Unemployment reduction or labor force expansion? How partisanship matters for the design of active labor market policy in Europe

Unemployment reduction or labor force expansion? How partisanship matters for the design of... Abstract Comparative scholars fundamentally disagree about the impact of partisan politics in modern welfare states, particularly in certain ‘new’ policy areas such as active labor market policy (ALMP). Using new data on 900 ALMP programs across Europe, this study attempts to reconcile a long-standing dispute between the traditional ‘power resources’ approach and the ‘insider/outsider’ approach pioneered by Rueda. The study argues that both left-wing and right-wing governments invest in ALMP but that politics still matter because parties’ preferences regarding unemployment differ. The left is more inclined to expand programs primarily designed to reduce unemployment, which exclusively target ‘core’ groups in, or at risk of, unemployment, and programs in which participants are no longer counted among the unemployed. In contrast, both sides are equally prone to expand programs that also—or instead—target people who are not yet participating in the labor market, which thus also—or instead—serve to increase labor supply. 1. Introduction It has been widely established that the last 15–20 years have seen an ‘activation turn’ of labor market policy across the OECD. Responding to the generally high unemployment and inactivity rates of the 1980s and early 1990s, governments have introduced a mix of demanding and enabling so-called ‘active labor market policy’ (ALMP) schemes aimed at the unemployed and some groups outside of the labor force (e.g. housewives and single parents) to facilitate their entry into employment. Although evaluations find mixed evidence regarding the effects of ALMPs on employment (Card et al., 2010; Kluve, 2010; Martin, 2014), ALMPs are consistently listed high on international policy agendas—most recently in the Employment Guidelines of the Europe 2020 strategy from 2010 and in the Social Investment Package launched by the European Commission in 2013. These developments have spurred a growing interest among comparative political scientists in the political and economic processes that determine the scale and orientation of ALMPs provided by governments across the EU and the OECD. Several books have been devoted to the policy area (e.g. Eichhorst et al., 2008; Weishaupt, 2011; Bonoli, 2013), and it is placed at the center of attention in most recent volumes that investigate the overall development of modern welfare states (e.g. Morel et al., 2012; Bonoli and Natali, 2012; Hemerijck, 2013; Thelen, 2014). Particular interest has been paid to the role of partisan politics for ALMP development; however, to date, scholars have not come anywhere close to a consensus regarding this matter. Over the past 10 years, studies have alternately reported evidence supporting hypotheses about positive, nonexistent and negative relationships between left-wing strength and the level of ALMP ‘effort’ exerted by the government. This lack of scholarly agreement was recently noted by Clasen et al. (2016, p. 33), who forcefully argued that most comparative research in the field has been fallen prey to conceptual under-specification of the dependent variable and that ‘[t]he process of uncovering the causal dynamics specific to this policy field is still in its infancy’. They suggest that scholars who search for variables to explain variations in ALMP development might want to look beyond the ‘usual suspects’, such as partisanship, and instead (or also) consider some rather different political institutional variables, for instance, those related to the nature of the national systems of public expenditure planning and control and the multi-actor delivery systems that characterize ALMPs. Whereas I largely subscribe to the diagnosis of Clasen et al. (2016) regarding the state of the field, I seek to demonstrate in this article that scholars run the risk of throwing the baby out with the bathwater if they fully adopt their proposed cure. I instead argue that if adequate attention is paid to the various purposes and target groups that ALMP programs potentially serve, our understanding of the impact of partisanship on ALMP design can be refined. Building on classic insights about variation in party preferences with regard to unemployment and labor shortages (Furåker, 1976; Hibbs, 1977), I outline a theory about how party preferences for ALMPs vary across programs that have different aims and thus different target groups and characteristics. On that basis, I derive two hypotheses, which are then tested on a new panel data set with unprecedented richness, covering almost 900 European ALMP programs over up to 15 years (European Commission, 2015). The results of these tests support a two-fold claim. First, whereas parties do not differ in their preferences for ALMP programs that, to some extent, serve to increase labor supply—and that thus also, or exclusively, target people who did not previously participate (at least not fully) in the labor market—the left is more likely than the right to also expand programs that primarily serve to reduce unemployment. These programs exclusively target people at the ‘core’ of the labor force: those who are unemployed and those who are employed but at the risk of losing their jobs. Second, the left is more inclined to expand particular ALMP programs for which participation entails a formal change in the participant’s labor market status—from unemployed to either inactive or employed. Together, these claims contradict a number of prevailing theories about the modern welfare state. First, they are at odds with the traditional ‘power resources’ school’s account of ALMPs (Esping-Andersen, 1990; Janoski, 1990; Rothstein, 1996; Huo et al., 2008) because, for many ALMP programs, right-wing governments are just as expansionary as left-wing governments. Second, they conflict with the ‘insider/outsider’ perspective that Rueda (2005, 2006, 2007) introduced as a critique of the ‘power resources’ approach, as right-wing governments match—but do not exceed—the commitment of left-wing governments for these types of programs. Third, on a more general note, the results suggest that it is too early to conclude that factors related to the ‘new politics’ of the welfare state approach—such as new, strong interest groups (Pierson, 1994), policy diffusion from international organizations (Armingeon, 2007) and the convergence of party preferences (Lindvall, 2010; Nelson, 2013; Tepe and Vanhuysse, 2013)—have rendered the traditional class-based political explanations of social and labor market policy obsolete. The following section highlights the inconclusiveness of previous research, outlines the theoretical argument and derives two hypotheses to be tested. In the third section, the new data set is introduced, and all variables are presented. The fourth section reports the results from a set of models, which lend support to both hypotheses; and the fifth section concludes. Descriptive statistics, robustness checks and extensions are provided in an Appendix, as well as in the online Supplementary Material. 2. Theory and hypotheses 2.1 Party politics and active labor market policy: a review As previously noted by multiple scholars, the cumulative evidence from the studies conducted over the past 10 years about the effect of partisan politics on ALMP ‘effort’ has been surprisingly inconclusive, especially given the similarity in the research questions and data applied (Bonoli, 2013; Tepe and Vanhuysse, 2013; Clasen et al., 2016). Based on the results these studies report, most of them can be divided into three groups.1 Those in the first group have found that left-wing influence is positively related to ALMP ‘effort’ (Huo et al., 2008; Iversen and Stephens, 2008; van Vliet and Koster, 2011). These results are consistent with the ‘power resources’ school’s account of ALMP development, in which social democratic governments are considered more inclined to expand ALMP to strengthen labor as an organized social force by contributing to lower levels of unemployment. This understanding of ALMP was long the dominant one among welfare state scholars (Esping-Andersen, 1990; Janoski, 1990, 1994; Rothstein, 1996; Boix, 1998). A second group of studies have reported that the government’s ideological underpinnings have no impact on ALMP ‘effort’ (Rueda, 2005; Franzese and Hays, 2006; Armingeon, 2007; Gaston and Rajaguru, 2008; Bonoli, 2013). Multiple explanations have been proposed for why ideology is not expected to have an impact on such efforts. Mucciaroni (1990), Swenson (2002), and Farnsworth (2012, 2013) have argued that ALMPs—particularly training programs and labor market services—are often in the interest of workers and employers alike. As such, they are less likely to be a subject of partisan dispute. Another explanation is that policy diffusion—spurred by mutual learning experiences, a broad consensus among policy experts, and the influence of employment strategies adopted by the EU and the OECD—has caused party preferences for labor market policy to converge (Franzese and Hays, 2006; Armingeon, 2007; Lindvall, 2010; Nelson, 2013). This ‘deep shift in thinking’ (Nelson, 2013, p. 272) since around 1998 is sometimes referred to as the ‘activation turn’ (Bonoli, 2010, p. 435). Finally, a few studies have found a negative relationship between left-wing influence and spending on (at least some categories of) ALMP. Drawing on insider–outsider theories of unemployment in economics, Rueda (2005, 2006, 2007) has made the influential claim that because unions and—for electoral reasons—social democratic parties tend to favor the interests of labor market insiders, they are unlikely to support, and might even oppose, ALMPs. According to this account, ALMPs do not favor insiders because they promote the employment entry of outsiders who can underbid insiders’ wage demands and simultaneously increase the tax burden (Rueda, 2005). In accordance with this theory, Rueda (2007) found that increased left-wing strength resulted in a decreased ALMP spending. Building on Rueda’s work, Tepe and Vanhuysse (2013) recently introduced the ‘left party disinvestment thesis’. Supporting a weak version of that thesis, they found that increased left-wing influence does not typically increase the overall ALMP expenditure; in addition, in line with a strong version of that thesis, they found that increased left-wing influence decreases spending on the one category of ALMPs—direct job creation programs—which they suggest will most likely benefit outsiders. Reporting separate analyses for different categories2 of ALMP programs—such as training programs, employment incentives and direct job creation programs—Tepe and Vanhuysse’s (2013) study, together with those by Bonoli (2010), Nelson (2013) and Vlandas (2013), mark a new wave of ALMP research that does not fit nicely into any of the three previous bodies of research. These authors begin to address the problem that ALMP programs differ in terms of their effects on individuals and on the labor market, and that therefore, they also presumably differ in terms of how different political actors value them. Thus, their studies clearly represent a sophisticated advance within the field. However, their joint results regarding the effect of left-wing strength on ALMP ‘effort’ are as inconclusive as those in the previous literature. For training programs, Nelson (2013) finds a positive effect, whereas Tepe and Vanhuysse (2013) and Vlandas (2013) find none. For direct job creation programs, Tepe and Vanhuysse (2013) find a negative effect; Vlandas (2013) finds none; and Nelson (2013) finds a positive effect prior to the ‘activation turn’. For employment incentives, Nelson (2013) finds a positive effect; Tepe and Vanhuysse (2013) find none; and Vlandas (2013) finds a negative effect. These inconclusive results partly stem from the difficulties involved in deriving hypotheses about party preferences for particular categories of ALMP programs because, as aptly demonstrated by Clasen et al. (2016, p. 30), programs that fall in the same administrative category may have ‘very distinctive aims and presumably very different political support coalitions’. 2.2 The role of ALMP: to reduce unemployment or to expand labor supply? In this article, I argue that the differences in ALMP programs’ objectives are key for understanding partisan differences in ALMP preferences. Possibly the first scholar to provide this insight was Furåker (1976), who classified the traditional categories of labor market policy interventions based on whether they were meant to serve either one or both of two possible purposes: reducing (or preventing) unemployment and/or reducing (or preventing) labor shortages.3 In Furåker’s model, as in those of some other contemporary scholars (e.g. Hibbs, 1977), sellers (i.e. workers) and buyers (i.e. employers) on the labor market vary in their preferences regarding unemployment. Accordingly, he suggested that the ways in which governments prioritize labor market measures depend on the extent to which each group has been able to influence government policy. Thus, left-wing governments, which tend to favor workers’ interests, are expected to prioritize measures that aim to reduce unemployment, whereas right-wing governments, which have closer ties to the business community, are expected to be more concerned about labor shortages. Research from the past two decades has provided an additional reason for why left-wing governments might tend to be more concerned about unemployment: electoral motivations stemming from issue ownership. At the ballot box, left-wing governments are often found to be penalized particularly harshly for unemployment (Powell and Whitten, 1993; Whitten and Palmer, 1999; van der Brug et al., 2007). Now, although I find Furåker’s framework largely compelling, I contend that the most important determinant of whether a labor market program is meant to reduce unemployment or labor shortages is not the program’s content but whether it targets people who are already participating in the labor market or those who are not. Indeed, contrary to Rueda’s (2006, p. 388) influential claim that ‘ALMPs unambiguously benefit outsiders’, I argue that whereas some categories of labor market programs—such as sheltered employment for the disabled—target people who are fairly homogeneous with respect to their ‘outsiderness’, most categories accommodate programs for which the primary target group includes those at the ‘core’ of the labor force and programs that target people who are on the fringes or even outside the labor market.4 For instance, Clasen et al. (2016, p. 30) show that the direct job creation category accommodates both types of programs. In addition, labor market training programs are known to accommodate many programs that target particularly disadvantaged groups;5 however, the education schemes and the accompanying work time reduction subsidies included in the so-called short-time work (STW) programs, which many European governments rolled out during the recent financial crisis, primarily targeted workers who were already employed but who ran the risk of becoming unemployed6 (Hijzen and Venn, 2011). These examples illustrate why the attention that Furåker pays to program categories might be misguided, and they might also partly explain why the results from the most recent wave of ALMP research are so inconclusive. To summarize, I propose shifting the focus away from program categories and argue that a program’s overarching aim—unemployment reduction or labor force expansion—is more likely to be the primary source of partisan conflict. Two hypotheses can be derived to test this claim. First, I expect that left-wing governments are more prone than right-wing governments to expand labor market programs that exclusively target people at the ‘core’ of the labor force who are unemployed or at risk of becoming unemployed. In contrast, in line with Boix’s (1998, p. 4) remark that ‘[i]n the first place, all parties prefer to develop policies that maximize growth’, I expect less of a partisan effect for programs that also, or exclusively, target ‘non-core’ groups. These programs do not necessarily primarily aim to reduce unemployment, but they might serve to increase the size of the labor force by making people who would otherwise be inactive—such as housewives, discouraged youth and people in early retirement—begin searching for a job.7 If such programs succeed, growth might increase both directly—through increased output—and indirectly—if the new employment mitigates bottlenecks caused by labor shortages. Moreover, these programs might have side effects, such as increased tax revenue and lower caseloads in other more expensive social security programs, which are attractive to all governments, irrespective of their ideologies or allegiances. Now, it is not obvious that increasing output by increasing labor force participation is in line with the traditional left-wing agenda, a defining feature of which is often held to be the decommodification of labor; i.e. the aim to provide citizens with an opportunity to opt out of work without a potential loss of income or welfare (Esping-Andersen, 1990). However, at odds with this understanding of the left-wing agenda, more recent comparative studies have found that social democratic parties emphasize policies that promote labor market participation over a long-term labor market exit (Huber and Stephens, 2001, Huo et al., 2008). These studies find that left-wing control of the cabinet is associated with those kinds of decommodification policies that do not reduce aggregate levels of employment—such as old-age pension entitlements and short-term unemployment benefit replacement rates—but unrelated to those kinds that provide strong work disincentives—such as the duration and the replacement rates of long-term unemployment benefits. Importantly, this revised understanding of the left-wing’s attitude toward decommodification is compatible with ALMPs that are aimed to increase labor force participation.8 The second hypothesis is that governments with different ideological makeups prefer different ALMP programs depending on what participation entails for an individual’s labor market status. For some types of programs (e.g. full-time training programs), enrollment typically implies that the participant is no longer immediately available for work and, in turn, that his or her unemployment spell is either broken (whereby the unemployment duration counter is reset to zero) or suspended (whereby the duration counter is paused until the participant leaves the program). In other programs, participation does not change one’s labor market status. Two partisan mechanisms might be at play here. First, according to basic search models of the labor market, as more people actively engage in job seeking, labor market competition increases—likely to the detriment of core workers, whose interests are favored by left-wing governments. Second, people who enroll in a program that breaks their unemployment spell are no longer counted among the unemployed. Thus, maintaining a stock of participants in such programs might artificially reduce the unemployment rate in the long term (while keeping the participants active in supposedly productive training or subsidized work). As already noted, a lower unemployment rate might strengthen the bargaining power of labor and be electorally beneficial to left-wing parties. Therefore, I hypothesize that left-wing governments are more inclined to expand ALMP programs that cause a break or suspension of the unemployment spell, whereas right-wing governments are more interested in supporting ALMP programs that keep participants in the labor supply. Table 1 summarizes the hypotheses presented above. Table 1. Hypotheses: program feature × partisanship interactions   H1: Core target group   H2: Broken unemployment     Core  Non-core  Broken  Continuous  Impact of Left-wing government strength on ALMP ‘effort’  +  0  +  –    H1: Core target group   H2: Broken unemployment     Core  Non-core  Broken  Continuous  Impact of Left-wing government strength on ALMP ‘effort’  +  0  +  –  Table 1. Hypotheses: program feature × partisanship interactions   H1: Core target group   H2: Broken unemployment     Core  Non-core  Broken  Continuous  Impact of Left-wing government strength on ALMP ‘effort’  +  0  +  –    H1: Core target group   H2: Broken unemployment     Core  Non-core  Broken  Continuous  Impact of Left-wing government strength on ALMP ‘effort’  +  0  +  –  3. Data and operationalization 3.1 The EU Labor Market Policy database The comparative research on ALMP has invariably used the OECD Labor Market Policy (LMP) database, which contains country-year observations on expenditures on a number of program categories, starting in 1985 for a subset of countries (Grubb and Puymoyen, 2008). However, this database lacks data on program characteristics, which are required to test the hypotheses introduced above. Fortunately, the European Commission (2015) collects data on labor market programs in the EU Member States and Norway that meet all the requirements. Importantly, in this database, the unit of observation is program-year rather than country-year. Therefore, information about a large set of qualitative program characteristics as well as annual summaries of expenditures and participants are reported annually for each program operating in each country, which makes individually analyzing each one of these hundreds of programs possible rather than simply obtaining country-level aggregates. Because, to the best of my knowledge, these data have not previously been applied in this field of research, a few limitations of the EU LMP database should be noted here. First, no data are available from before 1998. However, for the present purposes, this limitation only makes for a tougher test of my hypotheses because, as noted above, previous studies have found the effects of partisan politics to be smaller since the ‘activation turn’ around the turn of the century. Second, all data are reported via a questionnaire that is completed by national authorities, and approximately 10% of the reported quantitative data are based on estimations. Therefore, systematic cross-country differences in reporting and estimation practices might distort cross-country data comparisons. However, in the present study, this problem is mitigated by the fact that all regression models include program-fixed effects, which ensure that no between-country variation is used to estimate the parameters. Third, whereas the data are based in principle, on a full count of labor market programs as defined by Eurostat (2013), the database is only supposed to cover interventions at the national and regional levels. As argued by Clasen et al. (2016), the omission of local ALMPs might distort comparative analyses, yet I argue that the present study is spared from such problems because the unit of analysis is an individual program, not a country-level aggregate. Finally, whereas most program categories in the database contain only labor market interventions that ‘aim to benefit identifiable individuals’ (Eurostat, 2013, p. 7) and that are thus suitable for inclusion in this study, the labor market services category also covers functions that are not directly linked to individual participants, such as services for employers, administrative functions and general overhead. Therefore, this program category is excluded from the study, along with the two categories that are typically not considered ‘active’ labor market policies: out-of-work income maintenance and support and early retirement. Despite these issues, the data in the EU LMP database seem to be of high quality and satisfactorily comparable across interventions and years, particularly if only within-country or within-program variation is used in estimations. 