deviation). The negative coeffi- effects regression models of unemployment cient for unionization is not what would be on welfare benefits, institutions determining predicted by the orthodox view of labor mar- the insider/outsider divide, and control vari- ket rigidities and their impact on unemploy- ables. Model 1 includes all the main effects, ment. However, given the very mixed and Model 2 introduces the two-way interac- evidence of this literature, it is not entirely tions between unemployment benefits and the unprecedented (e.g., Belot and van Ours EPL ratio, unionization, and centralization. 2004).25 Moreover, as the following models Model 3 adds the three-way interaction will show, it is questionable whether mode- between unemployment benefits, unioniza- ling the impact of single institutions can tion, and centralization. Model 4 mirrors clarify how the institutional context might Model 2 and interacts minimum income ben- influence unemployment. The coefficients of efits with the EPL ratio, unionization, and the control variables are mostly in line with centralization. Finally, Model 5 adds the three- theoretical expectations. Expenditure on way interaction between minimum income ALMP is associated with lower unemploy- benefits, unionization, and centralization. ment and so is higher spending on childcare. Hence, Models 2 and 4 test the regulation Neither labor taxes nor bargaining coverage hypothesis, and Models 3 and 5 test the con- have significant coefficients. The significant figuration hypothesis. I use standardized coef- negative coefficient of the output gap indi- ficients, so the main effects show the cates that positive is association at the mean of the variable and associated with lower unemployment. Finally, their respective interaction variables. Further- there is a strong association between marital more, in the models with three-way interac- status and unemployment: the higher the rate tions, the two-way interactions express the of married individuals within a sociodemo- association at the mean of the components of graphic group, the lower its unemployment the three-way interaction. The coefficients can rate. Throughout the different specifications, be interpreted as the change in unemployment some of the control variables’ coefficients rates associated with a one standard deviation change. Because these changes do not change in the respective variable. The multi- strongly relate to the study’s central concerns, tude of interaction effects makes it more dif- I will focus strictly on the main variables of ficult to interpret the models; hence, I will interest and their interactions. display the main findings in graphical form. As noted earlier, I compute models that The baseline model (Model 1) shows a introduce interaction terms to test the hypoth- statistically significant negative association eses about the moderating effect of the insti- between unemployment benefits and the tutional insider/outsider divide. Models 2 unemployment rates of sociodemographic and 3 focus on unemployment benefits. groups (–1.1 percentage points per standard Model 2 includes the two-way interactions of Biegert 1051

Table 1. Fixed-Effects Regressions of Unemployment on Welfare Benefits and Their Interactions with the Institutional Insider/Outsider Divide

Model 1 Model 2 Model 3 Model 4 Model 5

Unemployment Benefits –.011** –.010** –.005 –.010** –.008* (.004) (.003) (.003) (.003) (.003) Unempl. Benefits × EPL ratio .008*** .009*** (.002) (.002) Unempl. Benefits × Unionization .010*** .015*** (.002) (.002) Unempl. Benefits × Centralization .002 .006* (.003) (.003) Unempl. Ben. × Union. × Cent. –.021*** (.003) Minimum Income Benefits .003 .003 .004 –.002 .007 (.003) (.003) (.003) (.003) (.004) Min. Inc. Ben. × EPL ratio .010*** .006*** (.002) (.002) Min. Inc. Ben. × Unionization .010*** .015*** (.003) (.003) Min. Inc. Ben. × Centralization .023*** .020*** (.004) (.004) Min. Inc. Ben. × Union. × Cent. –.023*** (.005) Union. × Centralization –.029*** –.021*** (.004) (.004) EPL ratio –.003* .008*** .007*** .000 –.001 (.001) (.002) (.002) (.001) (.001) Unionization –.032** –.032*** –.013 –.050*** –.045*** (.010) (.009) (.009) (.010) (.011) Centralization .028*** .019*** .021*** .006 .017*** (.005) (.005) (.005) (.005) (.005) Coverage –.001 .016** .015* .010 –.002 (.006) (.006) (.006) (.006) (.006) ALMP –.014*** –.016*** –.017*** –.013*** –.012*** (.002) (.002) (.002) (.001) (.001) Labor Taxes .003 .003 .003 .008** .009** (.003) (.003) (.003) (.003) (.003) Childcare –.009*** –.014*** –.013*** –.012*** –.005* (.002) (.002) (.003) (.002) (.002) Output Gap –.018*** –.015*** –.014*** –.016*** –.016*** (.001) (.001) (.001) (.001) (.001) Marital Status –.040*** –.040*** –.041*** –.040*** –.041*** (.004) (.003) (.003) (.003) (.003) Wave Dummies yes yes yes yes yes N 1,579 1,579 1,579 1,579 1,579 Observations 18,266 18,266 18,266 18,266 18,266 R2 (within) .335 .340 .348 .348 .355

Note: Coefficients and (panel robust standard errors) from OLS fixed-effects regressions. Constants not shown. *p < .05; **p < .01; ***p < .001 (two-tailed tests). unemployment benefits with the EPL ratio, The significant coefficients for the interaction unionization, and wage bargaining centralization. with the EPL ratio and unionization support 1052 American Sociological Review 82(5)

Figure 3. Marginal Effects of Unemployment Benefits on Unemployment Rates across Levels of EPL Ratio, Unionization, and Centralization Note: Marginal effects computed on the basis of Model 3 (see Table 1). The range of computed marginal effects is determined by empirical levels of EPL, unionization, and centralization.