3.2 Dependent variable: ALMP ‘effort’ Most comparative researchers base their indicators of governments’ ALMP ‘efforts’ on how much public spending (relative to GDP) is devoted to the policy area. In an effort to disentangle the effects of deliberate policy decisions from the effects of economic conditions, many scholars control for the ‘problem pressure’ by adjusting the rate of unemployment on either side of the regression equation. However, as discussed by Clasen et al. (2016), unemployment rates are ‘notoriously problematic in comparative analysis as they are expressed as a ratio of the labour force’. Because unemployment, inactivity and employment are communicating vessels, expansion in policy areas such as early retirement, higher education or part-time work that lead to a decline in the unemployment rate might generate ‘a largely artificial image of increasing ALMP “effort”’ (Clasen et al., 2016, p. 27). Moreover, as noted above, many ALMP programs mechanically alter the labor market status of their participants from unemployed to inactive or employed, which means that indicators of ‘effort’ that are adjusted according to unemployment may ‘be endogenously ratcheted upwards or downwards by increases or decreases in expenditure on measures that have a direct impact on the unemployment rate’ (Clasen et al., 2016, p. 28). Another problem with routinely adjusting spending for the unemployment rate is that being unemployed does not necessarily imply that one takes part in the programs provided by the government. An analysis of an indicator reported by Eurostat (2015a), which measures the share of the registered unemployed who participate in an ALMP program, reveals that the ‘activation rate’ varies considerably—between countries and within countries. For the 214 observed country-years, nested in 24 countries, the overall mean activation rate is 21.6% of the registered unemployed. The between-country standard deviation is 11.4 percentage points, and the within-country standard deviation is 6.1 percentage points. This variation indicates that simply adjusting ALMP spending for unemployment does not get us very far if we want to reliably assess how the treatment that an unemployed individual can expect to receive from the government varies between countries or over time. The literature devoted to traditional social insurance systems (e.g. Esping-Andersen, 1990; Korpi and Palme, 1998; Scruggs and Allan, 2006) recognizes these policies’ coverage rates and eligibility criteria as important dimensions. Similarly, I argue that who and how many individuals participate in ALMP programs warrants further attention. Whereas the problems discussed above should not necessarily lead us to dismiss ALMP spending altogether when conceptualizing ALMP ‘effort’, I argue that Esping-Andersen (1990, p. 20) was right to remark that if our aim as welfare state scholars ‘is to test causal theories that involve actors, we should begin with the demands that were actually promoted by those actors’ and that ‘it is difficult to imagine that anyone struggled for expenditure per se’. Because the total expenditure for any ALMP program (or a complete program portfolio) is constituted by two components—the number of participants and the average expenditure per participant—I argue that, for a given ALMP program, the number of participants is a more valid indicator of the government’s inclination to use the program than is total expenditure. Changes in the total expenditure per participant might also reflect, for instance, changes in the efficiency of operations, economies of scale and other factors that might not reflect government’s preferences as clearly as the number of people enrolled in the program. Therefore, I use the data on the average annual participant stock—henceforth denoted the Scope of the ALMP program—to construct the main dependent variable, whereas corresponding models for the program’s total annual Expenditure, measured in million Euros at constant 2005 prices, are reported as robustness checks in Table A5. Admittedly, the Scope indicator is not free of problems. First, it is impossible to tell if a given participant stock in an ALMP program accommodates a large number of short-term transient participants—those soon to enter employment or to transfer to another program—or a small group of participants who have been enrolled for a long time because they are part of a lengthy training scheme or because they are simply ‘trapped’ in the program. However, although this implies that the average participant stock is not a suitable indicator for assessing program content or efficiency, I maintain that Scope is a better proxy than Expenditure for the extent to which a government seeks to intervene in the labor market in a discretionary manner to achieve some particular objective; which, fortunately, is of interest in the present study. Second, whereas the expenditure data in the EU LMP database are considered relatively complete, more gaps can be found in the participant data (European Commission, 2015). Whereas one or more observations of expenditure data exist for 1270 programs, participant data only exist for 1113 programs. This is another reason to use Expenditure as a robustness check. Because the relationships between the programs’ Scope (as well as Expenditure) and the independent variables of interest are not expected to be linear (for good reason), the dependent variable is log-transformed.9 3.3 Independent variables The analyses include two program-level independent variables of particular interest: Core target group and Broken unemployment. Both are program-year dummy variables that are extracted from the qualitative data reports of the EU LMP database. The data required to produce them are nearly complete. Data are missing for only 0.6% of all observations, and for about half of those with missing data, all needed information can be inferred from the program descriptions included in the dataset. Core target group is assigned a 0 if the program is available for one or both of the two target groups ‘not registered’ and ‘other registered jobseekers’, and a 1 if it exclusively targets those who are ‘registered unemployed’ or ‘employed at risk of involuntary job loss’. According to Eurostat (2013, p. 45), ‘not registered’ ‘indicates where interventions are targeted at groups who are not in employment or where registration with the PES [Public Employment Service] is not a prerequisite for participation’.10 In practice, ‘other registered jobseekers’ ‘refers to persons who are unemployed (but do not qualify as registered unemployed), underemployed or inactive’. While the definitions of these two target groups cover a rather heterogeneous set of individuals, none of these individuals is unemployed according to the national definition—nor employed and at risk of unemployment. Therefore, programs that target—partly or exclusively—one of these two groups serve, at least to some extent, to increase labor force participation. Conversely, programs that only target ‘registered unemployed’ or ‘employed at risk’ primarily seek to reduce (or prevent) unemployment. To test hypothesis H1, I include an interaction term between Core target group and Left-wing government strength. The latter variable is defined, according to a well-established practice, as the number of cabinet posts held by social democrats and members of other leftist parties as a share of the total number of cabinet posts, weighted by the number of days in office in a given year. Data are retrieved from Armingeon et al. (2015). The second dummy variable, Broken unemployment, is assigned a 1 if participation in the program breaks or suspends the unemployment spell and a 0 if unemployment is continuous. Programs for which this variable is coded as ‘not relevant’—for instance, because they do not target the unemployed—are assigned a 0. To test hypothesis H2, I include the interaction term Broken unemployment × Left-wing government strength. To control for the potential confounding interactions between Left-wing government strength and the program category, I include interaction terms between Left-wing government strength and each of the main ALMP program categories: Training, Employment incentives, Sheltered and supported employment and rehabilitation, Direct job creation and Start-up incentives (with the interaction term that includes Training omitted and used as a reference category). I include program-fixed effects in all models to control for unobserved between-program heterogeneity, including any differences stemming from potential systematic cross-country differences in the way that participant and expenditure data are reported to the EU LMP database. I thereby also control for a number of largely time-invariant, country-level factors that have been found to affect ALMP ‘effort’ in previous studies, including welfare regime differences, trade openness, the degree of employer coordination and the involvement of social partners in policymaking.11 To control for within-country variation in ‘problem pressure’, I include an item for the logged number of Unemployed and inactive individuals from ages 15–64 years. I also include controls for Real GDP at constant 2005 prices, Real GDP growth, Government deficit and Government debt. The indicator of GDP is retrieved from Eurostat (2015b); data on unemployment and inactivity come from the EU Labor Force Survey (Eurostat, 2015c); and the other variables are provided by Armingeon et al. (2015). I also control for the Number of programs that are operating in the country during a particular year because—all other things being equal—the more ALMP programs in place at the same time, the smaller each individual program is likely to be. Finally, I add year dummies to control for possible common temporal shocks. For my final sample, I exclude 14 programs (145 observations) that have at least one reported break in the time series and 66 programs (477 observations) that have at least one gap in the data. However, as demonstrated by the robustness check in the Supplementary Material (Model S2), including these observations in the sample does not markedly change the results for the coefficients of interest. To assure full comparability between models with and without a lagged dependent variable, I also leave out the first observation of each panel in the models without a lagged dependent variable to attain an identical sample for both types of models. In doing so, I exclude approximately 1000 observations from the analysis, including 132 full programs that have been observed for only 1 year. These exclusions slightly change the composition of my sample, but should not pose any real problem to the analysis.12 The final sample comprises approximately 5600 program-years nested in nearly 900 programs, which are nested in 28 countries.13 Descriptive statistics for all included variables are reported in Table A2. Table A3 shows how the data are structured by presenting a detailed description of two program-year observations from two different programs in different countries. In the Supplementary Material, Table S1 documents how the panels and observations are distributed across countries and years, Figure S1 presents the frequencies of panels of varying length, across program categories and Table S2 describes the observation patterns for all panels included in the study. Table 2. Regression results   (1)  (2)  (3)  (4)  (5)  Lagged dependent variable        0.59***  0.69***          (0.04)  (0.01)  Left-wing government strength  0.50***  −0.47**  −0.41**  −0.35***  −0.33***    (0.16)  (0.18)  (0.19)  (0.12)  (0.11)  Core target group × Left-wing gov’t    0.57***  0.54***  0.39***  0.39***      (0.14)  (0.15)  (0.11)  (0.10)  Broken unemployment × Left-wing gov’t    0.77***  0.83***  0.53***  0.47***      (0.18)  (0.20)  (0.16)  (0.09)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.06***  0.06***  0.05***  0.03**  0.03***    (0.02)  (0.02)  (0.02)  (0.01)  (0.01)   Unemployed and inactive (log)  1.22**  1.35**  1.32**  1.25***  1.24***    (0.55)  (0.55)  (0.53)  (0.41)  (0.27)   Government deficit  −0.01  −0.01  −0.01  −0.00  −0.00    (0.01)  (0.01)  (0.01)  (0.00)  (0.01)   Government debt  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.00  −0.00  −0.00  −0.00  −0.00**    (0.01)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.21  −0.43**  −0.39**  −0.27**  −0.27***    (0.15)  (0.16)  (0.15)  (0.12)  (0.10)   Broken unemployment  −0.07  −0.20  −0.22  −0.08  −0.07    (0.19)  (0.17)  (0.17)  (0.12)  (0.11)   Employment incentives × Left-wing gov’t      −0.16  −0.06  −0.04        (0.21)  (0.13)  (0.10)   Sheltered empl. and rehab. × Left-wing gov’t      −0.49  −0.29  −0.26*        (0.29)  (0.20)  (0.15)   Direct job creation × Left-wing gov’t      0.50  0.34  0.31**        (0.54)  (0.31)  (0.15)   Start-up incentives × Left-wing gov’t      −0.16  −0.07  −0.03        (0.34)  (0.34)  (0.18)  Constant  −9.76  −11.39  −11.09  −14.38**      (8.29)  (8.26)  (8.02)  (6.15)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  5668  5668  5668  5668  5668  Programs  896  896  896  896  896    (1)  (2)  (3)  (4)  (5)  Lagged dependent variable        0.59***  0.69***          (0.04)  (0.01)  Left-wing government strength  0.50***  −0.47**  −0.41**  −0.35***  −0.33***    (0.16)  (0.18)  (0.19)  (0.12)  (0.11)  Core target group × Left-wing gov’t    0.57***  0.54***  0.39***  0.39***      (0.14)  (0.15)  (0.11)  (0.10)  Broken unemployment × Left-wing gov’t    0.77***  0.83***  0.53***  0.47***      (0.18)  (0.20)  (0.16)  (0.09)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.06***  0.06***  0.05***  0.03**  0.03***    (0.02)  (0.02)  (0.02)  (0.01)  (0.01)   Unemployed and inactive (log)  1.22**  1.35**  1.32**  1.25***  1.24***    (0.55)  (0.55)  (0.53)  (0.41)  (0.27)   Government deficit  −0.01  −0.01  −0.01  −0.00  −0.00    (0.01)  (0.01)  (0.01)  (0.00)  (0.01)   Government debt  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.00  −0.00  −0.00  −0.00  −0.00**    (0.01)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.21  −0.43**  −0.39**  −0.27**  −0.27***    (0.15)  (0.16)  (0.15)  (0.12)  (0.10)   Broken unemployment  −0.07  −0.20  −0.22  −0.08  −0.07    (0.19)  (0.17)  (0.17)  (0.12)  (0.11)   Employment incentives × Left-wing gov’t      −0.16  −0.06  −0.04        (0.21)  (0.13)  (0.10)   Sheltered empl. and rehab. × Left-wing gov’t      −0.49  −0.29  −0.26*        (0.29)  (0.20)  (0.15)   Direct job creation × Left-wing gov’t      0.50  0.34  0.31**        (0.54)  (0.31)  (0.15)   Start-up incentives × Left-wing gov’t      −0.16  −0.07  −0.03        (0.34)  (0.34)  (0.18)  Constant  −9.76  −11.39  −11.09  −14.38**      (8.29)  (8.26)  (8.02)  (6.15)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  5668  5668  5668  5668  5668  Programs  896  896  896  896  896  Model 1–4: Cluster-robust standard errors in parentheses (clustered at the country level). Model 5: Bootstrapped standard errors in parentheses. Bias correction initialized by Arellano and Bond estimator (Bruno 2005). *P < 0.10, **P < 0.05, ***P < 0.01. Table 2. Regression results   (1)  (2)  (3)  (4)  (5)  Lagged dependent variable        0.59***  0.69***          (0.04)  (0.01)  Left-wing government strength  0.50***  −0.47**  −0.41**  −0.35***  −0.33***    (0.16)  (0.18)  (0.19)  (0.12)  (0.11)  Core target group × Left-wing gov’t    0.57***  0.54***  0.39***  0.39***      (0.14)  (0.15)  (0.11)  (0.10)  Broken unemployment × Left-wing gov’t    0.77***  0.83***  0.53***  0.47***      (0.18)  (0.20)  (0.16)  (0.09)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.06***  0.06***  0.05***  0.03**  0.03***    (0.02)  (0.02)  (0.02)  (0.01)  (0.01)   Unemployed and inactive (log)  1.22**  1.35**  1.32**  1.25***  1.24***    (0.55)  (0.55)  (0.53)  (0.41)  (0.27)   Government deficit  −0.01  −0.01  −0.01  −0.00  −0.00    (0.01)  (0.01)  (0.01)  (0.00)  (0.01)   Government debt  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.00  −0.00  −0.00  −0.00  −0.00**    (0.01)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.21  −0.43**  −0.39**  −0.27**  −0.27***    (0.15)  (0.16)  (0.15)  (0.12)  (0.10)   Broken unemployment  −0.07  −0.20  −0.22  −0.08  −0.07    (0.19)  (0.17)  (0.17)  (0.12)  (0.11)   Employment incentives × Left-wing gov’t      −0.16  −0.06  −0.04        (0.21)  (0.13)  (0.10)   Sheltered empl. and rehab. × Left-wing gov’t      −0.49  −0.29  −0.26*        (0.29)  (0.20)  (0.15)   Direct job creation × Left-wing gov’t      0.50  0.34  0.31**        (0.54)  (0.31)  (0.15)   Start-up incentives × Left-wing gov’t      −0.16  −0.07  −0.03        (0.34)  (0.34)  (0.18)  Constant  −9.76  −11.39  −11.09  −14.38**      (8.29)  (8.26)  (8.02)  (6.15)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  5668  5668  5668  5668  5668  Programs  896  896  896  896  896    (1)  (2)  (3)  (4)  (5)  Lagged dependent variable        0.59***  0.69***          (0.04)  (0.01)  Left-wing government strength  0.50***  −0.47**  −0.41**  −0.35***  −0.33***    (0.16)  (0.18)  (0.19)  (0.12)  (0.11)  Core target group × Left-wing gov’t    0.57***  0.54***  0.39***  0.39***      (0.14)  (0.15)  (0.11)  (0.10)  Broken unemployment × Left-wing gov’t    0.77***  0.83***  0.53***  0.47***      (0.18)  (0.20)  (0.16)  (0.09)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.06***  0.06***  0.05***  0.03**  0.03***    (0.02)  (0.02)  (0.02)  (0.01)  (0.01)   Unemployed and inactive (log)  1.22**  1.35**  1.32**  1.25***  1.24***    (0.55)  (0.55)  (0.53)  (0.41)  (0.27)   Government deficit  −0.01  −0.01  −0.01  −0.00  −0.00    (0.01)  (0.01)  (0.01)  (0.00)  (0.01)   Government debt  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.00  −0.00  −0.00  −0.00  −0.00**    (0.01)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.21  −0.43**  −0.39**  −0.27**  −0.27***    (0.15)  (0.16)  (0.15)  (0.12)  (0.10)   Broken unemployment  −0.07  −0.20  −0.22  −0.08  −0.07    (0.19)  (0.17)  (0.17)  (0.12)  (0.11)   Employment incentives × Left-wing gov’t      −0.16  −0.06  −0.04        (0.21)  (0.13)  (0.10)   Sheltered empl. and rehab. × Left-wing gov’t      −0.49  −0.29  −0.26*        (0.29)  (0.20)  (0.15)   Direct job creation × Left-wing gov’t      0.50  0.34  0.31**        (0.54)  (0.31)  (0.15)   Start-up incentives × Left-wing gov’t      −0.16  −0.07  −0.03        (0.34)  (0.34)  (0.18)  Constant  −9.76  −11.39  −11.09  −14.38**      (8.29)  (8.26)  (8.02)  (6.15)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  5668  5668  5668  5668  5668  Programs  896  896  896  896  896  Model 1–4: Cluster-robust standard errors in parentheses (clustered at the country level). Model 5: Bootstrapped standard errors in parentheses. Bias correction initialized by Arellano and Bond estimator (Bruno 2005). *P < 0.10, **P < 0.05, ***P < 0.01. 4. Descriptive evidence Before turning to the modeling exercises, consider some initial descriptive evidence for the two hypotheses, presented in Figure 1. For each of the 28 studied countries, each graph’s x-axis plots Left-wing government strength averaged over all the years for which data for the country is available in the EU LMP database (ranging from 6 to 16 years between 1998 and 2013). For each corresponding country and time period, the panel on the left plots the average total ALMP expenditure as a share of GDP. No apparent association exists between the prevalence of left-wing cabinets and total ALMP expenditure during the observed period. In contrast, the center panel and the panel on the right—each of which splits the total expenditure into two parts—show that a clear correlation exists between Left-wing government strength and the share of the total ALMP expenditure allocated to programs that fall in the Core target group and Broken unemployment categories. In line with both hypotheses, the stronger the Left-wing government strength, the larger share of overall spending that is devoted toward programs in the Core target group (r = 0.30, P = 0.12) and Broken unemployment (r = 0.51, P < 0.01) categories. Figure 1. View largeDownload slide Left-wing government strength and ALMP in 28 European countries. Note: For definitions, see Section 3. Indicators are averaged over all years for which data for the country are available in the EU LMP database: 1998—2013: AT, BE, DK, FI, IE, IT, NL, NO, PT, SE; 1998–2012: DK, ES, FR, LU; 1998–2011: UK; 1998–2010: EL; 2002–2013: CZ; 2003–2013: EE, LT, LV; 2003–2012: HU, RO; 2004–2013: BG, SK; 2005–2013: SI; 2005–2012: PL; 2006–2012: MT; 2006–2011: CY. Sources: European Commission (2015), Armingeon et al. (2015) and author’s calculations. Figure 1. View largeDownload slide Left-wing government strength and ALMP in 28 European countries. Note: For definitions, see Section 3. Indicators are averaged over all years for which data for the country are available in the EU LMP database: 1998—2013: AT, BE, DK, FI, IE, IT, NL, NO, PT, SE; 1998–2012: DK, ES, FR, LU; 1998–2011: UK; 1998–2010: EL; 2002–2013: CZ; 2003–2013: EE, LT, LV; 2003–2012: HU, RO; 2004–2013: BG, SK; 2005–2013: SI; 2005–2012: PL; 2006–2012: MT; 2006–2011: CY. Sources: European Commission (2015), Armingeon et al. (2015) and author’s calculations. Clearly, these associations are no more than simple correlations, and, as discussed above, numerous factors potentially influence both the cabinet’s composition and the composition of the ALMP portfolio, thereby confounding the relationships displayed in Figure 1. Studying these relationships in a more rigorous manner requires a more advanced econometric analysis, which is the topic of the next section. 5. Estimation strategy and results This section reports the results from a set of models that can all be represented—with various restrictions—with the same general equation:   Ypct=γYpct−1+αp+δt +βLct+λ′Ppct+θ′(Lct× Ppct)+ ρ′Cct+ɛpct (1) Here, Ypct is the Scope of program p in country c in year t; Ypct-1 is its first lag; αp is a program-specific intercept; and δt is a year-fixed effect. Lct is Left-wing government strength in country c in year t; Ppct is a vector of program-level variables; λ is a vector of the main effects of these variables; and θ is a vector of the effects of the interaction between Lct and each of the variables in Ppct. Cct is a vector of time-varying, country-level controls; ρ is a vector of the effects of those controls; and ɛpct is the error term. Because the program-fixed effects included in these models do not necessarily completely control for within-cluster correlation (Cameron and Miller, 2015), I use cluster-robust standard errors whenever possible. As the more conservative option for nested data involves clustering on a higher level, I cluster on the 28 countries included rather than on the program level. A Breusch-Pagan LM test confirms that considerable between-program heterogeneity exists in the data and that a random effects or fixed effects model is thus preferable to pooled OLS regression. Hausman tests performed on various model specifications consistently reject one of the key assumptions underpinning the random effects model, i.e., that all independent variables are uncorrelated with the residuals. Therefore, I opt for the consistent—although less efficient—fixed effects model, which only makes use of within-program variation. Model 1 in Table 2 reports that, on average, Left-wing government strength has a positive and statistically significant effect on the Scope of ALMP programs when the country-level and program-level factors discussed above are controlled for and all potential interactions are restricted to zero. However, the positive coefficients for the two interaction terms included in Model 2 seemingly indicate that the effect of left-wing power varies between programs in line with the hypotheses. To rule out any suspicion that these results are driven by confounding interactions between Left-wing government strength and program category, Model 3 controls for that risk by adding four additional interaction terms.14 Including them does not drastically alter the results.15 The coefficients estimated in a panel-data fixed effects model without a lagged dependent variable, such as Model 3—for which the identifying assumption is that all potential confounders are time-invariant—tend to be too large if time-varying confounders are also present (Angrist and Pischke, 2009, ch. 5.4). Because inflated coefficients cannot be ruled out in the present case, Model 4 includes a lagged dependent variable. Consequently, all the coefficients of interest are reduced substantially. Now, as first noted by Nickell (1981), OLS estimates from models with both fixed effects and a lagged dependent variable tend to be biased due to the correlation between the lagged dependent variable and the error term. Therefore, the main model, Model 5, applies a version of the bias-corrected LSDV estimator (LSDVc) (Kiviet, 1995), which has been found to be reasonably accurate when evaluated on unbalanced, short panels, such as those in the data used here (Flannery and Hankins, 2013). The coefficients of interest in Model 5 are slightly smaller than those in Model 4, and, because the LSDVc estimator developed by Bruno (2005) only reports bootstrapped standard errors, these standard errors are markedly smaller for some coefficients than the cluster-robust standard errors used in the other models. However, even if standard errors of a similar size to those in Model 4 were applied to the corresponding coefficients in Model 5, the statistical significance of the coefficients of interest would remain. Although the positive and strongly significant interaction effects reported from all models in Table 2 lend preliminary support to both hypotheses, a closer examination of how Left-wing government strength’s estimated effect on program Scope varies between programs with different characteristics is needed before any conclusions can be drawn. The results of such an examination, based on the estimates from Model 5, are presented graphically in Figure 2. Figure 2. View largeDownload slide Partial conditional effects of left-wing government on ALMP ‘effort’. Note: Partial effects of a change from 0% to 100% Left-wing government strength on Scope conditional on a program characteristic, with all other variables held at their means. Error bars denote a 95% confidence interval. All effects are estimated from Model 5. Raw effect estimates, standard deviations and P-values are reported in the Table A4. Figure 2. View largeDownload slide Partial conditional effects of left-wing government on ALMP ‘effort’. Note: Partial effects of a change from 0% to 100% Left-wing government strength on Scope conditional on a program characteristic, with all other variables held at their means. Error bars denote a 95% confidence interval. All effects are estimated from Model 5. Raw effect estimates, standard deviations and P-values are reported in the Table A4. Whereas the coefficients for all the independent variables reported in Table 2 represent effects on the exponent of the natural logarithm of Scope, Figure 2 plots exponentiated effects that allow a more straightforward interpretation. Each bar represents an estimate of the average percentage change in the Scope—i.e. the average percentage change in the participant stock—of an ALMP program that results from a ‘full’ cabinet change—from a cabinet in which left-wing parties hold no seats to one in which they hold all seats—conditional on the particular program characteristic of interest and with all other variables set at their means. First, consider the bar on the left in the panel on the left. It indicates that, for a program that targets only ‘core’ groups, the estimated average effect of a change of government is a 48% increase in Scope.16 In contrast to this rather substantial effect, the bar on the right in the panel on the left shows that the estimated effect of Left-wing government strength on Scope for programs that target ‘non-core’ groups is less than 1% and far from being statistically significant. These results strongly corroborate hypothesis H1. As shown in the panel on the right, a full shift to a left-wing government is estimated to have a considerable effect on programs that entail a broken unemployment spell. For an average program with this feature, the estimated effect is a 46% increase in Scope. For programs that do not break the unemployment spell, the effect is –9%, which is in the hypothesized direction but not large enough to be statistically significant. Taken together, these results lend fairly strong support to hypothesis H2. In sum, the overall results provide evidence of substantial heterogeneous partisan effects on the governments’ ALMP ‘effort’ in programs with varying characteristics, which is in line with the theoretical argument outlined above. In addition, Table A5 provides an identical set of models applied to the logged annual program Expenditure, which produces coefficients that are only slightly smaller than those for Scope in Table 2 and that are all significant to the same extent. Moreover, as reported in Table A6, with one interesting exception,17 the estimates from Model 5 are found to be in the same direction and of a similar or larger value when the sample is split into five country clusters: Scandinavian, Continental, Anglo-Saxon, Southern and Central and Eastern European. Although the coefficients in these models are not significant in most cases, they provide a preliminary indication that no particular country or welfare regime drives the results. Additional models of Scope reported in the Supplementary Material (Tables S3–S7), suggest that the results are robust to other measures of left-wing power, to the inclusion of only one of the two key hypothesized variables and its interaction at a time, and to a number of sample adjustments. Finally, the Supplementary Material (Table S8) reports a set of models that follow the standard approach in the field by aggregating—for each of the four program categories—annual expenditures on the country-year level and expressing them as shares of GDP. In line with expectations, the results from these models indicate that, on the country-year level, Left-wing government strength is specifically associated with Core target group programs and Broken unemployment programs. 6. Concluding discussion This article aims to make two contributions to the comparative ALMP literature. First, it introduces the rich program-level ALMP data in the EU LMP database, which allows for a much more detailed analysis of ALMP programs across Europe compared with the aggregate expenditure data that currently dominates the field. Second, it advances an argument about how the impact of partisanship on ALMP varies across programs with varying characteristics, which differs from arguments in all three dominant strands of the literature. Whereas the historically dominant ‘power resources’ account holds that ALMPs are primarily promoted by the left (Esping-Andersen, 1990; Janoski, 1990; Janoski, 1994; Rothstein, 1996; Boix, 1998; Huo et al., 2008), the results presented here show that, for programs that target groups outside the ‘core’ of the labor force—which accounted for up to 40% of the observations—right-wing governments are just as expansionary as left-wing governments. In addition, the results are also at odds with the understanding of ALMP advanced in ‘insider/outsider’ perspective that Rueda (2005, 2006, 2007) introduced as a critique of the ‘power resources’ approach. At first glance, the positive interaction effect found between left-wing strength and ALMP ‘effort’ in programs that target people at the ‘core’ of the labor force seems to corroborate Rueda’s ‘insider/outsider’ hypothesis and the ‘left party disinvestment thesis’ introduced by Tepe and Vanhuysse (2013). However, the size of the interaction effect reported in Figure 2 indicates that right-wing governments match—but do not exceed—left-wing governments’ commitment to programs that also target ‘non-core’ groups. This study thus provides no evidence that left-wing governments cater less to the interests of ‘outsiders’ than right-wing governments. My understanding of why the ‘power resources’ approach and the ‘insider/outsider’ approach ascribe different ALMP preferences to parties with different ideologies is that these approaches make different implicit assumptions about the overall objective of such policies. On the one hand, in the ‘power resources’ account, ALMPs aim to reduce and/or prevent unemployment—whereby they are expected to strengthen the position of organized labor; on the other hand, in the ‘insider/outsider’ account, ALMPs seek to bring outsiders into the labor market—whereby they are expected to challenge the status of the generally better-organized workers at the ‘core’ of the labor force. The present study reconciles these perspectives by stressing that ALMPs can serve either purpose. In line with the ‘power resources’ account, left-wing governments are more inclined to expand two types of programs: those that seek to reduce or prevent unemployment among people at the ‘core’ of the labor force and those for which enrollment implies a temporary exit from open unemployment. In addition, in line with the ‘insider/outsider’ approach, left-wing parties do indeed cater to the interests of ‘core’ groups (by targeting them with ALMPs). However, this study finds that left-wing and right-wing governments are equally inclined to expand programs that also aim to increase the labor market participation of ‘non-core’ groups, which is at odds with the ‘insider/outsider’ approach. As increased labor supply is conducive to growth, the results are in line with Boix’s (1998, p. 11) basic argument that all parties have a preference for growth-enhancing policies, although the results do not support his assertion that right-wing governments ‘reject any sort of public capital formation policies’ (to which Boix counts human capital-enhancing ALMPs). Insofar as these results indicate that present-day right-wing governments are more inclined to channel public resources toward training, employment subsidies and other human capital-enhancing policies for marginal groups as a means of increasing their labor supply (instead of relying on traditional tax-reducing strategies alone), they perhaps corroborate the third strand of the literature, which argues that policy convergence has occurred in recent decades (Lindvall, 2010; Bonoli, 2013; Tepe and Vanhuysse, 2013; Nelson, 2013). However, the results clearly suggest that it is too soon to conclude, as some do, that partisan politics have lost their relevance in the ‘post-activation turn’ welfare state. Instead, the results illustrate how a traditional ideological conflict extends into the realm of ALMP programs and finds new expressions in their detailed policy settings. As such, comparative scholars need to move beyond measures of aggregate spending and more closely examine the ways in which a government that is elected to administer a modern welfare state can recalibrate (rather than revoke) the large policy portfolios that it inherits to serve its own particular objectives. Notably, the design of the present study—which, to achieve a reliable identification strategy, only exploits within-country variation—is limited because it produces effect estimates that are averaged across countries. As recent studies have reported cross-country variation in the effect of partisanship on ALMP (e.g. Vlandas, 2013), a more comprehensive exploration of how the party preferences theorized in the present study might vary across institutional environments seems like a promising endeavor for future research. The EU’s possible influence on the ALMP portfolios of its member states—particularly those in Central and Eastern Europe and those affected by the new stricter fiscal rules of the EMU—also deserves further investigation. Finally, more research into the political determinants of the composition of the ALMP program portfolios is needed. Under what circumstances are programs with different characteristics established, recast and eventually shut down? The ample and still largely unused data in the EU LMP database might very well hold the answers to those and similar questions. Supplementary material Supplementary material is available at Socio-Economic Review Journal online. 1 For a more detailed review of most of these studies, see Tepe and Vanhuysse (2013). 2 A description of the traditional categories of ALMP programs is provided in Table A1. 3 In Furåker’s (1976, p. 106) classification scheme, direct job creation, start-up incentives and early retirement measures serve to reduce unemployment but not labor shortages, whereas measures to increase the labor market participation of people who are not currently in the labor force and to expand labor immigration serve to reduce labor shortages but not unemployment. Moving grants and certain types of labor market training, as well as most placement services, serve to reduce both unemployment and labor shortages by improving matching. Some labor market policy measures, including unemployment benefits and general placement services for people who are already employed, serve none of the purposes. 4 Here, in contrast to Furåker (1976), I argue that moving grants, labor market training, and placement services do not always serve to reduce unemployment; these programs only reduce employment when they are targeted toward people who are unemployed or at risk of becoming unemployed. 5 One example is the Italian ‘work-entry contract’, Contratto di inserimento lavorativo, through which employers can provide apprenticeships to youth and certain categories of disadvantaged workers, such as the long-term unemployed, the disabled and women who live in problem areas (European Commission, 2015). 6 In some analyses [for instance, Thelen (2014, ch. 4)], STW schemes are explicitly distinguished from ALMP schemes, based precisely on their distinct target groups. This approach is unfortunate for two reasons: first, it relies on the unverified assumption that all other ALMPs target ‘outsiders’. Second, it does not fit well with the way that these policies are reported and aggregated in the widely-used labor-market policy databases, which also include programs that target the ‘employed at risk of unemployment’, such as STW programs (Grubb and Puymoyen, 2008; Eurostat, 2013). 7 Thus, programs that target people who are not already participating in the labor force might even increase the unemployment rate, to the extent that people begin to supply their labor but fail to find employment. 8 Another possible objection is that, if programs that target the unemployed are effective in putting people back into work, then output will increase, thereby encouraging right-wing governments to support these programs as well. Although this is true it does not necessarily mean that, for a given output increase caused by people reentering work from unemployment as a result of a particular program, right-wing governments will value that program as much as left-wing governments. To be clear, I do not mean to suggest that right-wing governments have no interest in programs that target the unemployed; but, because the possible distributional and electoral effects of reducing unemployment tend to benefit left-wing parties more, I argue that these will likely be more prone to support such programs. Another reason for that is that some programs that target the unemployed, such as training or direct job creation, might be seen as enablers of the kind of short-term decommodification of core workers favored by left-wing parties, as they might guarantee participants an alternative, temporary, nonmarket-based, source of income (while simultaneously serving as a legitimizing check of their willingness to work). I am grateful to one of the reviewers for making this point. 9 Residual analyses presented in the Supplementary Material (Figure S2) confirm that this operation is essential: the residuals from these regressions approximately follow a normal distribution only if it the dependent variable is log-transformed. 10 One example is the recruitment subsidy Nystartsjobb, established in Sweden in 2007. Eligible participants include not only the registered unemployed but also anyone who has been absent from the labor market for a long time (typically for more than 1 year) or who is a newly arrived immigrant (European Commission, 2015). 11 The program-level fixed effects also mitigate the plausible concern that not all parties classified as left-wing parties across the 28 countries are necessarily positioned farther to the left on the left–right continuum than all right-wing parties. Because only within-country information is used in all models, only the relative position of left-wing and right-wing parties within each country is important for this indicator of left-wing strength to spare this study from such concerns. 12 Kaplan–Meier survival estimates (not reported here) indicate that 98% of the programs in these categories survive beyond one year and that 89% survive for more than 2 years. Model S3 in the Supplementary Material is identical to Model 3 (i.e. it does not include a lagged dependent variable) but it is run on the larger sample and returns almost identical results. 13 To be precise, a few dozen programs are divided into two or more components because a single program might comprise expenditure and participant data that need to be divided between more than one program category. Here, each of these components is treated as a separate unit of observation. 14 The main effects of Core target group and Broken unemployment, unlike those of the four program categories, are not fully absorbed by the program-fixed effects because considerable within-program variation exists in these policy settings over time. Programs could conceivably be deliberately modified over time. For instance, according to the data, 16 of the 27 operating programs in Sweden changed from 1 to 0 in Broken unemployment between 2012 and 2013. These changes correspond well to legislative changes introduced in 2013, which increased the job search requirements for all recipients of unemployment insurance benefits. Still, in other cases these changes may reflect reporting errors. However, few haphazard changes occur; for only 7 of the 93 programs for which there is one or more change in Core target group and/or Broken unemployment over time, the policy setting switches back again at a later time. As a robustness check, Model S4 in the Supplementary Material omits these 93 programs. The sample is reduced by 751 observations, but the coefficients for the variables of interest are only marginally changed and are still statistically significant. 15 For the sake of parsimony, no model includes the interactions between the program category and Core target group or between the program category and Broken unemployment, as all eight are insignificant. 16 The exponentiated effect of coefficient β is calculated by taking the base of natural logarithm, e, to the power of β. As the estimated marginal effects of a change to a Left-wing government on Scope for a program with a Core target group is 0.393 in Model 5, the exponentiated effect is e0.393 = 1.481. 17 In continental countries, Broken unemployment × Left-wing government strength is markedly smaller. Acknowledgements Previous versions of this article have been presented at annual meetings of the Swedish Political Science Association in Stockholm (2015), the Swedish Network for Social Policy and Welfare Research in Lund (2015) and the Midwest Political Science Association in Chicago (2016) as well as at various seminars at Uppsala University. I thank the participants in these seminars, Per Andersson, Ingrid Esser, Daniel Fredriksson, Anders Lindbom, Karl-Oskar Lindgren, Sven Oskarsson, Joakim Palme and two anonymous reviewers for providing helpful comments and advice. Financial support from Uppsala Center for Labor Studies is gratefully acknowledged. References Angrist J. D., Pischke J.-S. ( 2009) Mostly Harmless Econometrics: An Empiricist’s Companion , Princeton, NJ, Princeton University Press. Armingeon K. ( 2007) ‘ Active Labour Market Policy, International Organizations and Domestic Politics’, Journal of European Public Policy , 14, 905– 932. Google Scholar CrossRef Search ADS   Armingeon K., Isler C., Knöpfel L., Weisstanner D., Engler S. ( 2015) Comparative Political Data Set, 1960-2013 , Bern, Institute of Political Science and Publishing House, University of Bern. Boix C. ( 1998) Political Parties, Growth and Equality: Conservative and Social Democratic Economic Strategies in the World Economy , Cambridge, Cambridge University Press. Bonoli G. ( 2010) ‘ The Political Economy of Active Labor-Market Policy’, Politics & Society , 38, 435– 457. Google Scholar CrossRef Search ADS   Bonoli G. ( 2013) The Origins of Active Social Policy: Labour Market and Childcare Policies in a Comparative Perspective , Oxford, UK, Oxford University Press. Bonoli G., Natali D. ( 2012) The Politics of the New Welfare State , Oxford, UK, Oxford University Press. Bruno G. S. F. ( 2005) ‘Estimation and Inference in Dynamic Unbalanced Panel-Data Models with a Small Number of Individuals’, STATA Journal , 5, 473– 500. Cameron A. C., Miller D. L. ( 2015) ‘ A Practitioner’s Guide to Cluster-Robust Inference’, Journal of Human Resources , 50, 317– 372. Google Scholar CrossRef Search ADS   Card D., Kluve J., Weber A. ( 2010) ‘ Active Labour Market Policy Evaluations: A Meta-Analysis’, Economic Journal , 120, F452– F477. Clasen J., Clegg D., Goerne A. ( 2016) ‘ Comparative Social Policy Analysis and Active Labour Market Policy: Putting Quality before Quantity’, Journal of Social Policy , 45, 21– 38. Google Scholar CrossRef Search ADS   Eichhorst W., Kaufmann O., Konle-Seidl R., Reinhard H.-J. ( 2008) ‘Bringing the Jobless into Work? An Introduction to Activation Policies’. In Eichhorst W., Kaufmann O., Konle-Seidl R. (eds) Bringing the Jobless into Work? Experiences with Activation Schemes in Europe and the US , Berlin and Heidelberg, Springer. Esping-Andersen G. ( 1990) The Three Worlds of Welfare Capitalism , Cambridge, Polity and Publishing House. European Commission ( 2015). ‘Reference Metadata: Labour Market Policy (source: DG EMPL) (lmp)’, accessed at http://ec.europa.eu/eurostat/cache/metadata/en/lmp_esms.htm on June 4, 2015. Eurostat ( 2013) Labour Market Policy Statistics. Methodology 2013, Report, Eurostat. Eurostat ( 2015a) ‘Dataset: Activation of Registered Unemployed (source: DG EMPL)’, accessed at http://ec.europa.eu/eurostat/en/web/products-datasets/-/LMP_IND_ACTRU on November 18, 2015. Eurostat ( 2015b) ‘Dataset: GDP and Main Components (Output, Expenditure, and Income)’, accessed at http://ec.europa.eu/eurostat/en/web/products-datasets/-/NAMA_10_GDP on July 28, 2015. Eurostat ( 2015c) ‘Dataset: Population by Sex, Age, Nationality and Labour Status (1 000)’, accessed at http://ec.europa.eu/eurostat/en/web/products-datasets/-/LFSA_PGANWS on July 28, 2015. Farnsworth K. ( 2012) Social Versus Corporate Welfare: Competing Needs and Interests within the Welfare State , London, UK, Palgrave Macmillan. Farnsworth K. ( 2013) ‘ Bringing Corporate Welfare In’, Journal of Social Policy , 42, 1– 22. Google Scholar CrossRef Search ADS   Flannery M. J., Hankins K. W. ( 2013) ‘ Estimating Dynamic Panel Models in Corporate Finance’, Journal of Corporate Finance , 19, 1– 19. Google Scholar CrossRef Search ADS   Franzese R. J., Hays J. C. ( 2006) ‘Strategic Interaction among EU Governments in Active Labor Market Policy-Making: Subsidiarity and Policy Coordination under the European Employment Strategy’, European Union Politics , 7, 167– 189. Google Scholar CrossRef Search ADS   Furåker B. ( 1976) ‘Stat och arbetsmarknad: Studier i svensk rörlighetspolitik [State and Labor Market: Studies in Swedish Mobility Politics]’, Arkiv Avhandlingsserie, 0347-4909, 5. Gaston N., Rajaguru G. ( 2008) ‘ The Rise (and Fall) of Labour Market Programmes: Domestic vs. Global Factors’, Oxford Economic Papers , 60, 619– 648. Google Scholar CrossRef Search ADS   Grubb D., Puymoyen A. ( 2008) Long Time Series for Public Expenditure on Labour Market Programmes , Report, OECD Publishing. Hemerijck A. ( 2013) Changing Welfare States , Oxford, Oxford University Press. Hibbs D. A. ( 1977) ‘ Political Parties and Macroeconomic Policy’, The American Political Science Review , 71, 1467– 1487. Google Scholar CrossRef Search ADS   Hijzen A., Venn D. ( 2011) The Role of Short-Time Work Schemes during the 2008-09 Recession , Report, OECD Publishing. Huber E., Stephens J. D. ( 2001) Development and Crisis of the Welfare State: Parties and Policies in Global Markets , Chicago, University of Chicago Press. Huo J., Nelson M., Stephens J. D. ( 2008) ‘Decommodification and Activation in Social Democratic Policy: Resolving the Paradox’, Journal of European Social Policy , 18, 5– 20. Google Scholar CrossRef Search ADS   Iversen T., Stephens J. D. ( 2008) ‘ Partisan Politics, the Welfare State, and Three Worlds of Human Capital Formation’, Comparative Political Studies , 41, 600– 637. Google Scholar CrossRef Search ADS   Janoski T. ( 1990) The Political Economy of Unemployment: Active Labor Market Policy in West Germany and the United States . Berkeley, CA, University of California Press. Janoski T. ( 1994) ‘Direct State Intervention in the Labor Market: The Explanation of Active Labor Market Policy from 1950 to 1988 in Social Democratic, Conservative and Liberal Regimes’. In Janoski T., Hicks A. (eds) The Comparative Political Economy of the Welfare State , New York, NY, Cambridge University Press, pp. 54– 92. Kiviet J. F. ( 1995) ‘ On Bias, Inconsistency, and Efficiency of Various Estimators in Dynamic Panel Data Models’, Journal of Econometrics , 68, 53– 78. Google Scholar CrossRef Search ADS   Kluve J. ( 2010) ‘ The Effectiveness of European Active Labor Market Programs’, Labour Economics , 17, 904– 918. Google Scholar CrossRef Search ADS   Korpi W., Palme J. ( 1998) ‘ The Paradox of Redistribution and Strategies of Equality: Welfare State Institutions, Inequality, and Poverty in the Western Countries’, American Sociological Review , 63, 661– 687. Google Scholar CrossRef Search ADS   Lindvall J. ( 2010) Mass Unemployment and the State , Oxford, Oxford University Press. Martin J. P. ( 2014) Activation and Active Labour Market Policies in OECD Countries: Stylized Facts and Evidence on Their Effectiveness, IZA Policy Paper No. 84, IZA. Morel N., Palier B., Palme J. ( 2012) Towards a Social Investment Welfare State?: Ideas, Policies and Challenges , Bristol, Policy and Publishing House. Mucciaroni G. ( 1990) The Political Failure of Employment Policy, 1945–1982 , Pittsburgh, PA, University of Pittsburgh Press. Nelson M. ( 2013) ‘ Making Markets with Active Labor Market Policies: the Influence of Political Parties, Welfare State Regimes, and Economic Change on Spending on Different Types of Policies’, European Political Science Review , 5, 255– 277. Google Scholar CrossRef Search ADS   Nickell S. ( 1981) ‘ Biases in Dynamic Models with Fixed Effects’, Econometrica , 49, 1417– 1426. Google Scholar CrossRef Search ADS   Pierson P. ( 1994) Dismantling the Welfare State?: Reagan, Thatcher and the Politics of Retrenchment , Cambridge, Cambridge University Press. Powell G. B.Jr., Whitten G. D. ( 1993) ‘ A Cross-National Analysis of Economic Voting: Taking Account of the Political Context’, American Journal of Political Science , 37, 391– 414. Google Scholar CrossRef Search ADS   Rothstein B. ( 1996) The Social Democratic State: the Swedish Model and the Bureaucratic Problem of Social Reforms , Pittsburgh, PA, University of Pittsburgh Press. Rueda D. ( 2005) ‘ Insider–Outsider Politics in Industrialized Democracies: The Challenge to Social Democratic Parties’, American Political Science Review , 99, 61– 74. Google Scholar CrossRef Search ADS   Rueda D. ( 2006) ‘ Social Democracy and Active Labour-Market Policies: Insiders, Outsiders and the Politics of Employment Promotion’, British Journal of Political Science , 36, 385– 406. Google Scholar CrossRef Search ADS   Rueda D. ( 2007) Social Democracy inside Out: Partisanship and Labor Market Policy in Industrialized Democracies , Oxford, Oxford University Press. Scruggs L., Allan J. ( 2006) ‘ Welfare-State Decommodification in 18 OECD Countries: A Replication and Revision’, Journal of European Social Policy , 16, 55– 72. Google Scholar CrossRef Search ADS   Swenson P. A. ( 2002) Capitalists against Markets: The Making of Labor Markets and Welfare States in the United States and Sweden , New York, Oxford University Press. Tepe M., Vanhuysse P. ( 2013) ‘ Parties, Unions and Activation Strategies: The Context- Dependent Politics of Active Labour Market Policy Spending’, Political Studies , 61, 480– 504. Google Scholar CrossRef Search ADS   Thelen K. A. ( 2014) Varieties of Liberalization and the New Politics of Social Solidarity , New York, NY, Cambridge University Press. van der Brug W., van der EijK C., Franklin M. ( 2007) The Economy and the Vote: Economic Conditions and Elections in Fifteen Countries , Cambridge, UK, Cambridge University Press. van Vliet O., Koster F. ( 2011) ‘ Europeanization and the Political Economy of Active Labour Market Policies’, European Union Politics , 12, 217– 239. Google Scholar CrossRef Search ADS   Vlandas T. ( 2013) ‘ Mixing Apples with Oranges? Partisanship and Active Labour Market Policies in Europe’, Journal of European Social Policy , 23, 3– 20. Google Scholar CrossRef Search ADS   Weishaupt T. J. ( 2011) From the Manpower Revolution to the Activation Paradigm: Explaining Institutional Continuity and Change in an Integrating Europe , Amsterdam, Amsterdam University Press. Whitten G. D., Palmer H. D. ( 1999) ‘ Cross-National Analyses of Economic Voting’, Electoral Studies , 18, 49– 67. Google Scholar CrossRef Search ADS   Appendix Table A1 provides descriptions of the six traditional ALMP program categories and statistics on the share of observations in each category that has Core target group and Broken unemployment characteristics. Table A2 presents summary statistics for all the baseline variables included in the models in the article. Table A3 illustrates the structure of the data set by presenting all the utilized program-year-level data for two observations of two different programs—one in Sweden and one in Germany. Table A4 reports further detail regarding the partial conditional effects of Left-wing government strength on Scope that are presented in Figure 2. Table A5 reports five models that are identical to those in Table 2, but they are applied to the logged annual program Expenditure instead of Scope. The model with the lagged dependent variable (A4) and the LSDV(c) model (A5) both produce coefficients of interest that are only slightly smaller than those for Scope and that are still statistically significant. Finally, the models in Table A6 split the sample into five country clusters to assess whether any particular welfare regime drives the results. Precision is reduced in most models, but with few exceptions—notably that Broken unemployment × Left-wing government strength is smaller for the continental countries—the estimates are in the same direction and of a similar or larger size. Table A1. Distribution of program characteristics across ALMP program categories       Characteristics (%)   Category  Description  Observation  Core target group  Broken unemployment  Training  Covers measures that aim to improve participants’ employability through training. The measures should include some evidence of classroom teaching or supervision, specifically for the purpose of instruction. Short courses that only develop a person’s ability to get a job are considered labor market services and fall outside this category.  