the regulation hypothesis; they indicate that benefits on unemployment rates across the the association between unemployment ben- levels of the EPL ratio, unionization, and efits and unemployment becomes more posi- wage bargaining centralization based on tive as the disparity in the EPL ratio and Model 3. The x-axis is restricted to the empir- unionization increases. In Model 3, the two- ical values of the three institutional arrange- way interaction between unemployment ben- ments. Figure 3 shows that increasing efits and centralization is also significant and unemployment benefits by one standard devi- positive. However, the three-way interaction ation is associated with a reduction in unem- of unemployment benefits, unionization, and ployment of almost 3 percentage points in a centralization is significant and negative. In context of very low EPL (Panel A). This sum with the two-way interactions and the association comes closer to 0 as the EPL ratio main effects, this indicates that the associa- rises. It only becomes significantly positive tion between unemployment benefits and when the EPL ratio is extremely high; at such unemployment does not rise with unioniza- levels, a one standard deviation increase in tion when centralization is high. Thus, while unemployment benefits is associated with Model 2 lends support to the regulation about a 1 percentage point increase in unem- hypothesis, Model 3 indicates that the con- ployment. Panel B shows that unionization figuration hypothesis is more precise by and centralization are complementary mod- showing that the moderating effect of the erators. When centralization is low (defined wage bargaining process depends on its spe- as one standard deviation below the mean), cific constellation. lower unionization leads to a more negative To illustrate these results, Figure 3 dis- association between unemployment benefits plays the marginal effects of unemployment and unemployment (up to around –5 percentage Biegert 1053 points per standard deviation). In the same percentage points. By contrast, in a context of context, higher unionization leads to a more high centralization, the association is initially positive association between unemployment positive (at about 4 percentage points per benefits and unemployment (up to about 6 standard deviation) but gets slightly smaller percentage points per standard deviation). Yet and ultimately becomes insignificant as when centralization is high (defined as one unionization rises. The findings for minimum standard deviation above the mean), unioni- income benefits differ from those for unem- zation does not moderate the association ployment benefits; there is a positive associa- between unemployment benefits and unem- tion between minimum income benefits and ployment rates. In such cases, the coefficient unemployment for all but the very highest never significantly differs from 0, indicating levels of unionization when centralization is no association between unemployment bene- high. The results still support the configura- fits and unemployment. tion hypothesis, because minimum income Models 4 and 5 focus on minimum income benefits are associated with lower unemploy- benefits. Model 4 shows significant positive ment when both unionization and centraliza- coefficients for the interactions between min- tion are high compared to when just one of imum income benefits and the EPL ratio, these factors is high. But in absolute terms, unionization, and wage bargaining centraliza- minimum income benefits are still positively tion. This supports the regulation hypothesis. associated with unemployment except in Yet again, the significant negative coefficient extremely unionized cases. for the three-way interaction between mini- In summary, the findings support the prop- mum income benefits, unionization, and cen- osition that the effect of welfare benefits on tralization added in Model 5 confirms that the employment is moderated by the institutional configuration hypothesis is more precise. insider/outsider divide. Specifically, Models Figure 4 displays the marginal effects of 2 and 4 show that the association between minimum income benefits across levels of welfare benefits and unemployment gets EPL ratio, unionization, and centralization worse as regulation increases, which seem- based on Model 5. The pattern here is similar ingly confirms the regulation hypothesis. Yet to the pattern for unemployment benefits. As in accordance with the configuration hypoth- the EPL ratio increases, the association esis, Models 3 and 5 qualify these findings. between minimum income benefits and unem- The insider/outsider divide does not simply ployment goes from being negative to posi- increase with higher levels of regulation. tive. Whereas minimum income benefits show Instead, the specific constellation of institu- a negative association with unemployment at tions determines the institutional insider/out- extremely low levels of the EPL ratio, of up to sider divide and thus moderates the association about –1.5 percentage points per standard between benefits and unemployment. deviation, the association is positive at extremely high levels of the EPL ratio, at Stylized Regimes, Country around 1.5 percentage points. Again, Panel B Case Examples, and Long-Term shows that the moderating influence of union- ization and centralization depends on the Associations respective level of the other institution. The To make these results more concrete, I calcu- association between minimum income bene- late predictive figures of the association fits and unemployment becomes more posi- between the respective benefit and unemploy- tive for increasing levels of unionization in a ment for four stylized regime types based on context of low wage bargaining centralization. Models 3 and 5 (Table 1). The stylized Here, a low level of unionization leads to a regimes are based on configurations of the negative association of up to –6 percentage three institutions that determine the insider/ points, whereas a high level of unionization outsider divide: the EPL ratio, unionization, can lead to an association of up to around 7 and centralization. A “low” value is one 1054 American Sociological Review 82(5)

Figure 4. Marginal Effects of Minimum Income Benefits on Unemployment Rates across Levels of EPL Ratio, Unionization, and Centralization Note: Marginal effects computed on the basis of Model 5 (see Table 1). The range of computed marginal effects is determined by empirical levels of EPL, unionization, and centralization. standard deviation below the mean. A “high” 0, the figures illustrate Models 3 and 5 in value is one standard deviation above the Table 1 for the stylized regimes. At t = 1 to 5, mean. These figures further illustrate the the figures show the association between a main findings and also explore the associa- change in benefits and a change in unemploy- tion between a change in welfare benefits and ment one to five years onward.26 the development of unemployment in the Figure 5 shows unemployment benefits long-term. They allow me to investigate the have by far the most negative association timing implications of the disincentive per- with unemployment when there is a low level spective and the job-search-subsidy perspec- of all three institutions that determine the tive. This is relevant because disincentives to insider/outsider divide. However, as the find- work should arise immediately after benefit ings for the other regime types show, it is the levels increase. By contrast, job-search- way the regulating institutions are configured subsidy effects are long-term in character; that determines whether the association they are based on the assumption that better becomes positive. In a regime with a low EPL job matches and the improved employability ratio, in which the wage bargaining process of jobseekers will ultimately lead to lower combines high unionization with high cen- unemployment (Morel et al. 2012; Pollmann- tralization, the association is still negative. By Schult and Büchel 2005). Figures 5 and 6 contrast, there is a positive association depict results for the stylized regimes and between unemployment benefits and unem- lagged models of benefit changes (modeled ployment in the two regimes that combine a using lagged benefit indicators from 0 to 5 high EPL ratio with either low unionization year lags; full models can be found in Tables and high centralization or high unionization A2 and A3 in the online supplement). At t = and low centralization. Biegert 1055

Figure 5. Lagged Marginal Effects (up to five years) of Unemployment Benefits on Unemployment Rates across Configurations of EPL Ratio, Unionization, and Centralization Note: Marginal effects computed on the basis of Model 3 (see Table 1) including lagged indicators of unemployment benefits up to five years (see Table A2 in the online supplement). Low EPL ratio, unionization, centralization = mean – standard deviation. High EPL ratio, unionization, centralization = mean + standard deviation.