1997  47  62  Employment incentives  Covers measures that facilitate the recruitment of unemployed persons and other target groups or that help ensure the continued employment of persons at risk of involuntary job loss. It refers to subsidies for open market jobs that might exist or be created without the public subsidy and that will hopefully be sustainable after the end of the subsidy period.  1865  72  84  Sheltered and supported employment and rehabilitation  Covers measures that aim to promote the labor market integration of persons with reduced working capacity through sheltered or supported employment or through rehabilitation.  690  43  64  Direct job creation  Covers measures that create additional jobs, usually of community benefit or social use, to find employment for the long-term unemployed or persons who are otherwise difficult to place. It refers to subsidies for temporary, nonmarket jobs that would not exist or be created without public intervention.  714  75  69  Start-up incentives  Covers measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed. Assistance may take the form of direct cash benefits or indirect support, including loans, facility provision and business advice.  402  60  91  Labor market services  Refers to labor market interventions where participants’ main activity is job-search related and where participation usually does not result in a change in labor market status. Services also cover Public Employment Service functions that are not directly linked to participants, such as services for employers, administrative functions and general overhead.  –  –  –  Total    5668  60  72        Characteristics (%)   Category  Description  Observation  Core target group  Broken unemployment  Training  Covers measures that aim to improve participants’ employability through training. The measures should include some evidence of classroom teaching or supervision, specifically for the purpose of instruction. Short courses that only develop a person’s ability to get a job are considered labor market services and fall outside this category.  1997  47  62  Employment incentives  Covers measures that facilitate the recruitment of unemployed persons and other target groups or that help ensure the continued employment of persons at risk of involuntary job loss. It refers to subsidies for open market jobs that might exist or be created without the public subsidy and that will hopefully be sustainable after the end of the subsidy period.  1865  72  84  Sheltered and supported employment and rehabilitation  Covers measures that aim to promote the labor market integration of persons with reduced working capacity through sheltered or supported employment or through rehabilitation.  690  43  64  Direct job creation  Covers measures that create additional jobs, usually of community benefit or social use, to find employment for the long-term unemployed or persons who are otherwise difficult to place. It refers to subsidies for temporary, nonmarket jobs that would not exist or be created without public intervention.  714  75  69  Start-up incentives  Covers measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed. Assistance may take the form of direct cash benefits or indirect support, including loans, facility provision and business advice.  402  60  91  Labor market services  Refers to labor market interventions where participants’ main activity is job-search related and where participation usually does not result in a change in labor market status. Services also cover Public Employment Service functions that are not directly linked to participants, such as services for employers, administrative functions and general overhead.  –  –  –  Total    5668  60  72  Source: Eurostat (2013), European Commission (2015). The EU LMP database also includes the two other categories Out-of-work income maintenance and support and Early retirement. These LMP categories are usually referred to as ‘support’ or ‘passive’ schemes. Table A1. Distribution of program characteristics across ALMP program categories       Characteristics (%)   Category  Description  Observation  Core target group  Broken unemployment  Training  Covers measures that aim to improve participants’ employability through training. The measures should include some evidence of classroom teaching or supervision, specifically for the purpose of instruction. Short courses that only develop a person’s ability to get a job are considered labor market services and fall outside this category.  1997  47  62  Employment incentives  Covers measures that facilitate the recruitment of unemployed persons and other target groups or that help ensure the continued employment of persons at risk of involuntary job loss. It refers to subsidies for open market jobs that might exist or be created without the public subsidy and that will hopefully be sustainable after the end of the subsidy period.  1865  72  84  Sheltered and supported employment and rehabilitation  Covers measures that aim to promote the labor market integration of persons with reduced working capacity through sheltered or supported employment or through rehabilitation.  690  43  64  Direct job creation  Covers measures that create additional jobs, usually of community benefit or social use, to find employment for the long-term unemployed or persons who are otherwise difficult to place. It refers to subsidies for temporary, nonmarket jobs that would not exist or be created without public intervention.  714  75  69  Start-up incentives  Covers measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed. Assistance may take the form of direct cash benefits or indirect support, including loans, facility provision and business advice.  402  60  91  Labor market services  Refers to labor market interventions where participants’ main activity is job-search related and where participation usually does not result in a change in labor market status. Services also cover Public Employment Service functions that are not directly linked to participants, such as services for employers, administrative functions and general overhead.  –  –  –  Total    5668  60  72        Characteristics (%)   Category  Description  Observation  Core target group  Broken unemployment  Training  Covers measures that aim to improve participants’ employability through training. The measures should include some evidence of classroom teaching or supervision, specifically for the purpose of instruction. Short courses that only develop a person’s ability to get a job are considered labor market services and fall outside this category.  1997  47  62  Employment incentives  Covers measures that facilitate the recruitment of unemployed persons and other target groups or that help ensure the continued employment of persons at risk of involuntary job loss. It refers to subsidies for open market jobs that might exist or be created without the public subsidy and that will hopefully be sustainable after the end of the subsidy period.  1865  72  84  Sheltered and supported employment and rehabilitation  Covers measures that aim to promote the labor market integration of persons with reduced working capacity through sheltered or supported employment or through rehabilitation.  690  43  64  Direct job creation  Covers measures that create additional jobs, usually of community benefit or social use, to find employment for the long-term unemployed or persons who are otherwise difficult to place. It refers to subsidies for temporary, nonmarket jobs that would not exist or be created without public intervention.  714  75  69  Start-up incentives  Covers measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed. Assistance may take the form of direct cash benefits or indirect support, including loans, facility provision and business advice.  402  60  91  Labor market services  Refers to labor market interventions where participants’ main activity is job-search related and where participation usually does not result in a change in labor market status. Services also cover Public Employment Service functions that are not directly linked to participants, such as services for employers, administrative functions and general overhead.  –  –  –  Total    5668  60  72  Source: Eurostat (2013), European Commission (2015). The EU LMP database also includes the two other categories Out-of-work income maintenance and support and Early retirement. These LMP categories are usually referred to as ‘support’ or ‘passive’ schemes. Table A2. Summary statistics Variable  Mean  Std. Dev.  Min  Max  N  Dependent variables   Scope (log)  7.77  2.51  0  14.84  5668   Expenditure (log)  2.41  2.64  −4.77  8.81  7025  Country-level variables   Left-wing government strength  0.35  0.34  0  1  5668   Real GDP  6.63E+7  8.07E+7  6.08E+5  3.1E+8  5668   Real GDP growth  1.69  3.10  −14.81  11.62  5668   Unemployed and inactive (log)  14.9  1.29  11.58  16.79  5668   Government deficit  −2.07  4.71  −32.55  18.70  5668   Government debt  69.26  32.85  7.234  144.0  5668   Number of programs in operation  33.67  16.22  4  66  5668  Program-level variables   Core target group  0.60  0.49  0  1  5668   Broken unemployment  0.72  0.45  0  1  5668   Training  0.35  0.48  0  1  5668   Employment incentives  0.33  0.47  0  1  5668   Sheltered employment and rehabilitation  0.12  0.33  0  1  5668   Direct job creation  0.13  0.33  0  1  5668   Start-up incentives  0.07  0.26  0  1  5668  Variable  Mean  Std. Dev.  Min  Max  N  Dependent variables   Scope (log)  7.77  2.51  0  14.84  5668   Expenditure (log)  2.41  2.64  −4.77  8.81  7025  Country-level variables   Left-wing government strength  0.35  0.34  0  1  5668   Real GDP  6.63E+7  8.07E+7  6.08E+5  3.1E+8  5668   Real GDP growth  1.69  3.10  −14.81  11.62  5668   Unemployed and inactive (log)  14.9  1.29  11.58  16.79  5668   Government deficit  −2.07  4.71  −32.55  18.70  5668   Government debt  69.26  32.85  7.234  144.0  5668   Number of programs in operation  33.67  16.22  4  66  5668  Program-level variables   Core target group  0.60  0.49  0  1  5668   Broken unemployment  0.72  0.45  0  1  5668   Training  0.35  0.48  0  1  5668   Employment incentives  0.33  0.47  0  1  5668   Sheltered employment and rehabilitation  0.12  0.33  0  1  5668   Direct job creation  0.13  0.33  0  1  5668   Start-up incentives  0.07  0.26  0  1  5668  Table A2. Summary statistics Variable  Mean  Std. Dev.  Min  Max  N  Dependent variables   Scope (log)  7.77  2.51  0  14.84  5668   Expenditure (log)  2.41  2.64  −4.77  8.81  7025  Country-level variables   Left-wing government strength  0.35  0.34  0  1  5668   Real GDP  6.63E+7  8.07E+7  6.08E+5  3.1E+8  5668   Real GDP growth  1.69  3.10  −14.81  11.62  5668   Unemployed and inactive (log)  14.9  1.29  11.58  16.79  5668   Government deficit  −2.07  4.71  −32.55  18.70  5668   Government debt  69.26  32.85  7.234  144.0  5668   Number of programs in operation  33.67  16.22  4  66  5668  Program-level variables   Core target group  0.60  0.49  0  1  5668   Broken unemployment  0.72  0.45  0  1  5668   Training  0.35  0.48  0  1  5668   Employment incentives  0.33  0.47  0  1  5668   Sheltered employment and rehabilitation  0.12  0.33  0  1  5668   Direct job creation  0.13  0.33  0  1  5668   Start-up incentives  0.07  0.26  0  1  5668  Variable  Mean  Std. Dev.  Min  Max  N  Dependent variables   Scope (log)  7.77  2.51  0  14.84  5668   Expenditure (log)  2.41  2.64  −4.77  8.81  7025  Country-level variables   Left-wing government strength  0.35  0.34  0  1  5668   Real GDP  6.63E+7  8.07E+7  6.08E+5  3.1E+8  5668   Real GDP growth  1.69  3.10  −14.81  11.62  5668   Unemployed and inactive (log)  14.9  1.29  11.58  16.79  5668   Government deficit  −2.07  4.71  −32.55  18.70  5668   Government debt  69.26  32.85  7.234  144.0  5668   Number of programs in operation  33.67  16.22  4  66  5668  Program-level variables   Core target group  0.60  0.49  0  1  5668   Broken unemployment  0.72  0.45  0  1  5668   Training  0.35  0.48  0  1  5668   Employment incentives  0.33  0.47  0  1  5668   Sheltered employment and rehabilitation  0.12  0.33  0  1  5668   Direct job creation  0.13  0.33  0  1  5668   Start-up incentives  0.07  0.26  0  1  5668  Table A3. Comparison of two program-year observations from two ALMP programs Observation ID  SE84_2008  DE16_2008  Name in English  New start jobs  Recruitment subsidy for new businesses  Intervention ID (panel ID)  SE84  DE16  Year  2008  2008  Country  Sweden  Germany  Program category  4. Employment incentives  4. Employment incentives  Year started  2007  1998  Year ended  (ongoing as of 2013)  2010  Target groups   Registered unemployed  All; LTU; disabled; immigrants/ethnic min.  All; public priorities and other   Other registered jobseekers  All; disabled; immigrants/ethnic min.  –   Not registered  All; disabled; immigrants/ethnic min.  –   Employed  –  –  Core target group  Non-core  Core  Treatment of unemployment spell  Broken  Broken  Scope (Participant-years)  15 921  4655  Expenditure (M €, 2005 prices)  77.6  50.1  Scope (log)  ln(15,921) = 9.68  ln(4,655) = 8.45  Expenditure (log)  ln(77.6) = 4.35  ln(50.1) = 3.91  Observation ID  SE84_2008  DE16_2008  Name in English  New start jobs  Recruitment subsidy for new businesses  Intervention ID (panel ID)  SE84  DE16  Year  2008  2008  Country  Sweden  Germany  Program category  4. Employment incentives  4. Employment incentives  Year started  2007  1998  Year ended  (ongoing as of 2013)  2010  Target groups   Registered unemployed  All; LTU; disabled; immigrants/ethnic min.  All; public priorities and other   Other registered jobseekers  All; disabled; immigrants/ethnic min.  –   Not registered  All; disabled; immigrants/ethnic min.  –   Employed  –  –  Core target group  Non-core  Core  Treatment of unemployment spell  Broken  Broken  Scope (Participant-years)  15 921  4655  Expenditure (M €, 2005 prices)  77.6  50.1  Scope (log)  ln(15,921) = 9.68  ln(4,655) = 8.45  Expenditure (log)  ln(77.6) = 4.35  ln(50.1) = 3.91  Source:European Commission (2015). Table A3. Comparison of two program-year observations from two ALMP programs Observation ID  SE84_2008  DE16_2008  Name in English  New start jobs  Recruitment subsidy for new businesses  Intervention ID (panel ID)  SE84  DE16  Year  2008  2008  Country  Sweden  Germany  Program category  4. Employment incentives  4. Employment incentives  Year started  2007  1998  Year ended  (ongoing as of 2013)  2010  Target groups   Registered unemployed  All; LTU; disabled; immigrants/ethnic min.  All; public priorities and other   Other registered jobseekers  All; disabled; immigrants/ethnic min.  –   Not registered  All; disabled; immigrants/ethnic min.  –   Employed  –  –  Core target group  Non-core  Core  Treatment of unemployment spell  Broken  Broken  Scope (Participant-years)  15 921  4655  Expenditure (M €, 2005 prices)  77.6  50.1  Scope (log)  ln(15,921) = 9.68  ln(4,655) = 8.45  Expenditure (log)  ln(77.6) = 4.35  ln(50.1) = 3.91  Observation ID  SE84_2008  DE16_2008  Name in English  New start jobs  Recruitment subsidy for new businesses  Intervention ID (panel ID)  SE84  DE16  Year  2008  2008  Country  Sweden  Germany  Program category  4. Employment incentives  4. Employment incentives  Year started  2007  1998  Year ended  (ongoing as of 2013)  2010  Target groups   Registered unemployed  All; LTU; disabled; immigrants/ethnic min.  All; public priorities and other   Other registered jobseekers  All; disabled; immigrants/ethnic min.  –   Not registered  All; disabled; immigrants/ethnic min.  –   Employed  –  –  Core target group  Non-core  Core  Treatment of unemployment spell  Broken  Broken  Scope (Participant-years)  15 921  4655  Expenditure (M €, 2005 prices)  77.6  50.1  Scope (log)  ln(15,921) = 9.68  ln(4,655) = 8.45  Expenditure (log)  ln(77.6) = 4.35  ln(50.1) = 3.91  Source:European Commission (2015). Table A4. Partial conditional effects of left-wing government on ALMP ‘effort’ (Scope) Program characteristic   Δy/Δx  Std. Dev.  P-value  e^(Δy/Δx)  Core  0.393  0.064  0.000  1.482  Non-core  0.006  0.078  0.934  1.006  Broken  0.376  0.053  0.000  1.457  Continuous  −0.097  0.087  0.264  0.907  Program characteristic   Δy/Δx  Std. Dev.  P-value  e^(Δy/Δx)  Core  0.393  0.064  0.000  1.482  Non-core  0.006  0.078  0.934  1.006  Broken  0.376  0.053  0.000  1.457  Continuous  −0.097  0.087  0.264  0.907  Effects of a change from 0% to 100% left-wing cabinet seats, with all other variables at their means. Estimations produced from Model 5. Table A4. Partial conditional effects of left-wing government on ALMP ‘effort’ (Scope) Program characteristic   Δy/Δx  Std. Dev.  P-value  e^(Δy/Δx)  Core  0.393  0.064  0.000  1.482  Non-core  0.006  0.078  0.934  1.006  Broken  0.376  0.053  0.000  1.457  Continuous  −0.097  0.087  0.264  0.907  Program characteristic   Δy/Δx  Std. Dev.  P-value  e^(Δy/Δx)  Core  0.393  0.064  0.000  1.482  Non-core  0.006  0.078  0.934  1.006  Broken  0.376  0.053  0.000  1.457  Continuous  −0.097  0.087  0.264  0.907  Effects of a change from 0% to 100% left-wing cabinet seats, with all other variables at their means. Estimations produced from Model 5. Table A5. Regression results for models of Expenditure   (A1)  (A2)  (A3)  (A4)  (A5)  Lagged dependent variable        0.60***  0.70***          (0.03)  (0.01)  Left-wing government strength  0.58***  −0.06  0.03  −0.12  −0.15    (0.11)  (0.22)  (0.22)  (0.16)  (0.11)  Core target group × Left-wing gov’t    0.27  0.26  0.34***  0.34***      (0.18)  (0.20)  (0.11)  (0.08)  Broken unemployment × Left-wing gov’t    0.60***  0.65***  0.39**  0.37***      (0.20)  (0.19)  (0.15)  (0.10)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.01  0.01  0.01  0.02  0.02***    (0.01)  (0.01)  (0.01)  (0.01)  (0.01)   Unemployed and inactive (log)  1.57**  1.66**  1.59**  1.53***  1.59***    (0.67)  (0.68)  (0.66)  (0.38)  (0.23)   Government deficit  −0.02  −0.02  −0.02  −0.01  −0.01    (0.01)  (0.01)  (0.01)  (0.01)  (0.00)   Government debt  −0.01***  −0.01***  −0.01***  −0.01***  −0.01***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.01  −0.01  −0.01  −0.00  −0.00    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.12  −0.24*  −0.21  −0.23***  −0.23***    (0.14)  (0.14)  (0.13)  (0.07)  (0.09)   Broken unemployment  0.08  −0.05  −0.07  −0.03  −0.01    (0.12)  (0.12)  (0.12)  (0.07)  (0.10)   Employment incentives × Left-wing gov’t      −0.19  −0.10  −0.06        (0.14)  (0.11)  (0.09)   Sheltered empl. and rehab. × Left-wing gov’t      −0.44**  −0.31**  −0.29**        (0.18)  (0.12)  (0.13)   Direct job creation × Left-wing gov’t      0.42  0.34  0.32**        (0.48)  (0.29)  (0.14)   Start-up incentives × Left-wing gov’t      −0.36  −0.18  −0.16        (0.39)  (0.27)  (0.13)  Constant  −20.41**  −21.47**  −20.55**  −20.93***      (9.91)  (10.01)  (9.83)  (5.67)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  7025  7025  7025  7025  7025  Programs  1052  1052  1052  1052  1052    (A1)  (A2)  (A3)  (A4)  (A5)  Lagged dependent variable        0.60***  0.70***          (0.03)  (0.01)  Left-wing government strength  0.58***  −0.06  0.03  −0.12  −0.15    (0.11)  (0.22)  (0.22)  (0.16)  (0.11)  Core target group × Left-wing gov’t    0.27  0.26  0.34***  0.34***      (0.18)  (0.20)  (0.11)  (0.08)  Broken unemployment × Left-wing gov’t    0.60***  0.65***  0.39**  0.37***      (0.20)  (0.19)  (0.15)  (0.10)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.01  0.01  0.01  0.02  0.02***    (0.01)  (0.01)  (0.01)  (0.01)  (0.01)   Unemployed and inactive (log)  1.57**  1.66**  1.59**  1.53***  1.59***    (0.67)  (0.68)  (0.66)  (0.38)  (0.23)   Government deficit  −0.02  −0.02  −0.02  −0.01  −0.01    (0.01)  (0.01)  (0.01)  (0.01)  (0.00)   Government debt  −0.01***  −0.01***  −0.01***  −0.01***  −0.01***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.01  −0.01  −0.01  −0.00  −0.00    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.12  −0.24*  −0.21  −0.23***  −0.23***    (0.14)  (0.14)  (0.13)  (0.07)  (0.09)   Broken unemployment  0.08  −0.05  −0.07  −0.03  −0.01    (0.12)  (0.12)  (0.12)  (0.07)  (0.10)   Employment incentives × Left-wing gov’t      −0.19  −0.10  −0.06        (0.14)  (0.11)  (0.09)   Sheltered empl. and rehab. × Left-wing gov’t      −0.44**  −0.31**  −0.29**        (0.18)  (0.12)  (0.13)   Direct job creation × Left-wing gov’t      0.42  0.34  0.32**        (0.48)  (0.29)  (0.14)   Start-up incentives × Left-wing gov’t      −0.36  −0.18  −0.16        (0.39)  (0.27)  (0.13)  Constant  −20.41**  −21.47**  −20.55**  −20.93***      (9.91)  (10.01)  (9.83)  (5.67)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  7025  7025  7025  7025  7025  Programs  1052  1052  1052  1052  1052  Models A1–A4: Cluster-robust standard errors in parentheses (clustered at the country level). Model A5: Bootstrapped standard errors in parentheses. Bias correction initialized by Arellano and Bond estimator (Bruno 2005). *P < 0.10, **P < 0.05, ***P < 0.01. Table A5. Regression results for models of Expenditure   (A1)  (A2)  (A3)  (A4)  (A5)  Lagged dependent variable        0.60***  0.70***          (0.03)  (0.01)  Left-wing government strength  0.58***  −0.06  0.03  −0.12  −0.15    (0.11)  (0.22)  (0.22)  (0.16)  (0.11)  Core target group × Left-wing gov’t    0.27  0.26  0.34***  0.34***      (0.18)  (0.20)  (0.11)  (0.08)  Broken unemployment × Left-wing gov’t    0.60***  0.65***  0.39**  0.37***      (0.20)  (0.19)  (0.15)  (0.10)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.01  0.01  0.01  0.02  0.02***    (0.01)  (0.01)  (0.01)  (0.01)  (0.01)   Unemployed and inactive (log)  1.57**  1.66**  1.59**  1.53***  1.59***    (0.67)  (0.68)  (0.66)  (0.38)  (0.23)   Government deficit  −0.02  −0.02  −0.02  −0.01  −0.01    (0.01)  (0.01)  (0.01)  (0.01)  (0.00)   Government debt  −0.01***  −0.01***  −0.01***  −0.01***  −0.01***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.01  −0.01  −0.01  −0.00  −0.00    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.12  −0.24*  −0.21  −0.23***  −0.23***    (0.14)  (0.14)  (0.13)  (0.07)  (0.09)   Broken unemployment  0.08  −0.05  −0.07  −0.03  −0.01    (0.12)  (0.12)  (0.12)  (0.07)  (0.10)   Employment incentives × Left-wing gov’t      −0.19  −0.10  −0.06        (0.14)  (0.11)  (0.09)   Sheltered empl. and rehab. × Left-wing gov’t      −0.44**  −0.31**  −0.29**        (0.18)  (0.12)  (0.13)   Direct job creation × Left-wing gov’t      0.42  0.34  0.32**        (0.48)  (0.29)  (0.14)   Start-up incentives × Left-wing gov’t      −0.36  −0.18  −0.16        (0.39)  (0.27)  (0.13)  Constant  −20.41**  −21.47**  −20.55**  −20.93***      (9.91)  (10.01)  (9.83)  (5.67)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  7025  7025  7025  7025  7025  Programs  1052  1052  1052  1052  1052    (A1)  (A2)  (A3)  (A4)  (A5)  Lagged dependent variable        0.60***  0.70***          (0.03)  (0.01)  Left-wing government strength  0.58***  −0.06  0.03  −0.12  −0.15    (0.11)  (0.22)  (0.22)  (0.16)  (0.11)  Core target group × Left-wing gov’t    0.27  0.26  0.34***  0.34***      (0.18)  (0.20)  (0.11)  (0.08)  Broken unemployment × Left-wing gov’t    0.60***  0.65***  0.39**  0.37***      (0.20)  (0.19)  (0.15)  (0.10)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.01  0.01  0.01  0.02  0.02***    (0.01)  (0.01)  (0.01)  (0.01)  (0.01)   Unemployed and inactive (log)  1.57**  1.66**  1.59**  1.53***  1.59***    (0.67)  (0.68)  (0.66)  (0.38)  (0.23)   Government deficit  −0.02  −0.02  −0.02  −0.01  −0.01    (0.01)  (0.01)  (0.01)  (0.01)  (0.00)   Government debt  −0.01***  −0.01***  −0.01***  −0.01***  −0.01***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.01  −0.01  −0.01  −0.00  −0.00    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.12  −0.24*  −0.21  −0.23***  −0.23***    (0.14)  (0.14)  (0.13)  (0.07)  (0.09)   Broken unemployment  0.08  −0.05  −0.07  −0.03  −0.01    (0.12)  (0.12)  (0.12)  (0.07)  (0.10)   Employment incentives × Left-wing gov’t      −0.19  −0.10  −0.06        (0.14)  (0.11)  (0.09)   Sheltered empl. and rehab. × Left-wing gov’t      −0.44**  −0.31**  −0.29**        (0.18)  (0.12)  (0.13)   Direct job creation × Left-wing gov’t      0.42  0.34  0.32**        (0.48)  (0.29)  (0.14)   Start-up incentives × Left-wing gov’t      −0.36  −0.18  −0.16        (0.39)  (0.27)  (0.13)  Constant  −20.41**  −21.47**  −20.55**  −20.93***      (9.91)  (10.01)  (9.83)  (5.67)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  7025  7025  7025  7025  7025  Programs  1052  1052  1052  1052  1052  Models A1–A4: Cluster-robust standard errors in parentheses (clustered at the country level). Model A5: Bootstrapped standard errors in parentheses. Bias correction initialized by Arellano and Bond estimator (Bruno 2005). *P < 0.10, **P < 0.05, ***P < 0.01. Table A6. Regression results for models of Scope, by welfare regime cluster   (A6)  (A7)  (A8)  (A9)  (A10)  Lagged dependent variable  0.69***  0.65***  0.59***  0.65***  0.35***    (0.07)  (0.08)  (0.11)  (0.09)  (0.04)  Left-wing government strength  −0.58  −0.20  −0.60  −0.54*  −0.85**    (0.40)  (0.27)  (1.68)  (0.30)  (0.40)  Core target group × Left-wing gov’t  0.31  0.48**  0.83  0.73***  0.33    (0.23)  (0.23)  (1.23)  (0.28)  (0.29)  Broken unemployment × Left-wing gov’t  0.63*  0.13  2.91  0.38  0.89***    (0.33)  (0.26)  (1.75)  (0.28)  (0.31)  Country-level controls             Real GDP  0.00  −0.00***  −0.00  −0.00  0.00*    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.04**  −0.02  −0.12  0.08**  0.05***    (0.02)  (0.02)  (0.11)  (0.03)  (0.01)   Unemployed and inactive (log)  −1.05  0.28  3.85  0.19  −0.06    (0.73)  (0.78)  (4.30)  (1.56)  (0.98)   Government deficit  −0.00  −0.03**  0.16  −0.01  0.04*    (0.02)  (0.02)  (0.11)  (0.02)  (0.02)   Government debt  0.01  −0.01  0.26  −0.00  0.02**    (0.00)  (0.00)  (0.17)  (0.01)  (0.01)   Number of programs in operation  −0.02  −0.00  −0.00  −0.01*  −0.00    (0.02)  (0.01)  (0.06)  (0.01)  (0.01)  Program-level controls             Core target group  −0.53**  −0.23  0.50  −0.20  −1.16***    (0.25)  (0.19)  (0.65)  (0.24)  (0.30)   Broken unemployment  −0.16  0.66*  −0.02  −0.84***  −0.65    (0.13)  (0.34)  (0.18)  (0.18)  (0.47)   Employment incentives × Left-wing gov’t  0.24  0.09  0.95  −0.22  −0.53    (0.35)  (0.25)  (0.97)  (0.27)  (0.41)   Sheltered empl. and rehab. × Left-wing gov’t  −0.14  −0.46**  0.29  −0.21  −0.50    (0.25)  (0.22)  (1.02)  (0.26)  (0.39)   Direct job creation × Left-wing gov’t  0.11  0.72**  0.04  −0.45  −0.09    (0.36)  (0.32)  (0.74)  (0.30)  (0.52)   Start-up incentives × Left-wing gov’t  0.05  0.18  1.02  0.07  −0.40    (0.25)  (0.60)  (1.24)  (0.34)  (0.45)  Constant  17.56*  0.06  −61.98  1.04  6.43    (10.08)  (11.97)  (62.39)  (23.82)  (14.17)  Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  929  2095  333  1143  1168  Programs  112  276  41  222  245    (A6)  (A7)  (A8)  (A9)  (A10)  Lagged dependent variable  0.69***  0.65***  0.59***  0.65***  0.35***    (0.07)  (0.08)  (0.11)  (0.09)  (0.04)  Left-wing government strength  −0.58  −0.20  −0.60  −0.54*  −0.85**    (0.40)  (0.27)  (1.68)  (0.30)  (0.40)  Core target group × Left-wing gov’t  0.31  0.48**  0.83  0.73***  0.33    (0.23)  (0.23)  (1.23)  (0.28)  (0.29)  Broken unemployment × Left-wing gov’t  0.63*  0.13  2.91  0.38  0.89***    (0.33)  (0.26)  (1.75)  (0.28)  (0.31)  Country-level controls             Real GDP  0.00  −0.00***  −0.00  −0.00  0.00*    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.04**  −0.02  −0.12  0.08**  0.05***    (0.02)  (0.02)  (0.11)  (0.03)  (0.01)   Unemployed and inactive (log)  −1.05  0.28  3.85  0.19  −0.06    (0.73)  (0.78)  (4.30)  (1.56)  (0.98)   Government deficit  −0.00  −0.03**  0.16  −0.01  0.04*    (0.02)  (0.02)  (0.11)  (0.02)  (0.02)   Government debt  0.01  −0.01  0.26  −0.00  0.02**    (0.00)  (0.00)  (0.17)  (0.01)  (0.01)   Number of programs in operation  −0.02  −0.00  −0.00  −0.01*  −0.00    (0.02)  (0.01)  (0.06)  (0.01)  (0.01)  Program-level controls             Core target group  −0.53**  −0.23  0.50  −0.20  −1.16***    (0.25)  (0.19)  (0.65)  (0.24)  (0.30)   Broken unemployment  −0.16  0.66*  −0.02  −0.84***  −0.65    (0.13)  (0.34)  (0.18)  (0.18)  (0.47)   Employment incentives × Left-wing gov’t  0.24  0.09  0.95  −0.22  −0.53    (0.35)  (0.25)  (0.97)  (0.27)  (0.41)   Sheltered empl. and rehab. × Left-wing gov’t  −0.14  −0.46**  0.29  −0.21  −0.50    (0.25)  (0.22)  (1.02)  (0.26)  (0.39)   Direct job creation × Left-wing gov’t  0.11  0.72**  0.04  −0.45  −0.09    (0.36)  (0.32)  (0.74)  (0.30)  (0.52)   Start-up incentives × Left-wing gov’t  0.05  0.18  1.02  0.07  −0.40    (0.25)  (0.60)  (1.24)  (0.34)  (0.45)  Constant  17.56*  0.06  −61.98  1.04  6.43    (10.08)  (11.97)  (62.39)  (23.82)  (14.17)  Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  929  2095  333  1143  1168  Programs  112  276  41  222  245  Cluster-robust standard errors in parentheses (clustered at the program level). *P < 0.10, **P < 0.05, ***P < 0.01. Model A6: Scandinavian (DK, FI, NO, SE). Model A7: Continental (AT, BE, DE, FR, NL, LU). Model A8: Anglo-Saxon (IE, UK). Model A9: Southern (CY, EL, ES, IT, MT, PT). Model A10: Central and Eastern European (BG, CZ, EE, HU, LT, LV, PL, RO, SI, SK). Table A6. Regression results for models of Scope, by welfare regime cluster   (A6)  (A7)  (A8)  (A9)  (A10)  Lagged dependent variable  0.69***  0.65***  0.59***  0.65***  0.35***    (0.07)  (0.08)  (0.11)  (0.09)  (0.04)  Left-wing government strength  −0.58  −0.20  −0.60  −0.54*  −0.85**    (0.40)  (0.27)  (1.68)  (0.30)  (0.40)  Core target group × Left-wing gov’t  0.31  0.48**  0.83  0.73***  0.33    (0.23)  (0.23)  (1.23)  (0.28)  (0.29)  Broken unemployment × Left-wing gov’t  0.63*  0.13  2.91  0.38  0.89***    (0.33)  (0.26)  (1.75)  (0.28)  (0.31)  Country-level controls             Real GDP  0.00  −0.00***  −0.00  −0.00  0.00*    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.04**  −0.02  −0.12  0.08**  0.05***    (0.02)  (0.02)  (0.11)  (0.03)  (0.01)   Unemployed and inactive (log)  −1.05  0.28  3.85  0.19  −0.06    (0.73)  (0.78)  (4.30)  (1.56)  (0.98)   Government deficit  −0.00  −0.03**  0.16  −0.01  0.04*    (0.02)  (0.02)  (0.11)  (0.02)  (0.02)   Government debt  0.01  −0.01  0.26  −0.00  0.02**    (0.00)  (0.00)  (0.17)  (0.01)  (0.01)   Number of programs in operation  −0.02  −0.00  −0.00  −0.01*  −0.00    (0.02)  (0.01)  (0.06)  (0.01)  (0.01)  Program-level controls             Core target group  −0.53**  −0.23  0.50  −0.20  −1.16***    (0.25)  (0.19)  (0.65)  (0.24)  (0.30)   Broken unemployment  −0.16  0.66*  −0.02  −0.84***  −0.65    (0.13)  (0.34)  (0.18)  (0.18)  (0.47)   Employment incentives × Left-wing gov’t  0.24  0.09  0.95  −0.22  −0.53    (0.35)  (0.25)  (0.97)  (0.27)  (0.41)   Sheltered empl. and rehab. × Left-wing gov’t  −0.14  −0.46**  0.29  −0.21  −0.50    (0.25)  (0.22)  (1.02)  (0.26)  (0.39)   Direct job creation × Left-wing gov’t  0.11  0.72**  0.04  −0.45  −0.09    (0.36)  (0.32)  (0.74)  (0.30)  (0.52)   Start-up incentives × Left-wing gov’t  0.05  0.18  1.02  0.07  −0.40    (0.25)  (0.60)  (1.24)  (0.34)  (0.45)  Constant  17.56*  0.06  −61.98  1.04  6.43    (10.08)  (11.97)  (62.39)  (23.82)  (14.17)  Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  929  2095  333  1143  1168  Programs  112  276  41  222  245    (A6)  (A7)  (A8)  (A9)  (A10)  Lagged dependent variable  0.69***  0.65***  0.59***  0.65***  0.35***    (0.07)  (0.08)  (0.11)  (0.09)  (0.04)  Left-wing government strength  −0.58  −0.20  −0.60  −0.54*  −0.85**    (0.40)  (0.27)  (1.68)  (0.30)  (0.40)  Core target group × Left-wing gov’t  0.31  0.48**  0.83  0.73***  0.33    (0.23)  (0.23)  (1.23)  (0.28)  (0.29)  Broken unemployment × Left-wing gov’t  0.63*  0.13  2.91  0.38  0.89***    (0.33)  (0.26)  (1.75)  (0.28)  (0.31)  Country-level controls             Real GDP  0.00  −0.00***  −0.00  −0.00  0.00*    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.04**  −0.02  −0.12  0.08**  0.05***    (0.02)  (0.02)  (0.11)  (0.03)  (0.01)   Unemployed and inactive (log)  −1.05  0.28  3.85  0.19  −0.06    (0.73)  (0.78)  (4.30)  (1.56)  (0.98)   Government deficit  −0.00  −0.03**  0.16  −0.01  0.04*    (0.02)  (0.02)  (0.11)  (0.02)  (0.02)   Government debt  0.01  −0.01  0.26  −0.00  0.02**    (0.00)  (0.00)  (0.17)  (0.01)  (0.01)   Number of programs in operation  −0.02  −0.00  −0.00  −0.01*  −0.00    (0.02)  (0.01)  (0.06)  (0.01)  (0.01)  Program-level controls             Core target group  −0.53**  −0.23  0.50  −0.20  −1.16***    (0.25)  (0.19)  (0.65)  (0.24)  (0.30)   Broken unemployment  −0.16  0.66*  −0.02  −0.84***  −0.65    (0.13)  (0.34)  (0.18)  (0.18)  (0.47)   Employment incentives × Left-wing gov’t  0.24  0.09  0.95  −0.22  −0.53    (0.35)  (0.25)  (0.97)  (0.27)  (0.41)   Sheltered empl. and rehab. × Left-wing gov’t  −0.14  −0.46**  0.29  −0.21  −0.50    (0.25)  (0.22)  (1.02)  (0.26)  (0.39)   Direct job creation × Left-wing gov’t  0.11  0.72**  0.04  −0.45  −0.09    (0.36)  (0.32)  (0.74)  (0.30)  (0.52)   Start-up incentives × Left-wing gov’t  0.05  0.18  1.02  0.07  −0.40    (0.25)  (0.60)  (1.24)  (0.34)  (0.45)  Constant  17.56*  0.06  −61.98  1.04  6.43    (10.08)  (11.97)  (62.39)  (23.82)  (14.17)  Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  929  2095  333  1143  1168  Programs  112  276  41  222  245  Cluster-robust standard errors in parentheses (clustered at the program level). *P < 0.10, **P < 0.05, ***P < 0.01. Model A6: Scandinavian (DK, FI, NO, SE). Model A7: Continental (AT, BE, DE, FR, NL, LU). Model A8: Anglo-Saxon (IE, UK). Model A9: Southern (CY, EL, ES, IT, MT, PT). Model A10: Central and Eastern European (BG, CZ, EE, HU, LT, LV, PL, RO, SI, SK). © The Author 2017. Published by Oxford University Press and the Society for the Advancement of Socio-Economics. All rights reserved. For Permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Socio-Economic Review Oxford University Press

Unemployment reduction or labor force expansion? How partisanship matters for the design of active labor market policy in Europe

Socio-Economic Review , Volume Advance Article – Apr 26, 2017

Loading next page...
 
/lp/ou_press/unemployment-reduction-or-labor-force-expansion-how-partisanship-4uLwJWj5Hl
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press and the Society for the Advancement of Socio-Economics. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
1475-1461
eISSN
1475-147X
D.O.I.
10.1093/ser/mwx014
Publisher site
See Article on Publisher Site

Abstract

Abstract Comparative scholars fundamentally disagree about the impact of partisan politics in modern welfare states, particularly in certain ‘new’ policy areas such as active labor market policy (ALMP). Using new data on 900 ALMP programs across Europe, this study attempts to reconcile a long-standing dispute between the traditional ‘power resources’ approach and the ‘insider/outsider’ approach pioneered by Rueda. The study argues that both left-wing and right-wing governments invest in ALMP but that politics still matter because parties’ preferences regarding unemployment differ. The left is more inclined to expand programs primarily designed to reduce unemployment, which exclusively target ‘core’ groups in, or at risk of, unemployment, and programs in which participants are no longer counted among the unemployed. In contrast, both sides are equally prone to expand programs that also—or instead—target people who are not yet participating in the labor market, which thus also—or instead—serve to increase labor supply. 1. Introduction It has been widely established that the last 15–20 years have seen an ‘activation turn’ of labor market policy across the OECD. Responding to the generally high unemployment and inactivity rates of the 1980s and early 1990s, governments have introduced a mix of demanding and enabling so-called ‘active labor market policy’ (ALMP) schemes aimed at the unemployed and some groups outside of the labor force (e.g. housewives and single parents) to facilitate their entry into employment. Although evaluations find mixed evidence regarding the effects of ALMPs on employment (Card et al., 2010; Kluve, 2010; Martin, 2014), ALMPs are consistently listed high on international policy agendas—most recently in the Employment Guidelines of the Europe 2020 strategy from 2010 and in the Social Investment Package launched by the European Commission in 2013. These developments have spurred a growing interest among comparative political scientists in the political and economic processes that determine the scale and orientation of ALMPs provided by governments across the EU and the OECD. Several books have been devoted to the policy area (e.g. Eichhorst et al., 2008; Weishaupt, 2011; Bonoli, 2013), and it is placed at the center of attention in most recent volumes that investigate the overall development of modern welfare states (e.g. Morel et al., 2012; Bonoli and Natali, 2012; Hemerijck, 2013; Thelen, 2014). Particular interest has been paid to the role of partisan politics for ALMP development; however, to date, scholars have not come anywhere close to a consensus regarding this matter. Over the past 10 years, studies have alternately reported evidence supporting hypotheses about positive, nonexistent and negative relationships between left-wing strength and the level of ALMP ‘effort’ exerted by the government. This lack of scholarly agreement was recently noted by Clasen et al. (2016, p. 33), who forcefully argued that most comparative research in the field has been fallen prey to conceptual under-specification of the dependent variable and that ‘[t]he process of uncovering the causal dynamics specific to this policy field is still in its infancy’. They suggest that scholars who search for variables to explain variations in ALMP development might want to look beyond the ‘usual suspects’, such as partisanship, and instead (or also) consider some rather different political institutional variables, for instance, those related to the nature of the national systems of public expenditure planning and control and the multi-actor delivery systems that characterize ALMPs. Whereas I largely subscribe to the diagnosis of Clasen et al. (2016) regarding the state of the field, I seek to demonstrate in this article that scholars run the risk of throwing the baby out with the bathwater if they fully adopt their proposed cure. I instead argue that if adequate attention is paid to the various purposes and target groups that ALMP programs potentially serve, our understanding of the impact of partisanship on ALMP design can be refined. Building on classic insights about variation in party preferences with regard to unemployment and labor shortages (Furåker, 1976; Hibbs, 1977), I outline a theory about how party preferences for ALMPs vary across programs that have different aims and thus different target groups and characteristics. On that basis, I derive two hypotheses, which are then tested on a new panel data set with unprecedented richness, covering almost 900 European ALMP programs over up to 15 years (European Commission, 2015). The results of these tests support a two-fold claim. First, whereas parties do not differ in their preferences for ALMP programs that, to some extent, serve to increase labor supply—and that thus also, or exclusively, target people who did not previously participate (at least not fully) in the labor market—the left is more likely than the right to also expand programs that primarily serve to reduce unemployment. These programs exclusively target people at the ‘core’ of the labor force: those who are unemployed and those who are employed but at the risk of losing their jobs. Second, the left is more inclined to expand particular ALMP programs for which participation entails a formal change in the participant’s labor market status—from unemployed to either inactive or employed. Together, these claims contradict a number of prevailing theories about the modern welfare state. First, they are at odds with the traditional ‘power resources’ school’s account of ALMPs (Esping-Andersen, 1990; Janoski, 1990; Rothstein, 1996; Huo et al., 2008) because, for many ALMP programs, right-wing governments are just as expansionary as left-wing governments. Second, they conflict with the ‘insider/outsider’ perspective that Rueda (2005, 2006, 2007) introduced as a critique of the ‘power resources’ approach, as right-wing governments match—but do not exceed—the commitment of left-wing governments for these types of programs. Third, on a more general note, the results suggest that it is too early to conclude that factors related to the ‘new politics’ of the welfare state approach—such as new, strong interest groups (Pierson, 1994), policy diffusion from international organizations (Armingeon, 2007) and the convergence of party preferences (Lindvall, 2010; Nelson, 2013; Tepe and Vanhuysse, 2013)—have rendered the traditional class-based political explanations of social and labor market policy obsolete. The following section highlights the inconclusiveness of previous research, outlines the theoretical argument and derives two hypotheses to be tested. In the third section, the new data set is introduced, and all variables are presented. The fourth section reports the results from a set of models, which lend support to both hypotheses; and the fifth section concludes. Descriptive statistics, robustness checks and extensions are provided in an Appendix, as well as in the online Supplementary Material. 2. Theory and hypotheses 2.1 Party politics and active labor market policy: a review As previously noted by multiple scholars, the cumulative evidence from the studies conducted over the past 10 years about the effect of partisan politics on ALMP ‘effort’ has been surprisingly inconclusive, especially given the similarity in the research questions and data applied (Bonoli, 2013; Tepe and Vanhuysse, 2013; Clasen et al., 2016). Based on the results these studies report, most of them can be divided into three groups.1 Those in the first group have found that left-wing influence is positively related to ALMP ‘effort’ (Huo et al., 2008; Iversen and Stephens, 2008; van Vliet and Koster, 2011). These results are consistent with the ‘power resources’ school’s account of ALMP development, in which social democratic governments are considered more inclined to expand ALMP to strengthen labor as an organized social force by contributing to lower levels of unemployment. This understanding of ALMP was long the dominant one among welfare state scholars (Esping-Andersen, 1990; Janoski, 1990, 1994; Rothstein, 1996; Boix, 1998). A second group of studies have reported that the government’s ideological underpinnings have no impact on ALMP ‘effort’ (Rueda, 2005; Franzese and Hays, 2006; Armingeon, 2007; Gaston and Rajaguru, 2008; Bonoli, 2013). Multiple explanations have been proposed for why ideology is not expected to have an impact on such efforts. Mucciaroni (1990), Swenson (2002), and Farnsworth (2012, 2013) have argued that ALMPs—particularly training programs and labor market services—are often in the interest of workers and employers alike. As such, they are less likely to be a subject of partisan dispute. Another explanation is that policy diffusion—spurred by mutual learning experiences, a broad consensus among policy experts, and the influence of employment strategies adopted by the EU and the OECD—has caused party preferences for labor market policy to converge (Franzese and Hays, 2006; Armingeon, 2007; Lindvall, 2010; Nelson, 2013). This ‘deep shift in thinking’ (Nelson, 2013, p. 272) since around 1998 is sometimes referred to as the ‘activation turn’ (Bonoli, 2010, p. 435). Finally, a few studies have found a negative relationship between left-wing influence and spending on (at least some categories of) ALMP. Drawing on insider–outsider theories of unemployment in economics, Rueda (2005, 2006, 2007) has made the influential claim that because unions and—for electoral reasons—social democratic parties tend to favor the interests of labor market insiders, they are unlikely to support, and might even oppose, ALMPs. According to this account, ALMPs do not favor insiders because they promote the employment entry of outsiders who can underbid insiders’ wage demands and simultaneously increase the tax burden (Rueda, 2005). In accordance with this theory, Rueda (2007) found that increased left-wing strength resulted in a decreased ALMP spending. Building on Rueda’s work, Tepe and Vanhuysse (2013) recently introduced the ‘left party disinvestment thesis’. Supporting a weak version of that thesis, they found that increased left-wing influence does not typically increase the overall ALMP expenditure; in addition, in line with a strong version of that thesis, they found that increased left-wing influence decreases spending on the one category of ALMPs—direct job creation programs—which they suggest will most likely benefit outsiders. Reporting separate analyses for different categories2 of ALMP programs—such as training programs, employment incentives and direct job creation programs—Tepe and Vanhuysse’s (2013) study, together with those by Bonoli (2010), Nelson (2013) and Vlandas (2013), mark a new wave of ALMP research that does not fit nicely into any of the three previous bodies of research. These authors begin to address the problem that ALMP programs differ in terms of their effects on individuals and on the labor market, and that therefore, they also presumably differ in terms of how different political actors value them. Thus, their studies clearly represent a sophisticated advance within the field. However, their joint results regarding the effect of left-wing strength on ALMP ‘effort’ are as inconclusive as those in the previous literature. For training programs, Nelson (2013) finds a positive effect, whereas Tepe and Vanhuysse (2013) and Vlandas (2013) find none. For direct job creation programs, Tepe and Vanhuysse (2013) find a negative effect; Vlandas (2013) finds none; and Nelson (2013) finds a positive effect prior to the ‘activation turn’. For employment incentives, Nelson (2013) finds a positive effect; Tepe and Vanhuysse (2013) find none; and Vlandas (2013) finds a negative effect. These inconclusive results partly stem from the difficulties involved in deriving hypotheses about party preferences for particular categories of ALMP programs because, as aptly demonstrated by Clasen et al. (2016, p. 30), programs that fall in the same administrative category may have ‘very distinctive aims and presumably very different political support coalitions’. 