Figure 6 shows the associations between The most prominent country with a low minimum income benefits and unemploy- level of unionization and an above-average ment for the four stylized regimes. The main level of centralization is Germany. After difference between the findings on minimum experiencing historically high levels of unem- income benefit and those on unemployment ployment up to 2005, Germany saw its unem- benefit is that raising minimum income ben- ployment drop in the later 2000s. Looking at efits is associated with higher unemployment benefits, at first glance neither unemployment in three of the four stylized regimes instead of benefits nor minimum income benefits changed two. In a regime with a low EPL ratio, high much in this period. However, reforms between unionization, and high centralization, there is 2003 and 2005 shifted large sections of the a lower positive association between mini- unemployed from more generous unemploy- mum income benefits and unemployment ment benefits to comparatively lower mini- than is found in regimes with either high mum income benefits. The reforms also linked unionization and low centralization or low benefits more closely to active job-search unionization and high centralization. Yet, as efforts, which were increasingly strictly moni- discussed when presenting Figure 4, the asso- tored. At the same time, the reforms made it ciation does not become negative.27 easier for employers to create atypical jobs; for In the stylized regimes, the association instance, newly established firms could use between benefits and unemployment differs fixed-term contracts for up to four years with- strongly according to the configuration of labor out having to provide a valid reason, and market institutions and the extent to which they employers could now create “mini-jobs,” that divide insiders and outsiders. Can we observe is, jobs with low hours, low wages, and no these patterns in real-typical country cases? benefits. This period saw a steep increase in the 1056 American Sociological Review 82(5)

Figure 6. Lagged Marginal Effects (up to five years) of Minimum Income Benefits on Unemployment Rates across Configurations of EPL Ratio, Unionization, and Centralization Note: Marginal effects computed on the basis of Model 5 (see Table 1) including lagged indicators of minimum income benefits up to five years (see Table A3 in the online supplement). Low EPL ratio, unionization, centralization = mean – standard deviation. High EPL ratio, unionization, centralization = mean + standard deviation.

EPL ratio. The subsequent decrease in unem- Kingdom fit this constellation in the early ployment is often described as a result of 1990s, as did Italy and Slovenia in some years. expansion in atypical employment. This devel- Finland is the only country that was consist- opment is in line with this study’s findings: in ently in this category. As the models suggest, a the comparatively dualized German labor mar- steady retrenchment of unemployment benefits ket, benefit retrenchment was associated with and minimum income benefits throughout the lower unemployment as individuals were 1990s and 2000s accompanied a consistent forced to take jobs they would perhaps have decline in unemployment in Finland. declined in another context. The Nordic countries—Denmark, , Some countries, such as Ireland and Aus- and Sweden—are well known for their com- tria, became more similar to Germany in their bination of high unionization and centraliza- institutional configuration in the later 2000s tion. This configuration was also present in (mostly due to declining unionization), Belgium, Austria, and Ireland until these whereas others, like Slovakia, had combined countries became more similar to Germany in an above-average EPL ratio, lower unioniza- the later 2000s. All Nordic countries have cut tion, and relatively high centralization for a back their very generous welfare states to longer time. In line with the models, unem- some degree since the 1990s. Among them, ployment rose steeply in Slovakia until 2005 Sweden is an interesting case because it not and then decreased again when minimum only cut benefits more substantially than did income benefits were reduced. Denmark or Norway, but it also developed a The combination of above-average unioni- more dualized labor market. Unionization has zation and low centralization is rare. The United declined from above 85 percent in the 1990s Biegert 1057 to below 70 percent in the later 2000s. Most not have harmed the development of unem- dramatically, as part of a reform package in ployment in the United States. According to 2006 that significantly lowered unemploy- the models, the quick recovery in U.S. unem- ment benefits from 72.5 percent to 62 percent ployment after the crisis was aided rather than of former wages, wage bargaining was hindered by the temporary increase in the strongly decentralized and the use of fixed- generosity of the Emergency Unemployment term contracts was made easier (resulting in a Compensation measure. The pattern observed higher EPL ratio). After a severe economic in Switzerland, a country that also has low crisis, Swedish unemployment levels consist- levels of regulation in respect to the three ently decreased until the early 2000s, when institutions, fits the predictions better. Unem- they substantially rose again, which prompted ployment benefits did not change substan- the reforms. The changes were indeed accom- tially, but there was a steady decrease in panied by a subsequent decrease in unem- minimum income benefits in relation to aver- ployment, possibly because lowering benefits age wages. At the same time, unemployment works well in an increasingly dualized labor increased from very low levels of between 2 market with more atypical jobs. However, the and 4 percent in the 1990s to up to 6 percent results of this study suggest that Sweden has in the 2000s. turned its back on a model that could have As a last step, examining the lagged effects been equally successful in lowering unem- in Figure 5, it becomes evident that the asso- ployment but without increasing labor market ciation between a change in unemployment inequality. benefits and unemployment tends toward zero Finally, the United States is the most obvi- in all regime types. In some cases they are not ous country that combines low levels of all significantly different from zero after some three institutions that determine the insider/ years. Similarly, Figure 6 shows that the coef- outsider divide. Hence, we would expect ficients for minimum income benefits decrease increasing benefits to be associated with over time. Yet, they remain significantly dif- lower unemployment. Unemployment bene- ferent from zero except for the regime type fits only changed from 58.5 to 54.5 percent in with a high EPL ratio, high unionization, and the observation period, but minimum income low centralization. Because the associations benefits dropped from around 22 percent to tend to be strongest immediately after a about 15 percent of average wages. U.S. change in benefits, and because there are both unemployment rates dropped throughout the negative and positive associations, the find- 1990s, then briefly spiked in the early 2000s ings for the lagged models do not strongly before quickly decreasing; they only rose support either the disincentive perspective or again during the economic crisis. The model the job-search-subsidy perspective. In regime predictions are thus not completely borne out types with a high EPL ratio and either low in the United States. This is probably because unionization and high centralization or high the models do not aim to fully predict unem- unionization and low centralization, the posi- ployment but to estimate the association tive association diminishes over time, which between benefits and unemployment while we could interpret as an initial disincentive adjusting for potential confounding factors. effect that is outweighed by job-search-sub- Because some important factors that influ- sidy effects in the long run. This seems a ence trends in unemployment are not included stretch, however, as the associations come in the models, and because the ceteris paribus closer to zero in all four regime types. Instead, conditions of the multivariate analysis are not the change over time might be due to institu- met, it is possible for unemployment to tional changes that the models cannot take increase although the specifications of the into account. For instance, there may be addi- models would predict a decline. However, the tional benefits reform after some years, thus implication is that increasing benefits would biasing the results for all subsequent years. 1058 American Sociological Review 82(5)

Moreover, the time restrictions of the dataset groups. The empirical findings corroborate cause the case numbers to go down with each the main proposition of the article: the job- added lag. search-subsidy function of generous benefits Weighing the evidence, the findings indi- can indeed outweigh potential disincentives cate that the institutional context influences to work, lowering unemployment at the whether disincentive effects outweigh job- aggregate level. However, for job-search sub- search-subsidy effects or vice versa. The sidies to come into effect, the labor market empirical evidence suggests that for job- needs to offer jobseekers attractive opportuni- search subsidies to come into effect, the labor ties. The labor market’s capacity to provide market needs to offer jobseekers attractive this type of labor demand depends on the opportunities. When a large institutional positional advantages labor market insiders insider/outsider divide reduces quality job receive from the institutional context. When a opportunities, generous benefits are a disin- large institutional insider/outsider divide centive to job search. There is no evidence diminishes quality job opportunities, the dis- that the strong job protection and high wages incentive effects of generous benefits seem to enjoyed by insiders serve as an added incen- prevail. Within such institutional contexts, tive to look for these positions in countries meager benefits force jobseekers into employ- with a strong insider/outsider divide, so that ment, possibly accepting job offers they increased job-search efforts might ultimately would otherwise decline. make up for the lower job opportunities. In The study makes three distinct contribu- fact, only meager benefits force jobseekers to tions to the literature. First, it establishes the take job offers they would otherwise decline. institutional insider/outsider divide as a mod- erator of the relationship between benefits and unemployment. In a broader perspective, Discussion this supports scholarship that points out the This study investigated the effect of welfare existence of institutional interdependencies in benefits on unemployment. The dominant national economies (e.g., Amable 2003; Hall disincentive perspective contends that gener- and Soskice 2001; Iversen 2005). Second, by ous benefits act as a disincentive to work and testing the hypothesis with indicators for increase unemployment. By contrast, propo- unemployment benefits and minimum income nents of the job-search-subsidy perspective benefits, the analysis extends beyond prior argue that generous benefits lead to better job work’s focus on unemployment insurance. I matches and lower unemployment levels. find very similar patterns for the relationship This article showed that the relationship between both benefits and unemployment, depends on the institutional context. I demon- which suggests their impact on job-search strate that the disparity between EPL for per- processes follows similar mechanisms. Third, manent and temporary contract holders, and by using pseudo-panels, the study illustrates a the configuration of the wage bargaining rarely used modeling technique. Comparative process in terms of unionization and central- micro-level datasets are rarely in panel form, ization, determine the institutional insider/ so applying this method in future studies and outsider divide and thereby moderate how revisiting existing evidence might prove use- unemployment benefits and minimum income ful for comparative research. benefits affect unemployment. To test this Two important matters could not be tack- proposition, the analysis combined data from led within the scope of this study. First, the 20 European countries in the EULFS and analysis relied on a binary distinction between from the United States CPS from 1992 to employment and unemployment. Distinguish- 2009. I transformed the repeated cross- ing between different types of jobs might sectional data into pseudo-panels to use fixed provide further insights on the effect of ben- effects at the level of sociodemographic efits on labor markets (Kalleberg 2011). Biegert 1059

Previous studies indicate that in labor markets liberalized labor market. The results of this with a large insider/outsider divide, jobseek- study suggest that implementing such strate- ers are not only more likely to remain out of gies could help achieve low unemployment employment, they are also more likely to while also providing high levels of social enter atypical jobs (Biegert 2014; Fervers and security in times of joblessness. Schwander 2015; Schwander and Häuser- Second, neither retrenching benefits nor mann 2013). On the other hand, in several deregulating labor markets are necessarily countries, workers in low-wage jobs can successful strategies to tackle unemployment. receive welfare benefits. Hence, in certain Instead, the study raises an additional ques- contexts, benefits could subsidize low-wage tion, namely how to break down barriers jobs rather than providing job-search subsi- between insiders and outsiders while retaining dies.28 Future studies that explore how bene- the benefits of institutions that improve the fits and labor market institutions relate to the positions of insiders. For instance, we know quality of jobs could complement the present that strong unions are crucial for better work- analysis. Second, although the findings sug- ing conditions and low wage inequality (Brady gest that unemployment benefits and mini- et al. 2013; Freeman and Medoff 1984; West- mum income benefits work via similar ern and Rosenfeld 2011). The results imply mechanisms, the emerging differences that the insider/outsider divide does not sim- between the two warrant further study. This is ply move along a spectrum between flexibility especially relevant because the dualization and regulation, but in some configurations literature argues that exclusive benefits, such coordination does not inhibit the positive as unemployment insurance, can themselves effects of benefits for employment. The find- be an integral part of an insider/outsider labor ings point to constellations that combine high market (Emmenegger et al. 2012; Palier and levels of economic security with better work- Thelen 2010). ing conditions and high and equal wages, The present study adds a crucial dimension without excluding sections of the population to the existing research on the relationship from gainful employment. between the welfare state and unemployment. It establishes that lower levels of the institu- Acknowledgments tional insider/outsider divide make it possible I am very thankful to David Brady for extensive feedback for generous welfare benefits to have a posi- and way too much time and support. I am grateful to tive impact on unemployment. The results Giorgio Cutuli, Bernhard Ebbinghaus, Lena Hipp, David suggest that individuals are willing to work Howell, Gregory Jackson, Olli Kangas, the audiences at despite monetary incentives as long as there the ESPAnet conference in Poznan 2013, the ECSR con- ference in Berlin 2014, the BIGSSS conference in Bre- are attractive job opportunities. The findings men 2014, and the USP writing workshop for fruitful have two important policy implications. First, debates and inspiring suggestions. Finally, I thank the they tie in with the claims of the anonymous reviewers and ASR editors for many helpful literature (see Kalleberg 2011; Viebrock and comments and suggestions. Clasen 2008; Wilthagen and Tros 2004). This literature highlights the beneficial interaction Funding between a generous welfare state and a flexi- This research was partially funded by the German ble labor market. Proponents argue that such Research Foundation (Grant Number: EB 434/2-1). institutional constellations enable high levels of employment security, especially when Notes combined with ALMP. Similarly, Thelen 1. In many countries, benefit levels are reduced (2014) describes an “embedded flexibiliza- after a certain amount of time. The job-search- subsidy argument focuses on initial levels of ben- tion,” in which generous benefits are neces- efits for jobseekers, highlighting their effect on the sary to create employment security and to job-search period immediately after job loss (e.g., collectivize the social risk of job loss in a Nelson and Stephens 2012). Long-term unemployed 1060 American Sociological Review 82(5)