2.2 The role of ALMP: to reduce unemployment or to expand labor supply? In this article, I argue that the differences in ALMP programs’ objectives are key for understanding partisan differences in ALMP preferences. Possibly the first scholar to provide this insight was Furåker (1976), who classified the traditional categories of labor market policy interventions based on whether they were meant to serve either one or both of two possible purposes: reducing (or preventing) unemployment and/or reducing (or preventing) labor shortages.3 In Furåker’s model, as in those of some other contemporary scholars (e.g. Hibbs, 1977), sellers (i.e. workers) and buyers (i.e. employers) on the labor market vary in their preferences regarding unemployment. Accordingly, he suggested that the ways in which governments prioritize labor market measures depend on the extent to which each group has been able to influence government policy. Thus, left-wing governments, which tend to favor workers’ interests, are expected to prioritize measures that aim to reduce unemployment, whereas right-wing governments, which have closer ties to the business community, are expected to be more concerned about labor shortages. Research from the past two decades has provided an additional reason for why left-wing governments might tend to be more concerned about unemployment: electoral motivations stemming from issue ownership. At the ballot box, left-wing governments are often found to be penalized particularly harshly for unemployment (Powell and Whitten, 1993; Whitten and Palmer, 1999; van der Brug et al., 2007). Now, although I find Furåker’s framework largely compelling, I contend that the most important determinant of whether a labor market program is meant to reduce unemployment or labor shortages is not the program’s content but whether it targets people who are already participating in the labor market or those who are not. Indeed, contrary to Rueda’s (2006, p. 388) influential claim that ‘ALMPs unambiguously benefit outsiders’, I argue that whereas some categories of labor market programs—such as sheltered employment for the disabled—target people who are fairly homogeneous with respect to their ‘outsiderness’, most categories accommodate programs for which the primary target group includes those at the ‘core’ of the labor force and programs that target people who are on the fringes or even outside the labor market.4 For instance, Clasen et al. (2016, p. 30) show that the direct job creation category accommodates both types of programs. In addition, labor market training programs are known to accommodate many programs that target particularly disadvantaged groups;5 however, the education schemes and the accompanying work time reduction subsidies included in the so-called short-time work (STW) programs, which many European governments rolled out during the recent financial crisis, primarily targeted workers who were already employed but who ran the risk of becoming unemployed6 (Hijzen and Venn, 2011). These examples illustrate why the attention that Furåker pays to program categories might be misguided, and they might also partly explain why the results from the most recent wave of ALMP research are so inconclusive. To summarize, I propose shifting the focus away from program categories and argue that a program’s overarching aim—unemployment reduction or labor force expansion—is more likely to be the primary source of partisan conflict. Two hypotheses can be derived to test this claim. First, I expect that left-wing governments are more prone than right-wing governments to expand labor market programs that exclusively target people at the ‘core’ of the labor force who are unemployed or at risk of becoming unemployed. In contrast, in line with Boix’s (1998, p. 4) remark that ‘[i]n the first place, all parties prefer to develop policies that maximize growth’, I expect less of a partisan effect for programs that also, or exclusively, target ‘non-core’ groups. These programs do not necessarily primarily aim to reduce unemployment, but they might serve to increase the size of the labor force by making people who would otherwise be inactive—such as housewives, discouraged youth and people in early retirement—begin searching for a job.7 If such programs succeed, growth might increase both directly—through increased output—and indirectly—if the new employment mitigates bottlenecks caused by labor shortages. Moreover, these programs might have side effects, such as increased tax revenue and lower caseloads in other more expensive social security programs, which are attractive to all governments, irrespective of their ideologies or allegiances. Now, it is not obvious that increasing output by increasing labor force participation is in line with the traditional left-wing agenda, a defining feature of which is often held to be the decommodification of labor; i.e. the aim to provide citizens with an opportunity to opt out of work without a potential loss of income or welfare (Esping-Andersen, 1990). However, at odds with this understanding of the left-wing agenda, more recent comparative studies have found that social democratic parties emphasize policies that promote labor market participation over a long-term labor market exit (Huber and Stephens, 2001, Huo et al., 2008). These studies find that left-wing control of the cabinet is associated with those kinds of decommodification policies that do not reduce aggregate levels of employment—such as old-age pension entitlements and short-term unemployment benefit replacement rates—but unrelated to those kinds that provide strong work disincentives—such as the duration and the replacement rates of long-term unemployment benefits. Importantly, this revised understanding of the left-wing’s attitude toward decommodification is compatible with ALMPs that are aimed to increase labor force participation.8 The second hypothesis is that governments with different ideological makeups prefer different ALMP programs depending on what participation entails for an individual’s labor market status. For some types of programs (e.g. full-time training programs), enrollment typically implies that the participant is no longer immediately available for work and, in turn, that his or her unemployment spell is either broken (whereby the unemployment duration counter is reset to zero) or suspended (whereby the duration counter is paused until the participant leaves the program). In other programs, participation does not change one’s labor market status. Two partisan mechanisms might be at play here. First, according to basic search models of the labor market, as more people actively engage in job seeking, labor market competition increases—likely to the detriment of core workers, whose interests are favored by left-wing governments. Second, people who enroll in a program that breaks their unemployment spell are no longer counted among the unemployed. Thus, maintaining a stock of participants in such programs might artificially reduce the unemployment rate in the long term (while keeping the participants active in supposedly productive training or subsidized work). As already noted, a lower unemployment rate might strengthen the bargaining power of labor and be electorally beneficial to left-wing parties. Therefore, I hypothesize that left-wing governments are more inclined to expand ALMP programs that cause a break or suspension of the unemployment spell, whereas right-wing governments are more interested in supporting ALMP programs that keep participants in the labor supply. Table 1 summarizes the hypotheses presented above. Table 1. Hypotheses: program feature × partisanship interactions   H1: Core target group   H2: Broken unemployment     Core  Non-core  Broken  Continuous  Impact of Left-wing government strength on ALMP ‘effort’  +  0  +  –    H1: Core target group   H2: Broken unemployment     Core  Non-core  Broken  Continuous  Impact of Left-wing government strength on ALMP ‘effort’  +  0  +  –  Table 1. Hypotheses: program feature × partisanship interactions   H1: Core target group   H2: Broken unemployment     Core  Non-core  Broken  Continuous  Impact of Left-wing government strength on ALMP ‘effort’  +  0  +  –    H1: Core target group   H2: Broken unemployment     Core  Non-core  Broken  Continuous  Impact of Left-wing government strength on ALMP ‘effort’  +  0  +  –  3. Data and operationalization 3.1 The EU Labor Market Policy database The comparative research on ALMP has invariably used the OECD Labor Market Policy (LMP) database, which contains country-year observations on expenditures on a number of program categories, starting in 1985 for a subset of countries (Grubb and Puymoyen, 2008). However, this database lacks data on program characteristics, which are required to test the hypotheses introduced above. Fortunately, the European Commission (2015) collects data on labor market programs in the EU Member States and Norway that meet all the requirements. Importantly, in this database, the unit of observation is program-year rather than country-year. Therefore, information about a large set of qualitative program characteristics as well as annual summaries of expenditures and participants are reported annually for each program operating in each country, which makes individually analyzing each one of these hundreds of programs possible rather than simply obtaining country-level aggregates. Because, to the best of my knowledge, these data have not previously been applied in this field of research, a few limitations of the EU LMP database should be noted here. First, no data are available from before 1998. However, for the present purposes, this limitation only makes for a tougher test of my hypotheses because, as noted above, previous studies have found the effects of partisan politics to be smaller since the ‘activation turn’ around the turn of the century. Second, all data are reported via a questionnaire that is completed by national authorities, and approximately 10% of the reported quantitative data are based on estimations. Therefore, systematic cross-country differences in reporting and estimation practices might distort cross-country data comparisons. However, in the present study, this problem is mitigated by the fact that all regression models include program-fixed effects, which ensure that no between-country variation is used to estimate the parameters. Third, whereas the data are based in principle, on a full count of labor market programs as defined by Eurostat (2013), the database is only supposed to cover interventions at the national and regional levels. As argued by Clasen et al. (2016), the omission of local ALMPs might distort comparative analyses, yet I argue that the present study is spared from such problems because the unit of analysis is an individual program, not a country-level aggregate. Finally, whereas most program categories in the database contain only labor market interventions that ‘aim to benefit identifiable individuals’ (Eurostat, 2013, p. 7) and that are thus suitable for inclusion in this study, the labor market services category also covers functions that are not directly linked to individual participants, such as services for employers, administrative functions and general overhead. Therefore, this program category is excluded from the study, along with the two categories that are typically not considered ‘active’ labor market policies: out-of-work income maintenance and support and early retirement. Despite these issues, the data in the EU LMP database seem to be of high quality and satisfactorily comparable across interventions and years, particularly if only within-country or within-program variation is used in estimations. 3.2 Dependent variable: ALMP ‘effort’ Most comparative researchers base their indicators of governments’ ALMP ‘efforts’ on how much public spending (relative to GDP) is devoted to the policy area. In an effort to disentangle the effects of deliberate policy decisions from the effects of economic conditions, many scholars control for the ‘problem pressure’ by adjusting the rate of unemployment on either side of the regression equation. However, as discussed by Clasen et al. (2016), unemployment rates are ‘notoriously problematic in comparative analysis as they are expressed as a ratio of the labour force’. Because unemployment, inactivity and employment are communicating vessels, expansion in policy areas such as early retirement, higher education or part-time work that lead to a decline in the unemployment rate might generate ‘a largely artificial image of increasing ALMP “effort”’ (Clasen et al., 2016, p. 27). Moreover, as noted above, many ALMP programs mechanically alter the labor market status of their participants from unemployed to inactive or employed, which means that indicators of ‘effort’ that are adjusted according to unemployment may ‘be endogenously ratcheted upwards or downwards by increases or decreases in expenditure on measures that have a direct impact on the unemployment rate’ (Clasen et al., 2016, p. 28). Another problem with routinely adjusting spending for the unemployment rate is that being unemployed does not necessarily imply that one takes part in the programs provided by the government. An analysis of an indicator reported by Eurostat (2015a), which measures the share of the registered unemployed who participate in an ALMP program, reveals that the ‘activation rate’ varies considerably—between countries and within countries. For the 214 observed country-years, nested in 24 countries, the overall mean activation rate is 21.6% of the registered unemployed. The between-country standard deviation is 11.4 percentage points, and the within-country standard deviation is 6.1 percentage points. This variation indicates that simply adjusting ALMP spending for unemployment does not get us very far if we want to reliably assess how the treatment that an unemployed individual can expect to receive from the government varies between countries or over time. The literature devoted to traditional social insurance systems (e.g. Esping-Andersen, 1990; Korpi and Palme, 1998; Scruggs and Allan, 2006) recognizes these policies’ coverage rates and eligibility criteria as important dimensions. Similarly, I argue that who and how many individuals participate in ALMP programs warrants further attention. Whereas the problems discussed above should not necessarily lead us to dismiss ALMP spending altogether when conceptualizing ALMP ‘effort’, I argue that Esping-Andersen (1990, p. 20) was right to remark that if our aim as welfare state scholars ‘is to test causal theories that involve actors, we should begin with the demands that were actually promoted by those actors’ and that ‘it is difficult to imagine that anyone struggled for expenditure per se’. Because the total expenditure for any ALMP program (or a complete program portfolio) is constituted by two components—the number of participants and the average expenditure per participant—I argue that, for a given ALMP program, the number of participants is a more valid indicator of the government’s inclination to use the program than is total expenditure. Changes in the total expenditure per participant might also reflect, for instance, changes in the efficiency of operations, economies of scale and other factors that might not reflect government’s preferences as clearly as the number of people enrolled in the program. Therefore, I use the data on the average annual participant stock—henceforth denoted the Scope of the ALMP program—to construct the main dependent variable, whereas corresponding models for the program’s total annual Expenditure, measured in million Euros at constant 2005 prices, are reported as robustness checks in Table A5. Admittedly, the Scope indicator is not free of problems. First, it is impossible to tell if a given participant stock in an ALMP program accommodates a large number of short-term transient participants—those soon to enter employment or to transfer to another program—or a small group of participants who have been enrolled for a long time because they are part of a lengthy training scheme or because they are simply ‘trapped’ in the program. However, although this implies that the average participant stock is not a suitable indicator for assessing program content or efficiency, I maintain that Scope is a better proxy than Expenditure for the extent to which a government seeks to intervene in the labor market in a discretionary manner to achieve some particular objective; which, fortunately, is of interest in the present study. Second, whereas the expenditure data in the EU LMP database are considered relatively complete, more gaps can be found in the participant data (European Commission, 2015). Whereas one or more observations of expenditure data exist for 1270 programs, participant data only exist for 1113 programs. This is another reason to use Expenditure as a robustness check. Because the relationships between the programs’ Scope (as well as Expenditure) and the independent variables of interest are not expected to be linear (for good reason), the dependent variable is log-transformed.9 3.3 Independent variables The analyses include two program-level independent variables of particular interest: Core target group and Broken unemployment. Both are program-year dummy variables that are extracted from the qualitative data reports of the EU LMP database. The data required to produce them are nearly complete. Data are missing for only 0.6% of all observations, and for about half of those with missing data, all needed information can be inferred from the program descriptions included in the dataset. Core target group is assigned a 0 if the program is available for one or both of the two target groups ‘not registered’ and ‘other registered jobseekers’, and a 1 if it exclusively targets those who are ‘registered unemployed’ or ‘employed at risk of involuntary job loss’. According to Eurostat (2013, p. 45), ‘not registered’ ‘indicates where interventions are targeted at groups who are not in employment or where registration with the PES [Public Employment Service] is not a prerequisite for participation’.10 In practice, ‘other registered jobseekers’ ‘refers to persons who are unemployed (but do not qualify as registered unemployed), underemployed or inactive’. While the definitions of these two target groups cover a rather heterogeneous set of individuals, none of these individuals is unemployed according to the national definition—nor employed and at risk of unemployment. Therefore, programs that target—partly or exclusively—one of these two groups serve, at least to some extent, to increase labor force participation. Conversely, programs that only target ‘registered unemployed’ or ‘employed at risk’ primarily seek to reduce (or prevent) unemployment. To test hypothesis H1, I include an interaction term between Core target group and Left-wing government strength. The latter variable is defined, according to a well-established practice, as the number of cabinet posts held by social democrats and members of other leftist parties as a share of the total number of cabinet posts, weighted by the number of days in office in a given year. Data are retrieved from Armingeon et al. (2015). The second dummy variable, Broken unemployment, is assigned a 1 if participation in the program breaks or suspends the unemployment spell and a 0 if unemployment is continuous. Programs for which this variable is coded as ‘not relevant’—for instance, because they do not target the unemployed—are assigned a 0. To test hypothesis H2, I include the interaction term Broken unemployment × Left-wing government strength. To control for the potential confounding interactions between Left-wing government strength and the program category, I include interaction terms between Left-wing government strength and each of the main ALMP program categories: Training, Employment incentives, Sheltered and supported employment and rehabilitation, Direct job creation and Start-up incentives (with the interaction term that includes Training omitted and used as a reference category). I include program-fixed effects in all models to control for unobserved between-program heterogeneity, including any differences stemming from potential systematic cross-country differences in the way that participant and expenditure data are reported to the EU LMP database. I thereby also control for a number of largely time-invariant, country-level factors that have been found to affect ALMP ‘effort’ in previous studies, including welfare regime differences, trade openness, the degree of employer coordination and the involvement of social partners in policymaking.11 To control for within-country variation in ‘problem pressure’, I include an item for the logged number of Unemployed and inactive individuals from ages 15–64 years. I also include controls for Real GDP at constant 2005 prices, Real GDP growth, Government deficit and Government debt. The indicator of GDP is retrieved from Eurostat (2015b); data on unemployment and inactivity come from the EU Labor Force Survey (Eurostat, 2015c); and the other variables are provided by Armingeon et al. (2015). I also control for the Number of programs that are operating in the country during a particular year because—all other things being equal—the more ALMP programs in place at the same time, the smaller each individual program is likely to be. Finally, I add year dummies to control for possible common temporal shocks. For my final sample, I exclude 14 programs (145 observations) that have at least one reported break in the time series and 66 programs (477 observations) that have at least one gap in the data. However, as demonstrated by the robustness check in the Supplementary Material (Model S2), including these observations in the sample does not markedly change the results for the coefficients of interest. To assure full comparability between models with and without a lagged dependent variable, I also leave out the first observation of each panel in the models without a lagged dependent variable to attain an identical sample for both types of models. In doing so, I exclude approximately 1000 observations from the analysis, including 132 full programs that have been observed for only 1 year. These exclusions slightly change the composition of my sample, but should not pose any real problem to the analysis.12 The final sample comprises approximately 5600 program-years nested in nearly 900 programs, which are nested in 28 countries.13 Descriptive statistics for all included variables are reported in Table A2. Table A3 shows how the data are structured by presenting a detailed description of two program-year observations from two different programs in different countries. In the Supplementary Material, Table S1 documents how the panels and observations are distributed across countries and years, Figure S1 presents the frequencies of panels of varying length, across program categories and Table S2 describes the observation patterns for all panels included in the study. Table 2. Regression results   (1)  (2)  (3)  (4)  (5)  Lagged dependent variable        0.59***  0.69***          (0.04)  (0.01)  Left-wing government strength  0.50***  −0.47**  −0.41**  −0.35***  −0.33***    (0.16)  (0.18)  (0.19)  (0.12)  (0.11)  Core target group × Left-wing gov’t    0.57***  0.54***  0.39***  0.39***      (0.14)  (0.15)  (0.11)  (0.10)  Broken unemployment × Left-wing gov’t    0.77***  0.83***  0.53***  0.47***      (0.18)  (0.20)  (0.16)  (0.09)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.06***  0.06***  0.05***  0.03**  0.03***    (0.02)  (0.02)  (0.02)  (0.01)  (0.01)   Unemployed and inactive (log)  1.22**  1.35**  1.32**  1.25***  1.24***    (0.55)  (0.55)  (0.53)  (0.41)  (0.27)   Government deficit  −0.01  −0.01  −0.01  −0.00  −0.00    (0.01)  (0.01)  (0.01)  (0.00)  (0.01)   Government debt  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.00  −0.00  −0.00  −0.00  −0.00**    (0.01)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.21  −0.43**  −0.39**  −0.27**  −0.27***    (0.15)  (0.16)  (0.15)  (0.12)  (0.10)   Broken unemployment  −0.07  −0.20  −0.22  −0.08  −0.07    (0.19)  (0.17)  (0.17)  (0.12)  (0.11)   Employment incentives × Left-wing gov’t      −0.16  −0.06  −0.04        (0.21)  (0.13)  (0.10)   Sheltered empl. and rehab. × Left-wing gov’t      −0.49  −0.29  −0.26*        (0.29)  (0.20)  (0.15)   Direct job creation × Left-wing gov’t      0.50  0.34  0.31**        (0.54)  (0.31)  (0.15)   Start-up incentives × Left-wing gov’t      −0.16  −0.07  −0.03        (0.34)  (0.34)  (0.18)  Constant  −9.76  −11.39  −11.09  −14.38**      (8.29)  (8.26)  (8.02)  (6.15)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  5668  5668  5668  5668  5668  Programs  896  896  896  896  896    (1)  (2)  (3)  (4)  (5)  Lagged dependent variable        0.59***  0.69***          (0.04)  (0.01)  Left-wing government strength  0.50***  −0.47**  −0.41**  −0.35***  −0.33***    (0.16)  (0.18)  (0.19)  (0.12)  (0.11)  Core target group × Left-wing gov’t    0.57***  0.54***  0.39***  0.39***      (0.14)  (0.15)  (0.11)  (0.10)  Broken unemployment × Left-wing gov’t    0.77***  0.83***  0.53***  0.47***      (0.18)  (0.20)  (0.16)  (0.09)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.06***  0.06***  0.05***  0.03**  0.03***    (0.02)  (0.02)  (0.02)  (0.01)  (0.01)   Unemployed and inactive (log)  1.22**  1.35**  1.32**  1.25***  1.24***    (0.55)  (0.55)  (0.53)  (0.41)  (0.27)   Government deficit  −0.01  −0.01  −0.01  −0.00  −0.00    (0.01)  (0.01)  (0.01)  (0.00)  (0.01)   Government debt  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.00  −0.00  −0.00  −0.00  −0.00**    (0.01)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.21  −0.43**  −0.39**  −0.27**  −0.27***    (0.15)  (0.16)  (0.15)  (0.12)  (0.10)   Broken unemployment  −0.07  −0.20  −0.22  −0.08  −0.07    (0.19)  (0.17)  (0.17)  (0.12)  (0.11)   Employment incentives × Left-wing gov’t      −0.16  −0.06  −0.04        (0.21)  (0.13)  (0.10)   Sheltered empl. and rehab. × Left-wing gov’t      −0.49  −0.29  −0.26*        (0.29)  (0.20)  (0.15)   Direct job creation × Left-wing gov’t      0.50  0.34  0.31**        (0.54)  (0.31)  (0.15)   Start-up incentives × Left-wing gov’t      −0.16  −0.07  −0.03        (0.34)  (0.34)  (0.18)  Constant  −9.76  −11.39  −11.09  −14.38**      (8.29)  (8.26)  (8.02)  (6.15)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  5668  5668  5668  5668  5668  Programs  896  896  896  896  896  Model 1–4: Cluster-robust standard errors in parentheses (clustered at the country level). Model 5: Bootstrapped standard errors in parentheses. Bias correction initialized by Arellano and Bond estimator (Bruno 2005). *P < 0.10, **P < 0.05, ***P < 0.01. Table 2. Regression results   (1)  (2)  (3)  (4)  (5)  Lagged dependent variable        0.59***  0.69***          (0.04)  (0.01)  Left-wing government strength  0.50***  −0.47**  −0.41**  −0.35***  −0.33***    (0.16)  (0.18)  (0.19)  (0.12)  (0.11)  Core target group × Left-wing gov’t    0.57***  0.54***  0.39***  0.39***      (0.14)  (0.15)  (0.11)  (0.10)  Broken unemployment × Left-wing gov’t    0.77***  0.83***  0.53***  0.47***      (0.18)  (0.20)  (0.16)  (0.09)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.06***  0.06***  0.05***  0.03**  0.03***    (0.02)  (0.02)  (0.02)  (0.01)  (0.01)   Unemployed and inactive (log)  1.22**  1.35**  1.32**  1.25***  1.24***    (0.55)  (0.55)  (0.53)  (0.41)  (0.27)   Government deficit  −0.01  −0.01  −0.01  −0.00  −0.00    (0.01)  (0.01)  (0.01)  (0.00)  (0.01)   Government debt  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.00  −0.00  −0.00  −0.00  −0.00**    (0.01)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.21  −0.43**  −0.39**  −0.27**  −0.27***    (0.15)  (0.16)  (0.15)  (0.12)  (0.10)   Broken unemployment  −0.07  −0.20  −0.22  −0.08  −0.07    (0.19)  (0.17)  (0.17)  (0.12)  (0.11)   Employment incentives × Left-wing gov’t      −0.16  −0.06  −0.04        (0.21)  (0.13)  (0.10)   Sheltered empl. and rehab. × Left-wing gov’t      −0.49  −0.29  −0.26*        (0.29)  (0.20)  (0.15)   Direct job creation × Left-wing gov’t      0.50  0.34  0.31**        (0.54)  (0.31)  (0.15)   Start-up incentives × Left-wing gov’t      −0.16  −0.07  −0.03        (0.34)  (0.34)  (0.18)  Constant  −9.76  −11.39  −11.09  −14.38**      (8.29)  (8.26)  (8.02)  (6.15)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  5668  5668  5668  5668  5668  Programs  896  896  896  896  896    (1)  (2)  (3)  (4)  (5)  Lagged dependent variable        0.59***  0.69***          (0.04)  (0.01)  Left-wing government strength  0.50***  −0.47**  −0.41**  −0.35***  −0.33***    (0.16)  (0.18)  (0.19)  (0.12)  (0.11)  Core target group × Left-wing gov’t    0.57***  0.54***  0.39***  0.39***      (0.14)  (0.15)  (0.11)  (0.10)  Broken unemployment × Left-wing gov’t    0.77***  0.83***  0.53***  0.47***      (0.18)  (0.20)  (0.16)  (0.09)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.06***  0.06***  0.05***  0.03**  0.03***    (0.02)  (0.02)  (0.02)  (0.01)  (0.01)   Unemployed and inactive (log)  1.22**  1.35**  1.32**  1.25***  1.24***    (0.55)  (0.55)  (0.53)  (0.41)  (0.27)   Government deficit  −0.01  −0.01  −0.01  −0.00  −0.00    (0.01)  (0.01)  (0.01)  (0.00)  (0.01)   Government debt  −0.00  −0.00  −0.00  −0.00**  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.00  −0.00  −0.00  −0.00  −0.00**    (0.01)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.21  −0.43**  −0.39**  −0.27**  −0.27***    (0.15)  (0.16)  (0.15)  (0.12)  (0.10)   Broken unemployment  −0.07  −0.20  −0.22  −0.08  −0.07    (0.19)  (0.17)  (0.17)  (0.12)  (0.11)   Employment incentives × Left-wing gov’t      −0.16  −0.06  −0.04        (0.21)  (0.13)  (0.10)   Sheltered empl. and rehab. × Left-wing gov’t      −0.49  −0.29  −0.26*        (0.29)  (0.20)  (0.15)   Direct job creation × Left-wing gov’t      0.50  0.34  0.31**        (0.54)  (0.31)  (0.15)   Start-up incentives × Left-wing gov’t      −0.16  −0.07  −0.03        (0.34)  (0.34)  (0.18)  Constant  −9.76  −11.39  −11.09  −14.38**      (8.29)  (8.26)  (8.02)  (6.15)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  5668  5668  5668  5668  5668  Programs  896  896  896  896  896  Model 1–4: Cluster-robust standard errors in parentheses (clustered at the country level). Model 5: Bootstrapped standard errors in parentheses. Bias correction initialized by Arellano and Bond estimator (Bruno 2005). *P < 0.10, **P < 0.05, ***P < 0.01. 