individuals might be discouraged or not employable. albeit modestly less significant. In another robust- Because this affects the intensity of their job search, ness check (available from the author), I ran the job-search-subsidy effects are less likely. models using a multilevel specification that also 2. There are different definitions of insiders and outsid- confirmed the findings (for a similar approach, see ers in the literature. Early iterations of insider/outsider Jæger 2013). theory used a snapshot perspective on the labor force 8. Rather than modeling national unemployment to define who is an insider and who is an outsider. The more accurately, group-level unemployment rates employed are considered insiders and the unemployed approximate individual-level unemployment. In the are considered outsiders (Lindbeck and Snower absence of real individual panel data, this is a pref- 1988). Most of the dualization literature has modified erable way of modeling the proposed macro-micro this definition by including temporary workers in the mechanisms. When I aggregate the unemployment group of outsiders (e.g., Rueda 2005). Thus, employed rates of the various sociodemographic groups at the individuals can also be outsiders, but they have a country-year level, they are still highly correlated higher propensity of future unemployment. Some con- (.89) with the official numbers from the OECD, tributions to the dualization literature use a life-course which confirms the validity of the data used here. perspective (e.g., Schwander and Häusermann 2013). 9. I also use the data structure to run a robustness Outsiders are understood as individuals who are more check on the subsample of low-educated individu- likely to experience unemployment and precarious als to see whether disincentive effects are stronger work situations over their careers. According to this among individuals with lower wage expectations definition, unemployed individuals could be insiders (see Table B8 and Figure B2 [“Low-Educated Sam- if they only experience a short spell of unemployment ple”] in the online supplement). between permanent jobs. The life-course definition is 10. Because of the restriction to working-age (15 to helpful, for instance, when investigating the policy 64) individuals, 1928 is the first year an individual preferences of insiders and outsiders. Here, I am inter- could be born and still enter the sample in 1992. ested in how benefits affect the unemployed, that is, 1994 is the last year an individual could be born and individuals who are outsiders at this very moment, enter the sample before the end of 2009. which is why the snapshot definition is more appro- 11. I selected the countries according to the availabil- priate. More importantly, differences between the ity of micro data and macro indicators. These are snapshot definition and the life-course definition do the countries in alphabetical order: Austria, Bel- not directly affect the core of the proposed theoretical gium, Czech Republic, Denmark, Estonia, Fin- argument, because it is not about complete congruence land, France, Germany, Hungary, Ireland, Italy, the between the unemployed and outsiders but about how Netherlands, Norway, Portugal, Slovakia, Slovenia, benefits affect jobseekers’ search process and how the Spain, Sweden, Switzerland, United Kingdom, and insider/outsider divide affects the availability of attrac- the United States. Data collection for France starts tive jobs for jobseekers. in 1993; for Austria, Finland, and Sweden in 1995; 3. Iversen (1998) convincingly argues that the rela- for the Netherlands, Norway, and Switzerland in tionship between unions, the wage bargaining 1996; for Hungary in 1997; for Slovakia in 1998; system, and economic outcomes such as unemploy- for Estonia in 2003; and for Slovenia in 2004. Data ment depends on the given monetary policy regime. from 1998 are dropped for Ireland and the United To test whether monetary policy affects the rela- Kingdom due to missing information on education. tionships proposed here, I ran a robustness check 12. I ran robustness checks in which I coded only respon- adjusting for central independence (see Table dents who worked at least 35 hours a week or more B7 and Figure B2 [“Incl. Central Bank Indep.”] in as being employed. This alternative model reveals the online supplement). Inclusion of this indicator some differences from the main analysis, which I does not yield substantively different results. discuss in the online supplement (see also Table B9 4. Depending on the macro-economic context, benefi- and Figure B2 [“<35 Hours + Unemployed”]). cial outcomes for insiders could include alternatives 13. The definition of the dependent variable excludes to wage increases such as higher job security. Either jobless individuals who are not actively seeking a way, these bargaining outcomes will decrease qual- job, that is, the inactive. Robustness checks using the ity job opportunities for jobseekers. nonemployment rate and the inactivity rate as depen- 5. The code for the dataset and analyses is available on dent variables yield results that, in line with the theo- the author’s webpage. retical expectations, imply that inactive individuals 6. As a robustness check, I conducted the analysis are less affected by benefits and labor market institu- using random-effects models, which use between tions than are the unemployed (see Tables B10 and variation as well. The results do not substantively B11, Figure B3 [“Inactive” and “Nonemployed”], differ (see Table B12 and Figure B3 [“Random and the discussion in the online supplement). Effects Models”] in the online supplement). 14. To increase case numbers, we could think of any 7. Clustering standard errors at the country or coun- number of further time-constant determinants of try-year level, the main findings remain unchanged, cohort membership or use a more fine-grained Biegert 1061

grouping of birth cohorts. Some possibilities, such 22. The figures use the overall difference in the indica- as race or place of birth, are not available for the tors between two time points: 1992 and 2009. Thus, full sample. More importantly, this would reduce the figures do not show the substantial variance robustness in the measurement of aggregated time- in the years between these two time points within varying cohort-level variables, such as the unem- countries (see Table A1 in the online supplement). ployment rate. 23. In 2009, the unemployment rate in many countries 15. Because some countries consistently have larger was already affected by the global economic crisis. case numbers than others, this procedure might Using other time frames produces different change increase their weight in the analysis and thus bias rates. However, the overall pattern for the bivariate the results. I ran the models without the weights and relationships between benefits and unemployment did not find meaningful differences (see Table B13 rates is robust. and Figure B3 [“Without Weights”] in the online 24. Judging from the figures, there might be single supplement). countries strongly driving the association between 16. Other components of unemployment benefits might changes in benefits and changes in unemployment. affect their overall generosity, such as benefit dura- I ran the models for unemployment benefits with- tion, eligibility criteria, and coverage. I ran robust- out Italy and Sweden and the models for minimum ness checks with an indicator that comprehensively income benefits without the Czech Republic, Italy, included these components but was not available and Sweden (see Table B2 and Figure B1 [“With- for the full set of countries (Scruggs, Jahn, and out Potential Outliers”] in the online supplement). Kuitto 2014). I found some differences, which I Excluding these potential outliers did not substan- discuss in the online supplement (see Table B1 and tively change the main findings. Figure B1 [“Unemp. Benefit Generosity”]). 25. The coefficient became insignificant and dramati- 17. I ran robustness checks using the indicator for EPL cally diminished in size when I ran the models for permanent contracts instead of the ratio and con- without wave dummies. This might hint at potential trolling for EPL for temporary contracts. The differ- shortcomings in previous studies, as many did not ences in the results indicate that the disparity in EPL adjust for global trends (Belot and van Ours [2004] for permanent contracts and temporary contracts is being an exception). more important for the insider/outsider divide and 26. The lagged independent variable models deal with the effect of benefits on unemployment than the two more concerns. First, politicians might increase overall level of EPL (see Table B6 and Figure B2 welfare benefits to appease a growing population of [“EPL Permanent Contracts”] in the online supple- unemployed people. Hence, there might be reverse ment). causality. Second, because measurement on the micro 18. Centralization is a concept closely related to cor- level is spread over the whole year, information on poratism (although the two are not interchangeable, individuals’ labor market status might stem from see Calmfors and Driffill 1988). I ran robustness a time point before policy changes took place. The checks using an indicator for instead of order of events might thus be corrupted in some cases. centralization (Jahn 2014). I found only minor dif- 27. Additional analyses investigating potential outliers ferences, which I discuss in the online supplement indicate that this prediction changes and becomes (see Table B5 and Figure B1 [“Corporatism”]). more similar to the predictions for unemploy- 19. There are gaps in the time series for coverage. In ment benefits when excluding Sweden, Italy, and order not to lose observations, I interpolate values the Czech Republic (see Table B2 and Figure B1 linearly. [“Without Potential Outliers”] in the online supple- 20. Coverage might moderate unionization effects in ment). a similar way to centralization. Furthermore, some 28. See the discussion of robustness checks using an countries known for their strong insider/outsider alternative dependent variable that only codes indi- divides, such as France, Portugal, and Spain, show viduals with at least 35 work hours as employed high coverage although centralization is rather low. (Table B9 and Figure B2 [“<35 Hours + Unempl.”] I ran robustness checks including interactions with in the online supplement). coverage instead of centralization and found slight differences, which I discuss in the online supple- ment (see Table B4 and Figure B1 [“Coverage”]). References 21. Another potentially relevant confounder is the orga- 156 Congressional Record. 2010. Statement of Senator nization of unemployment benefits in a Ghent sys- Jon Kyl speaking for the Tax Extenders Act of 2009 on tem, where unions administer the benefits. I ran a March 1, 2010, to the U.S. Senate, H.R. 4213, 111th robustness check excluding the four Ghent countries Cong., 2nd sess. Congressional Record v. 156, pt. (Belgium, Finland, Denmark, Sweden); it did not 27:845. (https://www.congress.gov/crec/2010/03/01/ yield substantively different results (see Table B3 CREC-2010-03-01-senate.pdf). and Figure B1 [“Without Ghent Countries”] in the Allison, Paul D. 2009. Fixed Effects Regression Models, online supplement). Vol. 160. Los Angeles: Sage. 1062 American Sociological Review 82(5)