4. Descriptive evidence Before turning to the modeling exercises, consider some initial descriptive evidence for the two hypotheses, presented in Figure 1. For each of the 28 studied countries, each graph’s x-axis plots Left-wing government strength averaged over all the years for which data for the country is available in the EU LMP database (ranging from 6 to 16 years between 1998 and 2013). For each corresponding country and time period, the panel on the left plots the average total ALMP expenditure as a share of GDP. No apparent association exists between the prevalence of left-wing cabinets and total ALMP expenditure during the observed period. In contrast, the center panel and the panel on the right—each of which splits the total expenditure into two parts—show that a clear correlation exists between Left-wing government strength and the share of the total ALMP expenditure allocated to programs that fall in the Core target group and Broken unemployment categories. In line with both hypotheses, the stronger the Left-wing government strength, the larger share of overall spending that is devoted toward programs in the Core target group (r = 0.30, P = 0.12) and Broken unemployment (r = 0.51, P < 0.01) categories. Figure 1. View largeDownload slide Left-wing government strength and ALMP in 28 European countries. Note: For definitions, see Section 3. Indicators are averaged over all years for which data for the country are available in the EU LMP database: 1998—2013: AT, BE, DK, FI, IE, IT, NL, NO, PT, SE; 1998–2012: DK, ES, FR, LU; 1998–2011: UK; 1998–2010: EL; 2002–2013: CZ; 2003–2013: EE, LT, LV; 2003–2012: HU, RO; 2004–2013: BG, SK; 2005–2013: SI; 2005–2012: PL; 2006–2012: MT; 2006–2011: CY. Sources: European Commission (2015), Armingeon et al. (2015) and author’s calculations. Figure 1. View largeDownload slide Left-wing government strength and ALMP in 28 European countries. Note: For definitions, see Section 3. Indicators are averaged over all years for which data for the country are available in the EU LMP database: 1998—2013: AT, BE, DK, FI, IE, IT, NL, NO, PT, SE; 1998–2012: DK, ES, FR, LU; 1998–2011: UK; 1998–2010: EL; 2002–2013: CZ; 2003–2013: EE, LT, LV; 2003–2012: HU, RO; 2004–2013: BG, SK; 2005–2013: SI; 2005–2012: PL; 2006–2012: MT; 2006–2011: CY. Sources: European Commission (2015), Armingeon et al. (2015) and author’s calculations. Clearly, these associations are no more than simple correlations, and, as discussed above, numerous factors potentially influence both the cabinet’s composition and the composition of the ALMP portfolio, thereby confounding the relationships displayed in Figure 1. Studying these relationships in a more rigorous manner requires a more advanced econometric analysis, which is the topic of the next section. 5. Estimation strategy and results This section reports the results from a set of models that can all be represented—with various restrictions—with the same general equation:   Ypct=γYpct−1+αp+δt +βLct+λ′Ppct+θ′(Lct× Ppct)+ ρ′Cct+ɛpct (1) Here, Ypct is the Scope of program p in country c in year t; Ypct-1 is its first lag; αp is a program-specific intercept; and δt is a year-fixed effect. Lct is Left-wing government strength in country c in year t; Ppct is a vector of program-level variables; λ is a vector of the main effects of these variables; and θ is a vector of the effects of the interaction between Lct and each of the variables in Ppct. Cct is a vector of time-varying, country-level controls; ρ is a vector of the effects of those controls; and ɛpct is the error term. Because the program-fixed effects included in these models do not necessarily completely control for within-cluster correlation (Cameron and Miller, 2015), I use cluster-robust standard errors whenever possible. As the more conservative option for nested data involves clustering on a higher level, I cluster on the 28 countries included rather than on the program level. A Breusch-Pagan LM test confirms that considerable between-program heterogeneity exists in the data and that a random effects or fixed effects model is thus preferable to pooled OLS regression. Hausman tests performed on various model specifications consistently reject one of the key assumptions underpinning the random effects model, i.e., that all independent variables are uncorrelated with the residuals. Therefore, I opt for the consistent—although less efficient—fixed effects model, which only makes use of within-program variation. Model 1 in Table 2 reports that, on average, Left-wing government strength has a positive and statistically significant effect on the Scope of ALMP programs when the country-level and program-level factors discussed above are controlled for and all potential interactions are restricted to zero. However, the positive coefficients for the two interaction terms included in Model 2 seemingly indicate that the effect of left-wing power varies between programs in line with the hypotheses. To rule out any suspicion that these results are driven by confounding interactions between Left-wing government strength and program category, Model 3 controls for that risk by adding four additional interaction terms.14 Including them does not drastically alter the results.15 The coefficients estimated in a panel-data fixed effects model without a lagged dependent variable, such as Model 3—for which the identifying assumption is that all potential confounders are time-invariant—tend to be too large if time-varying confounders are also present (Angrist and Pischke, 2009, ch. 5.4). Because inflated coefficients cannot be ruled out in the present case, Model 4 includes a lagged dependent variable. Consequently, all the coefficients of interest are reduced substantially. Now, as first noted by Nickell (1981), OLS estimates from models with both fixed effects and a lagged dependent variable tend to be biased due to the correlation between the lagged dependent variable and the error term. Therefore, the main model, Model 5, applies a version of the bias-corrected LSDV estimator (LSDVc) (Kiviet, 1995), which has been found to be reasonably accurate when evaluated on unbalanced, short panels, such as those in the data used here (Flannery and Hankins, 2013). The coefficients of interest in Model 5 are slightly smaller than those in Model 4, and, because the LSDVc estimator developed by Bruno (2005) only reports bootstrapped standard errors, these standard errors are markedly smaller for some coefficients than the cluster-robust standard errors used in the other models. However, even if standard errors of a similar size to those in Model 4 were applied to the corresponding coefficients in Model 5, the statistical significance of the coefficients of interest would remain. Although the positive and strongly significant interaction effects reported from all models in Table 2 lend preliminary support to both hypotheses, a closer examination of how Left-wing government strength’s estimated effect on program Scope varies between programs with different characteristics is needed before any conclusions can be drawn. The results of such an examination, based on the estimates from Model 5, are presented graphically in Figure 2. Figure 2. View largeDownload slide Partial conditional effects of left-wing government on ALMP ‘effort’. Note: Partial effects of a change from 0% to 100% Left-wing government strength on Scope conditional on a program characteristic, with all other variables held at their means. Error bars denote a 95% confidence interval. All effects are estimated from Model 5. Raw effect estimates, standard deviations and P-values are reported in the Table A4. Figure 2. View largeDownload slide Partial conditional effects of left-wing government on ALMP ‘effort’. Note: Partial effects of a change from 0% to 100% Left-wing government strength on Scope conditional on a program characteristic, with all other variables held at their means. Error bars denote a 95% confidence interval. All effects are estimated from Model 5. Raw effect estimates, standard deviations and P-values are reported in the Table A4. Whereas the coefficients for all the independent variables reported in Table 2 represent effects on the exponent of the natural logarithm of Scope, Figure 2 plots exponentiated effects that allow a more straightforward interpretation. Each bar represents an estimate of the average percentage change in the Scope—i.e. the average percentage change in the participant stock—of an ALMP program that results from a ‘full’ cabinet change—from a cabinet in which left-wing parties hold no seats to one in which they hold all seats—conditional on the particular program characteristic of interest and with all other variables set at their means. First, consider the bar on the left in the panel on the left. It indicates that, for a program that targets only ‘core’ groups, the estimated average effect of a change of government is a 48% increase in Scope.16 In contrast to this rather substantial effect, the bar on the right in the panel on the left shows that the estimated effect of Left-wing government strength on Scope for programs that target ‘non-core’ groups is less than 1% and far from being statistically significant. These results strongly corroborate hypothesis H1. As shown in the panel on the right, a full shift to a left-wing government is estimated to have a considerable effect on programs that entail a broken unemployment spell. For an average program with this feature, the estimated effect is a 46% increase in Scope. For programs that do not break the unemployment spell, the effect is –9%, which is in the hypothesized direction but not large enough to be statistically significant. Taken together, these results lend fairly strong support to hypothesis H2. In sum, the overall results provide evidence of substantial heterogeneous partisan effects on the governments’ ALMP ‘effort’ in programs with varying characteristics, which is in line with the theoretical argument outlined above. In addition, Table A5 provides an identical set of models applied to the logged annual program Expenditure, which produces coefficients that are only slightly smaller than those for Scope in Table 2 and that are all significant to the same extent. Moreover, as reported in Table A6, with one interesting exception,17 the estimates from Model 5 are found to be in the same direction and of a similar or larger value when the sample is split into five country clusters: Scandinavian, Continental, Anglo-Saxon, Southern and Central and Eastern European. Although the coefficients in these models are not significant in most cases, they provide a preliminary indication that no particular country or welfare regime drives the results. Additional models of Scope reported in the Supplementary Material (Tables S3–S7), suggest that the results are robust to other measures of left-wing power, to the inclusion of only one of the two key hypothesized variables and its interaction at a time, and to a number of sample adjustments. Finally, the Supplementary Material (Table S8) reports a set of models that follow the standard approach in the field by aggregating—for each of the four program categories—annual expenditures on the country-year level and expressing them as shares of GDP. In line with expectations, the results from these models indicate that, on the country-year level, Left-wing government strength is specifically associated with Core target group programs and Broken unemployment programs. 6. Concluding discussion This article aims to make two contributions to the comparative ALMP literature. First, it introduces the rich program-level ALMP data in the EU LMP database, which allows for a much more detailed analysis of ALMP programs across Europe compared with the aggregate expenditure data that currently dominates the field. Second, it advances an argument about how the impact of partisanship on ALMP varies across programs with varying characteristics, which differs from arguments in all three dominant strands of the literature. Whereas the historically dominant ‘power resources’ account holds that ALMPs are primarily promoted by the left (Esping-Andersen, 1990; Janoski, 1990; Janoski, 1994; Rothstein, 1996; Boix, 1998; Huo et al., 2008), the results presented here show that, for programs that target groups outside the ‘core’ of the labor force—which accounted for up to 40% of the observations—right-wing governments are just as expansionary as left-wing governments. In addition, the results are also at odds with the understanding of ALMP advanced in ‘insider/outsider’ perspective that Rueda (2005, 2006, 2007) introduced as a critique of the ‘power resources’ approach. At first glance, the positive interaction effect found between left-wing strength and ALMP ‘effort’ in programs that target people at the ‘core’ of the labor force seems to corroborate Rueda’s ‘insider/outsider’ hypothesis and the ‘left party disinvestment thesis’ introduced by Tepe and Vanhuysse (2013). However, the size of the interaction effect reported in Figure 2 indicates that right-wing governments match—but do not exceed—left-wing governments’ commitment to programs that also target ‘non-core’ groups. This study thus provides no evidence that left-wing governments cater less to the interests of ‘outsiders’ than right-wing governments. My understanding of why the ‘power resources’ approach and the ‘insider/outsider’ approach ascribe different ALMP preferences to parties with different ideologies is that these approaches make different implicit assumptions about the overall objective of such policies. On the one hand, in the ‘power resources’ account, ALMPs aim to reduce and/or prevent unemployment—whereby they are expected to strengthen the position of organized labor; on the other hand, in the ‘insider/outsider’ account, ALMPs seek to bring outsiders into the labor market—whereby they are expected to challenge the status of the generally better-organized workers at the ‘core’ of the labor force. The present study reconciles these perspectives by stressing that ALMPs can serve either purpose. In line with the ‘power resources’ account, left-wing governments are more inclined to expand two types of programs: those that seek to reduce or prevent unemployment among people at the ‘core’ of the labor force and those for which enrollment implies a temporary exit from open unemployment. In addition, in line with the ‘insider/outsider’ approach, left-wing parties do indeed cater to the interests of ‘core’ groups (by targeting them with ALMPs). However, this study finds that left-wing and right-wing governments are equally inclined to expand programs that also aim to increase the labor market participation of ‘non-core’ groups, which is at odds with the ‘insider/outsider’ approach. As increased labor supply is conducive to growth, the results are in line with Boix’s (1998, p. 11) basic argument that all parties have a preference for growth-enhancing policies, although the results do not support his assertion that right-wing governments ‘reject any sort of public capital formation policies’ (to which Boix counts human capital-enhancing ALMPs). Insofar as these results indicate that present-day right-wing governments are more inclined to channel public resources toward training, employment subsidies and other human capital-enhancing policies for marginal groups as a means of increasing their labor supply (instead of relying on traditional tax-reducing strategies alone), they perhaps corroborate the third strand of the literature, which argues that policy convergence has occurred in recent decades (Lindvall, 2010; Bonoli, 2013; Tepe and Vanhuysse, 2013; Nelson, 2013). However, the results clearly suggest that it is too soon to conclude, as some do, that partisan politics have lost their relevance in the ‘post-activation turn’ welfare state. Instead, the results illustrate how a traditional ideological conflict extends into the realm of ALMP programs and finds new expressions in their detailed policy settings. As such, comparative scholars need to move beyond measures of aggregate spending and more closely examine the ways in which a government that is elected to administer a modern welfare state can recalibrate (rather than revoke) the large policy portfolios that it inherits to serve its own particular objectives. Notably, the design of the present study—which, to achieve a reliable identification strategy, only exploits within-country variation—is limited because it produces effect estimates that are averaged across countries. As recent studies have reported cross-country variation in the effect of partisanship on ALMP (e.g. Vlandas, 2013), a more comprehensive exploration of how the party preferences theorized in the present study might vary across institutional environments seems like a promising endeavor for future research. The EU’s possible influence on the ALMP portfolios of its member states—particularly those in Central and Eastern Europe and those affected by the new stricter fiscal rules of the EMU—also deserves further investigation. Finally, more research into the political determinants of the composition of the ALMP program portfolios is needed. Under what circumstances are programs with different characteristics established, recast and eventually shut down? The ample and still largely unused data in the EU LMP database might very well hold the answers to those and similar questions. Supplementary material Supplementary material is available at Socio-Economic Review Journal online. 1 For a more detailed review of most of these studies, see Tepe and Vanhuysse (2013). 2 A description of the traditional categories of ALMP programs is provided in Table A1. 3 In Furåker’s (1976, p. 106) classification scheme, direct job creation, start-up incentives and early retirement measures serve to reduce unemployment but not labor shortages, whereas measures to increase the labor market participation of people who are not currently in the labor force and to expand labor immigration serve to reduce labor shortages but not unemployment. Moving grants and certain types of labor market training, as well as most placement services, serve to reduce both unemployment and labor shortages by improving matching. Some labor market policy measures, including unemployment benefits and general placement services for people who are already employed, serve none of the purposes. 4 Here, in contrast to Furåker (1976), I argue that moving grants, labor market training, and placement services do not always serve to reduce unemployment; these programs only reduce employment when they are targeted toward people who are unemployed or at risk of becoming unemployed. 5 One example is the Italian ‘work-entry contract’, Contratto di inserimento lavorativo, through which employers can provide apprenticeships to youth and certain categories of disadvantaged workers, such as the long-term unemployed, the disabled and women who live in problem areas (European Commission, 2015). 6 In some analyses [for instance, Thelen (2014, ch. 4)], STW schemes are explicitly distinguished from ALMP schemes, based precisely on their distinct target groups. This approach is unfortunate for two reasons: first, it relies on the unverified assumption that all other ALMPs target ‘outsiders’. Second, it does not fit well with the way that these policies are reported and aggregated in the widely-used labor-market policy databases, which also include programs that target the ‘employed at risk of unemployment’, such as STW programs (Grubb and Puymoyen, 2008; Eurostat, 2013). 7 Thus, programs that target people who are not already participating in the labor force might even increase the unemployment rate, to the extent that people begin to supply their labor but fail to find employment. 8 Another possible objection is that, if programs that target the unemployed are effective in putting people back into work, then output will increase, thereby encouraging right-wing governments to support these programs as well. Although this is true it does not necessarily mean that, for a given output increase caused by people reentering work from unemployment as a result of a particular program, right-wing governments will value that program as much as left-wing governments. To be clear, I do not mean to suggest that right-wing governments have no interest in programs that target the unemployed; but, because the possible distributional and electoral effects of reducing unemployment tend to benefit left-wing parties more, I argue that these will likely be more prone to support such programs. Another reason for that is that some programs that target the unemployed, such as training or direct job creation, might be seen as enablers of the kind of short-term decommodification of core workers favored by left-wing parties, as they might guarantee participants an alternative, temporary, nonmarket-based, source of income (while simultaneously serving as a legitimizing check of their willingness to work). I am grateful to one of the reviewers for making this point. 9 Residual analyses presented in the Supplementary Material (Figure S2) confirm that this operation is essential: the residuals from these regressions approximately follow a normal distribution only if it the dependent variable is log-transformed. 10 One example is the recruitment subsidy Nystartsjobb, established in Sweden in 2007. Eligible participants include not only the registered unemployed but also anyone who has been absent from the labor market for a long time (typically for more than 1 year) or who is a newly arrived immigrant (European Commission, 2015). 11 The program-level fixed effects also mitigate the plausible concern that not all parties classified as left-wing parties across the 28 countries are necessarily positioned farther to the left on the left–right continuum than all right-wing parties. Because only within-country information is used in all models, only the relative position of left-wing and right-wing parties within each country is important for this indicator of left-wing strength to spare this study from such concerns. 12 Kaplan–Meier survival estimates (not reported here) indicate that 98% of the programs in these categories survive beyond one year and that 89% survive for more than 2 years. Model S3 in the Supplementary Material is identical to Model 3 (i.e. it does not include a lagged dependent variable) but it is run on the larger sample and returns almost identical results. 13 To be precise, a few dozen programs are divided into two or more components because a single program might comprise expenditure and participant data that need to be divided between more than one program category. Here, each of these components is treated as a separate unit of observation. 14 The main effects of Core target group and Broken unemployment, unlike those of the four program categories, are not fully absorbed by the program-fixed effects because considerable within-program variation exists in these policy settings over time. Programs could conceivably be deliberately modified over time. For instance, according to the data, 16 of the 27 operating programs in Sweden changed from 1 to 0 in Broken unemployment between 2012 and 2013. These changes correspond well to legislative changes introduced in 2013, which increased the job search requirements for all recipients of unemployment insurance benefits. Still, in other cases these changes may reflect reporting errors. However, few haphazard changes occur; for only 7 of the 93 programs for which there is one or more change in Core target group and/or Broken unemployment over time, the policy setting switches back again at a later time. As a robustness check, Model S4 in the Supplementary Material omits these 93 programs. The sample is reduced by 751 observations, but the coefficients for the variables of interest are only marginally changed and are still statistically significant. 15 For the sake of parsimony, no model includes the interactions between the program category and Core target group or between the program category and Broken unemployment, as all eight are insignificant. 16 The exponentiated effect of coefficient β is calculated by taking the base of natural logarithm, e, to the power of β. As the estimated marginal effects of a change to a Left-wing government on Scope for a program with a Core target group is 0.393 in Model 5, the exponentiated effect is e0.393 = 1.481. 