Amable, Bruno. 2003. The Diversity of Modern Capital- Card, David, Andrew Johnston, Pauline Leung, Alexandre ism. Oxford, UK: Oxford University Press. Mas, and Zhuan Pei. 2015. “The Effect of Unemploy- Atkinson, Anthony B., and John Micklewright. 1991. ment Benefits on the Duration of Unemployment Insur- “Unemployment Compensation and Labor Market ance Receipt: New Evidence from a Regression Kink Transitions: A Critical Review.” Journal of Economic Design in Missouri, 2003–2013.” American Economic Literature 29(4):1679–1727. Review 105(5):126–30. Avdagic, Sabina, and Paola Salardi. 2013. “Tenuous Calmfors, Lars, and John Driffill. 1988. “Central- Link: Labour Market Institutions and Unemployment ization of Wage Bargaining.” in Advanced and New Market Economies.” Socio- 3(6):13–61. Economic Review 11(4):739–69. Davidsson, Johan Bo, and Patrick Emmenegger. 2013. Baccaro, Lucio, and Diego Rei. 2007. “Institutional “Defending the Organisation, Not the Members: Determinants of Unemployment in OECD Countries: Unions and the Reform of Job Security Legislation Does the Deregulatory View Hold Water?” Interna- in .” European Journal of Political tional Organization 61(3):527–69. Research 52(3):339–63. Baker, Dean, Andrew Glyn, David R. Howell, and John Deaton, Angus. 1985. “Panel Data from Time Series of Schmitt. 2005. “Labor Market Institutions and Unem- Cross-Sections.” Journal of 30(1– ployment: A Critical Assessment of the Cross-Country 2):109–26. Evidence.” Pp. 72–118 in Fighting Unemployment: DiPrete, Thomas A., and Patricia A. McManus. 2000. The Limits of Orthodoxy, edited by D. “Family Change, Employment Transitions, and the Howell. New York: Oxford University Press. Welfare State: Household Income Dynamics in the Barbieri, Paolo. 2009. “Flexible Employment and United States and Germany.” American Sociological Inequality in Europe.” European Sociological Review Review 65(3):343–70. 25(6):621–28. Doeringer, Peter B., and Michael J. Piore. 1971. Internal Barbieri, Paolo, and Giorgio Cutuli. 2016. “Employment Labor Markets and Manpower Analysis. Lexington, Protection Legislation, Labour Market Dualism, and MA: D.C. Heath and Co. Inequality in Europe.” European Sociological Review Eichhorst, W., and P. Marx. 2011. “Reforming German 32(4):501–516. Labour Market Institutions: A Dual Path to Flexibil- Bassanini, Andrea, and Romain Duval. 2009. “Unemploy- ity.” Journal of European 21(1):73–87. ment, Institutions, and Reform Complementarities: Emmenegger, Patrick, Silja Häusermann, Bruno Palier, Re-Assessing the Aggregate Evidence for OECD Coun- and Martin Seeleib-Kaiser. 2012. The Age of Dualiza- tries.” Oxford Review of Economic Policy 25(1):40–59. tion: The Changing Face of Inequality in Deindustrial- Belot, Michèle, and Jan C. van Ours. 2004. “Does the izing Societies. Oxford, UK: Oxford University Press. Recent Success of Some OECD Countries in Low- Estevez-Abe, Margarita, Torben Iversen, and David Sos- ering Their Unemployment Rates Lie in the Clever kice. 2001. “ and the Formation of Design of Their Labor Market Reforms?” Oxford Skills: A Reinterpretation of the Welfare State.” Pp. Economic Papers 56(4):621–42. 145–83 in Varieties of Capitalism: The Institutional Bentolila, Samuel, Pierre Cahuc, Juan J. Dolado, and Foundations of Comparative Advantage, edited by Thomas Le Barbanchon. 2012. “Two-Tier Labour P. A. Hall and D. Soskice. Oxford, UK: Oxford Uni- Markets in the Great : France versus versity Press. Spain.” The Economic Journal 122(562):F155–F187. Fervers, Lukas, and Hanna Schwander. 2015. “Are Bertola, Giuseppe, Francine D. Blau, and Lawrence M. Outsiders Equally Out Everywhere? The Economic Kahn. 2007. “Labor Market Institutions and Demo- Disadvantage of Outsiders in Cross-National Per- graphic Employment Patterns.” Journal of Popula- spective.” European Journal of Industrial Relations tion Economics 20(4):833–67. 21(4):369–87. Biegert, Thomas. 2014. “On the Outside Looking In? Flood, Sarah, Miriam King, Steven Ruggles, and J. Rob- Transitions out of Non-Employment in the United ert Warren. 2015. Integrated Public Use Microdata Kingdom and Germany.” Journal of European Social Series, Current Population Survey: Version 4.0. Policy 24(1):3–18. [Machine-readable database]. Minneapolis: Univer- Blyth, Mark. 2002. Great Transformations: Economic sity of Minnesota. Ideas and Institutional Change in the Twentieth Cen- Freeman, Richard B., and James L. Medoff. 1984. What tury. Cambridge, UK: Cambridge University Press. Do Unions Do? New York: Basic. Bonoli, Giuliano. 2010. “The of Gangl, Markus. 2003. “The Only Way Is Up? Employ- Active Labor-Market Policy.” & Society ment Protection and Job Mobility among Recent 38(4):435–57. Entrants to European Labour Markets.” European Brady, David, Regina S. Baker, and Ryan Finnigan. Sociological Review 19(5):429–49. 2013. “When Unionization Disappears: State- Gangl, Markus. 2004. “Welfare States and the Scar Level Unionization and Working in the Effects of Unemployment: A Comparative Analysis United States.” American Sociological Review of the United States and .” American 78(5):872–96. Journal of 109(6):1319–64. Biegert 1063

Gangl, Markus. 2006. “Scar Effects of Unemployment: Iversen, Torben. 1999. Contested Economic Institutions: An Assessment of Institutional Complementarities.” The Politics of Macroeconomics and Wage Bargain- American Sociological Review 71(6):986–1013. ing in Advanced Democracies. Cambridge, UK: Gebel, Michael, and Johannes Giesecke. 2011. “Labor Cambridge University Press. Market Flexibility and Inequality: The Changing Iversen, Torben. 2005. Capitalism, Democracy, and Wel- Skill-Based Temporary Employment and Unemploy- fare. Cambridge, UK: Cambridge University Press. ment Risks in Europe.” Social Forces 90(1):17–40. Jacobs, David, and Lindsey Myers. 2014. “Union Gebel, Michael, and Johannes Giesecke. 2016. “Does Strength, , and Inequality: Contingent Help? The Impact of Employment Political Analyses of U.S. Income Differences since Protection Reforms on Youths’ Unemployment and 1950.” American Sociological Review 79(4):752–74. Temporary Employment Risks in Europe.” European Jæger, Mads Meier. 2013. “The Effect of Macroeconomic Sociological Review 32(4):486–500. and Social Conditions on the Demand for Redistri- Goldthorpe, John H. 1984. “The End of Convergence: bution: A Pseudo Panel Approach.” Journal of Euro- Corporatist and Dualist Tendencies in Modern West- pean Social Policy 23(2):149–63. ern Societies.” Pp. 315–43 in Order and Conflict Jahn, Detlef. 2014. “Changing of the Guard: Trends in in Contemporary Capitalism, edited by J. H. Gold- Corporatist Arrangements in 42 Highly Industrial- thorpe. Oxford, UK: Clarendon Press. ized Societies from 1960 to 2010.” Socio-Economic Gornick, Janet C., Marcia K. Meyers, and Katherin E. Review 14(1):47–71. Ross. 1997. “Supporting the Employment of Moth- Kalleberg, Arne L. 2011. Good Jobs, Bad Jobs: The Rise ers: Policy Variation across Fourteen Welfare States.” of Polarized and Precarious Employment Systems in Journal of European Social Policy 7(1):45–70. the United States, 1970s to 2000s. New York: Russell Hall, Peter A., and David Soskice. 2001. “An Introduc- Sage Foundation. tion to Varieties of Capitalism.” Pp. 1–68 in Variet- Kalleberg, Arne L., Michael Wallace, and Robert P. ies of Capitalism: The Institutional Foundations of Althauser. 1981. “Economic Segmentation, Worker Comparative Advantage, edited by P. A. Hall and D. Power, and Income Inequality.” American Journal of Soskice. Oxford, UK: Oxford University Press. Sociology 87(3):651–83. Häusermann, Silja, and Hanna Schwander. 2012. “Variet- Lalive, Rafael. 2007. “Unemployment Benefits, Unem- ies of Dualization? Labor Market Segmentation and ployment Duration, and Post-Unemployment Jobs: A Insider-Outsider Divides Across Regimes.” Pp. 27–51 Regression Discontinuity Approach.” American Eco- in The Age of Dualization: The Changing Face of nomic Review 97(2):108–112. Inequality in Deindustrializing Societies, edited by Layard, Richard, Stephen J. Nickell, and Richard Jack- P. Emmenegger, S. Häusermann, B. Palier, and M. man. 2005. Unemployment: Macroeconomic Perfor- Seeleib-Kaiser. Oxford, UK: Oxford University Press. mance and the Labour Market. Oxford, UK: Oxford Hicks, Alexander, and Lane Kenworthy. 1998. “Coopera- University Press. tion and Political Economic Performance in Affluent Lindbeck, Assar, and Denis J. Snower. 1988. The Insider- .” American Journal of Sociol- Outsider Theory of Employment and Unemployment. ogy 103(6):1631–72. Cambridge, MA: MIT Press. Hipp, Lena, and Kathrin Leuze. 2015. “Determinants of Meyer, Bruce D. 1990. “Unemployment Insur- Working Time Differences within Couples in Europe ance and Unemployment Spells.” Econometrica and the U.S.” KZfSS Kölner Zeitschrift für Soziologie 58(4):757–82. und Sozialpsychologie 67(4):659–84. Morel, Nathalie, Bruno Palier, and Joakim Palme. 2012. Howell, David R., and Bert M. Azizoglu. 2011. “Unem- Towards a Social Investment Welfare State? Ideas, ployment Benefits and Work Incentives: The US Policies and Challenges. Bristol, UK: Policy Press. Labour Market in the Great Recession.” Oxford Nelson, Kenneth. 2010. “Social Assistance and Minimum Review of Economic Policy 27(2):221–40. Income Benefits in Old and New EU Democracies.” Howell, David R., and Miriam Rehm. 2009. “Unemploy- International Journal of Social Welfare 19(4):367–78. ment Compensation and High European Unemploy- Nelson, Moira, and John D. Stephens. 2012. “Do Social ment: A Reassessment with New Benefit Indicators.” Investment Policies Produce More and Better Jobs?” Pp. Oxford Review of Economic Policy 25(1):60–93. 205–234 in Towards a Social Investment Welfare State? ILO. 1982. “Resolution Concerning Statistics of the Eco- Ideas, Policies and Challenges, edited by N. Morel, B. nomically Active Population, Employment, Unem- Palier, and J. Palme. Bristol, UK: Policy Press. ployment and Underemployment, Adopted by the Neugebauer, Martin. 2015. “The Introduction of Bach- Thirteenth International Conference of Labour Statis- elor Degrees and the Under-representation of Stu- ticians.” Retrieved February 2, 2016 (http://ec.europa dents from Low Social Origin in Higher Education .eu/eurostat/ramon/coded_files/ecacpop_en.pdf). in Germany: A Pseudo-Panel Approach.” European Iversen, Torben. 1998. “Wage Bargaining, Hard Money Sociological Review 31(5):591–602. and Economic Performance: Theory and Evidence Nickell, Stephen. 1997. “Unemployment and Labor Mar- for Organized Market Economies.” British Journal of ket Rigidities: Europe versus North America.” Jour- Political Science 28(1):31–61. nal of Economic Perspectives 11(3):55–74. 1064 American Sociological Review 82(5)

Noelke, Clemens. 2016. “Employment Protection Leg- van Ours, Jan C., and Milan Vodopivec. 2006. “How islation and the Youth Labour Market.” European Shortening the Potential Duration of Unemployment Sociological Review 32(4):471–85. Benefits Affects the Duration of Unemployment: Evi- OECD. 1994. “The OECD Jobs Study: Facts, Analysis, dence from a Natural Experiment.” Journal of Labor Strategies (1994).” Retrieved November 28, 2011 Economics 24(2):351–78. (http://www.oecd.org/dataoecd/42/51/1941679.pdf). van Vliet, Olaf, and Koen Caminada. 2012. “Unemploy- OECD. 2006. “Reassessing the Role of Policies and Insti- ment Replacement Rates Dataset among 34 Welfare tutions for Labour Market Performance: A Quantita- States, 1971–2009: An Update, Extension and Modi- tive Analysis.” Pp. 207–231 in Employment Outlook, fication of the Scruggs’ Welfare State Entitlements edited by OECD. Paris: OECD. Data Set.” Neujobs Special Report No.2/January Palier, Bruno, and Kathleen Thelen. 2010. “Institutionaliz- 2012. ing Dualism: Complementarities and Change in France Venn, Danielle. 2009. “Legislation, Collective Bargain- and Germany.” Politics & Society 38(1):119–48. ing and Enforcement.” Vol. No. 89. OECD Social, Pfeifer, Michaela. 2012. “Comparing Unemployment Pro- Employment and Migration Working Papers. tection and Social Assistance in 14 European Countries: Verbeek, Marno, and Francis Vella. 2005. “Estimating Four Worlds of Protection for People of Working Age.” Dynamic Models from Repeated Cross-Sections.” International Journal of Social Welfare 21(1):13–25. Journal of Econometrics 127(1):83–102. Pollmann-Schult, Matthias, and Felix Büchel. 2005. Viebrock, Elke, and Jochen Clasen. 2008. “Flexicurity “Unemployment Benefits, Unemployment Duration and : A Review.” Socio-Economic and Subsequent Job Quality: Evidence from West Review 7(2):305–331. Germany.” Acta Sociologica 48(1):21–39. Visser, Jelle. 2013. ICTWSS: Database on Institutional Rueda, David. 2005. “Insider–Outsider Politics in Industrial- Characteristics of Trade Unions, Wage Setting, ized Democracies: The Challenge to Social Democratic State Intervention and Social Pacts in 34 Countries Parties.” American Political Science Review 99(1):61–74. between 1960 and 2012. Amsterdam Institute for Rueda, David. 2014. “Dualization, Crisis and the Welfare Advanced Labour Studies AIAS, Amsterdam. State.” Socio-Economic Review 12(2):381–407. Western, Bruce. 1998. “Institutions and the Labor Mar- Rueda, David, and Jonas Pontusson. 2000. “Wage ket.” Pp. 224–43 in The New Institutionalism in Soci- Inequality and Varieties of Capitalism.” World Poli- ology, edited by M. C. Brinton and V. Nee. New York: tics 52(3):350–83. Russell Sage Foundation. Schmieder, Johannes F., and Till von Waechter. 2016. Western, Bruce, and Jake Rosenfeld. 2011. “Unions, “The Effects of Unemployment Insurance Benefits: Norms, and the Rise in U.S. Wage Inequality.” Ameri- New Evidence and Interpretation.” Annual Review of can Sociological Review 76(4):513–37. Economics 8(1):547–81. Wilthagen, Ton, and Frank Tros. 2004. “The Concept Schwander, Hanna, and Silja Häusermann. 2013. “Who of ‘Flexicurity’: A New Approach to Regulat- Is in and Who Is Out? A Risk-Based Conceptualiza- ing Employment and Labour Markets.” Transfer tion of Insiders and Outsiders.” Journal of European 10(2):166–86. Social Policy 23(3):248–69. Wright, Erik O. 2000. “Working-Class Power, Capitalist- Scruggs, Lyle, Detlef Jahn, and Kati Kuitto. 2014. “Compar- Class Interests, and Class Compromise.” American ative Welfare Entitlements Dataset 2. Version 2014-03.” Journal of Sociology 105(4):957–1002. University of Connecticut & University of Greifswald. Wulfgramm, Melike, and Lukas Fervers. 2015. “Unem- Sjöberg, Ola. 2000. “Unemployment and Unemployment ployment and Subsequent Employment Stability: Benefits in the OECD 1960–1990: An Empirical Test Does Labour Market Policy Matter?” Socio-Eco- of Neo-Classical Economic Theory.” Work, Employ- nomic Review 13(4):791–812. ment & Society 14(1):51–76. Sørensen, Aage B., and Arne L. Kalleberg. 1981. “An Outline of a Theory of the Matching of Persons to Jobs.” Pp. 49–74 in Sociological Perspectives on Thomas Biegert is a post-doctoral research fellow at Labor Markets, edited by I. Berg. New York: Aca- the WZB Berlin Social Science Center. In September demic Press. 2017 he will join the Department of Social Policy at Streeck, Wolfgang. 1992. Social Institutions and Eco- the London School of Economics and Political Science nomic Performance: Studies of Industrial Relations in as a Fellow in Social Policy. He completed his PhD at Advanced Capitalist Economies. London, UK: Sage. the University of Mannheim in 2014. He studies social Thelen, Kathleen. 2004. How Institutions Evolve: The inequality and stratification, labor markets, and wel- Political Economy of Skills in Comparative-Historical fare states, with a strong interest in quantitative meth- Perspective. New York: Cambridge University Press. ods. Other work on insiders and outsiders on the labor Thelen, Kathleen. 2014. Varieties of Liberalization and market authored by him has been published in Journal the New Politics of Social . New York: of European Social Policy and is forthcoming in Cambridge University Press. Socio-Economic Review.