17 In continental countries, Broken unemployment × Left-wing government strength is markedly smaller. Acknowledgements Previous versions of this article have been presented at annual meetings of the Swedish Political Science Association in Stockholm (2015), the Swedish Network for Social Policy and Welfare Research in Lund (2015) and the Midwest Political Science Association in Chicago (2016) as well as at various seminars at Uppsala University. I thank the participants in these seminars, Per Andersson, Ingrid Esser, Daniel Fredriksson, Anders Lindbom, Karl-Oskar Lindgren, Sven Oskarsson, Joakim Palme and two anonymous reviewers for providing helpful comments and advice. Financial support from Uppsala Center for Labor Studies is gratefully acknowledged. References Angrist J. D., Pischke J.-S. ( 2009) Mostly Harmless Econometrics: An Empiricist’s Companion , Princeton, NJ, Princeton University Press. Armingeon K. ( 2007) ‘ Active Labour Market Policy, International Organizations and Domestic Politics’, Journal of European Public Policy , 14, 905– 932. Google Scholar CrossRef Search ADS   Armingeon K., Isler C., Knöpfel L., Weisstanner D., Engler S. ( 2015) Comparative Political Data Set, 1960-2013 , Bern, Institute of Political Science and Publishing House, University of Bern. Boix C. ( 1998) Political Parties, Growth and Equality: Conservative and Social Democratic Economic Strategies in the World Economy , Cambridge, Cambridge University Press. Bonoli G. ( 2010) ‘ The Political Economy of Active Labor-Market Policy’, Politics & Society , 38, 435– 457. Google Scholar CrossRef Search ADS   Bonoli G. ( 2013) The Origins of Active Social Policy: Labour Market and Childcare Policies in a Comparative Perspective , Oxford, UK, Oxford University Press. Bonoli G., Natali D. ( 2012) The Politics of the New Welfare State , Oxford, UK, Oxford University Press. Bruno G. S. F. ( 2005) ‘Estimation and Inference in Dynamic Unbalanced Panel-Data Models with a Small Number of Individuals’, STATA Journal , 5, 473– 500. Cameron A. C., Miller D. L. ( 2015) ‘ A Practitioner’s Guide to Cluster-Robust Inference’, Journal of Human Resources , 50, 317– 372. Google Scholar CrossRef Search ADS   Card D., Kluve J., Weber A. ( 2010) ‘ Active Labour Market Policy Evaluations: A Meta-Analysis’, Economic Journal , 120, F452– F477. Clasen J., Clegg D., Goerne A. ( 2016) ‘ Comparative Social Policy Analysis and Active Labour Market Policy: Putting Quality before Quantity’, Journal of Social Policy , 45, 21– 38. Google Scholar CrossRef Search ADS   Eichhorst W., Kaufmann O., Konle-Seidl R., Reinhard H.-J. ( 2008) ‘Bringing the Jobless into Work? An Introduction to Activation Policies’. In Eichhorst W., Kaufmann O., Konle-Seidl R. (eds) Bringing the Jobless into Work? Experiences with Activation Schemes in Europe and the US , Berlin and Heidelberg, Springer. Esping-Andersen G. ( 1990) The Three Worlds of Welfare Capitalism , Cambridge, Polity and Publishing House. European Commission ( 2015). ‘Reference Metadata: Labour Market Policy (source: DG EMPL) (lmp)’, accessed at http://ec.europa.eu/eurostat/cache/metadata/en/lmp_esms.htm on June 4, 2015. Eurostat ( 2013) Labour Market Policy Statistics. Methodology 2013, Report, Eurostat. Eurostat ( 2015a) ‘Dataset: Activation of Registered Unemployed (source: DG EMPL)’, accessed at http://ec.europa.eu/eurostat/en/web/products-datasets/-/LMP_IND_ACTRU on November 18, 2015. Eurostat ( 2015b) ‘Dataset: GDP and Main Components (Output, Expenditure, and Income)’, accessed at http://ec.europa.eu/eurostat/en/web/products-datasets/-/NAMA_10_GDP on July 28, 2015. Eurostat ( 2015c) ‘Dataset: Population by Sex, Age, Nationality and Labour Status (1 000)’, accessed at http://ec.europa.eu/eurostat/en/web/products-datasets/-/LFSA_PGANWS on July 28, 2015. Farnsworth K. ( 2012) Social Versus Corporate Welfare: Competing Needs and Interests within the Welfare State , London, UK, Palgrave Macmillan. Farnsworth K. ( 2013) ‘ Bringing Corporate Welfare In’, Journal of Social Policy , 42, 1– 22. Google Scholar CrossRef Search ADS   Flannery M. J., Hankins K. W. ( 2013) ‘ Estimating Dynamic Panel Models in Corporate Finance’, Journal of Corporate Finance , 19, 1– 19. Google Scholar CrossRef Search ADS   Franzese R. J., Hays J. C. ( 2006) ‘Strategic Interaction among EU Governments in Active Labor Market Policy-Making: Subsidiarity and Policy Coordination under the European Employment Strategy’, European Union Politics , 7, 167– 189. Google Scholar CrossRef Search ADS   Furåker B. ( 1976) ‘Stat och arbetsmarknad: Studier i svensk rörlighetspolitik [State and Labor Market: Studies in Swedish Mobility Politics]’, Arkiv Avhandlingsserie, 0347-4909, 5. Gaston N., Rajaguru G. ( 2008) ‘ The Rise (and Fall) of Labour Market Programmes: Domestic vs. Global Factors’, Oxford Economic Papers , 60, 619– 648. Google Scholar CrossRef Search ADS   Grubb D., Puymoyen A. ( 2008) Long Time Series for Public Expenditure on Labour Market Programmes , Report, OECD Publishing. Hemerijck A. ( 2013) Changing Welfare States , Oxford, Oxford University Press. Hibbs D. A. ( 1977) ‘ Political Parties and Macroeconomic Policy’, The American Political Science Review , 71, 1467– 1487. Google Scholar CrossRef Search ADS   Hijzen A., Venn D. ( 2011) The Role of Short-Time Work Schemes during the 2008-09 Recession , Report, OECD Publishing. Huber E., Stephens J. D. ( 2001) Development and Crisis of the Welfare State: Parties and Policies in Global Markets , Chicago, University of Chicago Press. Huo J., Nelson M., Stephens J. D. ( 2008) ‘Decommodification and Activation in Social Democratic Policy: Resolving the Paradox’, Journal of European Social Policy , 18, 5– 20. Google Scholar CrossRef Search ADS   Iversen T., Stephens J. D. ( 2008) ‘ Partisan Politics, the Welfare State, and Three Worlds of Human Capital Formation’, Comparative Political Studies , 41, 600– 637. Google Scholar CrossRef Search ADS   Janoski T. ( 1990) The Political Economy of Unemployment: Active Labor Market Policy in West Germany and the United States . Berkeley, CA, University of California Press. Janoski T. ( 1994) ‘Direct State Intervention in the Labor Market: The Explanation of Active Labor Market Policy from 1950 to 1988 in Social Democratic, Conservative and Liberal Regimes’. In Janoski T., Hicks A. (eds) The Comparative Political Economy of the Welfare State , New York, NY, Cambridge University Press, pp. 54– 92. Kiviet J. F. ( 1995) ‘ On Bias, Inconsistency, and Efficiency of Various Estimators in Dynamic Panel Data Models’, Journal of Econometrics , 68, 53– 78. Google Scholar CrossRef Search ADS   Kluve J. ( 2010) ‘ The Effectiveness of European Active Labor Market Programs’, Labour Economics , 17, 904– 918. Google Scholar CrossRef Search ADS   Korpi W., Palme J. ( 1998) ‘ The Paradox of Redistribution and Strategies of Equality: Welfare State Institutions, Inequality, and Poverty in the Western Countries’, American Sociological Review , 63, 661– 687. Google Scholar CrossRef Search ADS   Lindvall J. ( 2010) Mass Unemployment and the State , Oxford, Oxford University Press. Martin J. P. ( 2014) Activation and Active Labour Market Policies in OECD Countries: Stylized Facts and Evidence on Their Effectiveness, IZA Policy Paper No. 84, IZA. Morel N., Palier B., Palme J. ( 2012) Towards a Social Investment Welfare State?: Ideas, Policies and Challenges , Bristol, Policy and Publishing House. Mucciaroni G. ( 1990) The Political Failure of Employment Policy, 1945–1982 , Pittsburgh, PA, University of Pittsburgh Press. Nelson M. ( 2013) ‘ Making Markets with Active Labor Market Policies: the Influence of Political Parties, Welfare State Regimes, and Economic Change on Spending on Different Types of Policies’, European Political Science Review , 5, 255– 277. Google Scholar CrossRef Search ADS   Nickell S. ( 1981) ‘ Biases in Dynamic Models with Fixed Effects’, Econometrica , 49, 1417– 1426. Google Scholar CrossRef Search ADS   Pierson P. ( 1994) Dismantling the Welfare State?: Reagan, Thatcher and the Politics of Retrenchment , Cambridge, Cambridge University Press. Powell G. B.Jr., Whitten G. D. ( 1993) ‘ A Cross-National Analysis of Economic Voting: Taking Account of the Political Context’, American Journal of Political Science , 37, 391– 414. Google Scholar CrossRef Search ADS   Rothstein B. ( 1996) The Social Democratic State: the Swedish Model and the Bureaucratic Problem of Social Reforms , Pittsburgh, PA, University of Pittsburgh Press. Rueda D. ( 2005) ‘ Insider–Outsider Politics in Industrialized Democracies: The Challenge to Social Democratic Parties’, American Political Science Review , 99, 61– 74. Google Scholar CrossRef Search ADS   Rueda D. ( 2006) ‘ Social Democracy and Active Labour-Market Policies: Insiders, Outsiders and the Politics of Employment Promotion’, British Journal of Political Science , 36, 385– 406. Google Scholar CrossRef Search ADS   Rueda D. ( 2007) Social Democracy inside Out: Partisanship and Labor Market Policy in Industrialized Democracies , Oxford, Oxford University Press. Scruggs L., Allan J. ( 2006) ‘ Welfare-State Decommodification in 18 OECD Countries: A Replication and Revision’, Journal of European Social Policy , 16, 55– 72. Google Scholar CrossRef Search ADS   Swenson P. A. ( 2002) Capitalists against Markets: The Making of Labor Markets and Welfare States in the United States and Sweden , New York, Oxford University Press. Tepe M., Vanhuysse P. ( 2013) ‘ Parties, Unions and Activation Strategies: The Context- Dependent Politics of Active Labour Market Policy Spending’, Political Studies , 61, 480– 504. Google Scholar CrossRef Search ADS   Thelen K. A. ( 2014) Varieties of Liberalization and the New Politics of Social Solidarity , New York, NY, Cambridge University Press. van der Brug W., van der EijK C., Franklin M. ( 2007) The Economy and the Vote: Economic Conditions and Elections in Fifteen Countries , Cambridge, UK, Cambridge University Press. van Vliet O., Koster F. ( 2011) ‘ Europeanization and the Political Economy of Active Labour Market Policies’, European Union Politics , 12, 217– 239. Google Scholar CrossRef Search ADS   Vlandas T. ( 2013) ‘ Mixing Apples with Oranges? Partisanship and Active Labour Market Policies in Europe’, Journal of European Social Policy , 23, 3– 20. Google Scholar CrossRef Search ADS   Weishaupt T. J. ( 2011) From the Manpower Revolution to the Activation Paradigm: Explaining Institutional Continuity and Change in an Integrating Europe , Amsterdam, Amsterdam University Press. Whitten G. D., Palmer H. D. ( 1999) ‘ Cross-National Analyses of Economic Voting’, Electoral Studies , 18, 49– 67. Google Scholar CrossRef Search ADS   Appendix Table A1 provides descriptions of the six traditional ALMP program categories and statistics on the share of observations in each category that has Core target group and Broken unemployment characteristics. Table A2 presents summary statistics for all the baseline variables included in the models in the article. Table A3 illustrates the structure of the data set by presenting all the utilized program-year-level data for two observations of two different programs—one in Sweden and one in Germany. Table A4 reports further detail regarding the partial conditional effects of Left-wing government strength on Scope that are presented in Figure 2. Table A5 reports five models that are identical to those in Table 2, but they are applied to the logged annual program Expenditure instead of Scope. The model with the lagged dependent variable (A4) and the LSDV(c) model (A5) both produce coefficients of interest that are only slightly smaller than those for Scope and that are still statistically significant. Finally, the models in Table A6 split the sample into five country clusters to assess whether any particular welfare regime drives the results. Precision is reduced in most models, but with few exceptions—notably that Broken unemployment × Left-wing government strength is smaller for the continental countries—the estimates are in the same direction and of a similar or larger size. Table A1. Distribution of program characteristics across ALMP program categories       Characteristics (%)   Category  Description  Observation  Core target group  Broken unemployment  Training  Covers measures that aim to improve participants’ employability through training. The measures should include some evidence of classroom teaching or supervision, specifically for the purpose of instruction. Short courses that only develop a person’s ability to get a job are considered labor market services and fall outside this category.  1997  47  62  Employment incentives  Covers measures that facilitate the recruitment of unemployed persons and other target groups or that help ensure the continued employment of persons at risk of involuntary job loss. It refers to subsidies for open market jobs that might exist or be created without the public subsidy and that will hopefully be sustainable after the end of the subsidy period.  1865  72  84  Sheltered and supported employment and rehabilitation  Covers measures that aim to promote the labor market integration of persons with reduced working capacity through sheltered or supported employment or through rehabilitation.  690  43  64  Direct job creation  Covers measures that create additional jobs, usually of community benefit or social use, to find employment for the long-term unemployed or persons who are otherwise difficult to place. It refers to subsidies for temporary, nonmarket jobs that would not exist or be created without public intervention.  714  75  69  Start-up incentives  Covers measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed. Assistance may take the form of direct cash benefits or indirect support, including loans, facility provision and business advice.  402  60  91  Labor market services  Refers to labor market interventions where participants’ main activity is job-search related and where participation usually does not result in a change in labor market status. Services also cover Public Employment Service functions that are not directly linked to participants, such as services for employers, administrative functions and general overhead.  –  –  –  Total    5668  60  72        Characteristics (%)   Category  Description  Observation  Core target group  Broken unemployment  Training  Covers measures that aim to improve participants’ employability through training. The measures should include some evidence of classroom teaching or supervision, specifically for the purpose of instruction. Short courses that only develop a person’s ability to get a job are considered labor market services and fall outside this category.  1997  47  62  Employment incentives  Covers measures that facilitate the recruitment of unemployed persons and other target groups or that help ensure the continued employment of persons at risk of involuntary job loss. It refers to subsidies for open market jobs that might exist or be created without the public subsidy and that will hopefully be sustainable after the end of the subsidy period.  1865  72  84  Sheltered and supported employment and rehabilitation  Covers measures that aim to promote the labor market integration of persons with reduced working capacity through sheltered or supported employment or through rehabilitation.  690  43  64  Direct job creation  Covers measures that create additional jobs, usually of community benefit or social use, to find employment for the long-term unemployed or persons who are otherwise difficult to place. It refers to subsidies for temporary, nonmarket jobs that would not exist or be created without public intervention.  714  75  69  Start-up incentives  Covers measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed. Assistance may take the form of direct cash benefits or indirect support, including loans, facility provision and business advice.  402  60  91  Labor market services  Refers to labor market interventions where participants’ main activity is job-search related and where participation usually does not result in a change in labor market status. Services also cover Public Employment Service functions that are not directly linked to participants, such as services for employers, administrative functions and general overhead.  –  –  –  Total    5668  60  72  Source: Eurostat (2013), European Commission (2015). The EU LMP database also includes the two other categories Out-of-work income maintenance and support and Early retirement. These LMP categories are usually referred to as ‘support’ or ‘passive’ schemes. Table A1. Distribution of program characteristics across ALMP program categories       Characteristics (%)   Category  Description  Observation  Core target group  Broken unemployment  Training  Covers measures that aim to improve participants’ employability through training. The measures should include some evidence of classroom teaching or supervision, specifically for the purpose of instruction. Short courses that only develop a person’s ability to get a job are considered labor market services and fall outside this category.  1997  47  62  Employment incentives  Covers measures that facilitate the recruitment of unemployed persons and other target groups or that help ensure the continued employment of persons at risk of involuntary job loss. It refers to subsidies for open market jobs that might exist or be created without the public subsidy and that will hopefully be sustainable after the end of the subsidy period.  1865  72  84  Sheltered and supported employment and rehabilitation  Covers measures that aim to promote the labor market integration of persons with reduced working capacity through sheltered or supported employment or through rehabilitation.  690  43  64  Direct job creation  Covers measures that create additional jobs, usually of community benefit or social use, to find employment for the long-term unemployed or persons who are otherwise difficult to place. It refers to subsidies for temporary, nonmarket jobs that would not exist or be created without public intervention.  714  75  69  Start-up incentives  Covers measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed. Assistance may take the form of direct cash benefits or indirect support, including loans, facility provision and business advice.  402  60  91  Labor market services  Refers to labor market interventions where participants’ main activity is job-search related and where participation usually does not result in a change in labor market status. Services also cover Public Employment Service functions that are not directly linked to participants, such as services for employers, administrative functions and general overhead.  –  –  –  Total    5668  60  72        Characteristics (%)   Category  Description  Observation  Core target group  Broken unemployment  Training  Covers measures that aim to improve participants’ employability through training. The measures should include some evidence of classroom teaching or supervision, specifically for the purpose of instruction. Short courses that only develop a person’s ability to get a job are considered labor market services and fall outside this category.  1997  47  62  Employment incentives  Covers measures that facilitate the recruitment of unemployed persons and other target groups or that help ensure the continued employment of persons at risk of involuntary job loss. It refers to subsidies for open market jobs that might exist or be created without the public subsidy and that will hopefully be sustainable after the end of the subsidy period.  1865  72  84  Sheltered and supported employment and rehabilitation  Covers measures that aim to promote the labor market integration of persons with reduced working capacity through sheltered or supported employment or through rehabilitation.  690  43  64  Direct job creation  Covers measures that create additional jobs, usually of community benefit or social use, to find employment for the long-term unemployed or persons who are otherwise difficult to place. It refers to subsidies for temporary, nonmarket jobs that would not exist or be created without public intervention.  714  75  69  Start-up incentives  Covers measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed. Assistance may take the form of direct cash benefits or indirect support, including loans, facility provision and business advice.  402  60  91  Labor market services  Refers to labor market interventions where participants’ main activity is job-search related and where participation usually does not result in a change in labor market status. Services also cover Public Employment Service functions that are not directly linked to participants, such as services for employers, administrative functions and general overhead.  –  –  –  Total    5668  60  72  Source: Eurostat (2013), European Commission (2015). The EU LMP database also includes the two other categories Out-of-work income maintenance and support and Early retirement. These LMP categories are usually referred to as ‘support’ or ‘passive’ schemes. Table A2. Summary statistics Variable  Mean  Std. Dev.  Min  Max  N  Dependent variables   Scope (log)  7.77  2.51  0  14.84  5668   Expenditure (log)  2.41  2.64  −4.77  8.81  7025  Country-level variables   Left-wing government strength  0.35  0.34  0  1  5668   Real GDP  6.63E+7  8.07E+7  6.08E+5  3.1E+8  5668   Real GDP growth  1.69  3.10  −14.81  11.62  5668   Unemployed and inactive (log)  14.9  1.29  11.58  16.79  5668   Government deficit  −2.07  4.71  −32.55  18.70  5668   Government debt  69.26  32.85  7.234  144.0  5668   Number of programs in operation  33.67  16.22  4  66  5668  Program-level variables   Core target group  0.60  0.49  0  1  5668   Broken unemployment  0.72  0.45  0  1  5668   Training  0.35  0.48  0  1  5668   Employment incentives  0.33  0.47  0  1  5668   Sheltered employment and rehabilitation  0.12  0.33  0  1  5668   Direct job creation  0.13  0.33  0  1  5668   Start-up incentives  0.07  0.26  0  1  5668  Variable  Mean  Std. Dev.  Min  Max  N  Dependent variables   Scope (log)  7.77  2.51  0  14.84  5668   Expenditure (log)  2.41  2.64  −4.77  8.81  7025  Country-level variables   Left-wing government strength  0.35  0.34  0  1  5668   Real GDP  6.63E+7  8.07E+7  6.08E+5  3.1E+8  5668   Real GDP growth  1.69  3.10  −14.81  11.62  5668   Unemployed and inactive (log)  14.9  1.29  11.58  16.79  5668   Government deficit  −2.07  4.71  −32.55  18.70  5668   Government debt  69.26  32.85  7.234  144.0  5668   Number of programs in operation  33.67  16.22  4  66  5668  Program-level variables   Core target group  0.60  0.49  0  1  5668   Broken unemployment  0.72  0.45  0  1  5668   Training  0.35  0.48  0  1  5668   Employment incentives  0.33  0.47  0  1  5668   Sheltered employment and rehabilitation  0.12  0.33  0  1  5668   Direct job creation  0.13  0.33  0  1  5668   Start-up incentives  0.07  0.26  0  1  5668  Table A2. Summary statistics Variable  Mean  Std. Dev.  Min  Max  N  Dependent variables   Scope (log)  7.77  2.51  0  14.84  5668   Expenditure (log)  2.41  2.64  −4.77  8.81  7025  Country-level variables   Left-wing government strength  0.35  0.34  0  1  5668   Real GDP  6.63E+7  8.07E+7  6.08E+5  3.1E+8  5668   Real GDP growth  1.69  3.10  −14.81  11.62  5668   Unemployed and inactive (log)  14.9  1.29  11.58  16.79  5668   Government deficit  −2.07  4.71  −32.55  18.70  5668   Government debt  69.26  32.85  7.234  144.0  5668   Number of programs in operation  33.67  16.22  4  66  5668  Program-level variables   Core target group  0.60  0.49  0  1  5668   Broken unemployment  0.72  0.45  0  1  5668   Training  0.35  0.48  0  1  5668   Employment incentives  0.33  0.47  0  1  5668   Sheltered employment and rehabilitation  0.12  0.33  0  1  5668   Direct job creation  0.13  0.33  0  1  5668   Start-up incentives  0.07  0.26  0  1  5668  Variable  Mean  Std. Dev.  Min  Max  N  Dependent variables   Scope (log)  7.77  2.51  0  14.84  5668   Expenditure (log)  2.41  2.64  −4.77  8.81  7025  Country-level variables   Left-wing government strength  0.35  0.34  0  1  5668   Real GDP  6.63E+7  8.07E+7  6.08E+5  3.1E+8  5668   Real GDP growth  1.69  3.10  −14.81  11.62  5668   Unemployed and inactive (log)  14.9  1.29  11.58  16.79  5668   Government deficit  −2.07  4.71  −32.55  18.70  5668   Government debt  69.26  32.85  7.234  144.0  5668   Number of programs in operation  33.67  16.22  4  66  5668  Program-level variables   Core target group  0.60  0.49  0  1  5668   Broken unemployment  0.72  0.45  0  1  5668   Training  0.35  0.48  0  1  5668   Employment incentives  0.33  0.47  0  1  5668   Sheltered employment and rehabilitation  0.12  0.33  0  1  5668   Direct job creation  0.13  0.33  0  1  5668   Start-up incentives  0.07  0.26  0  1  5668  Table A3. Comparison of two program-year observations from two ALMP programs Observation ID  SE84_2008  DE16_2008  Name in English  New start jobs  Recruitment subsidy for new businesses  Intervention ID (panel ID)  SE84  DE16  Year  2008  2008  Country  Sweden  Germany  Program category  4. Employment incentives  4. Employment incentives  Year started  2007  1998  Year ended  (ongoing as of 2013)  2010  Target groups   Registered unemployed  All; LTU; disabled; immigrants/ethnic min.  All; public priorities and other   Other registered jobseekers  All; disabled; immigrants/ethnic min.  –   Not registered  All; disabled; immigrants/ethnic min.  –   Employed  –  –  Core target group  Non-core  Core  Treatment of unemployment spell  Broken  Broken  Scope (Participant-years)  15 921  4655  Expenditure (M €, 2005 prices)  77.6  50.1  Scope (log)  ln(15,921) = 9.68  ln(4,655) = 8.45  Expenditure (log)  ln(77.6) = 4.35  ln(50.1) = 3.91  Observation ID  SE84_2008  DE16_2008  Name in English  New start jobs  Recruitment subsidy for new businesses  Intervention ID (panel ID)  SE84  DE16  Year  2008  2008  Country  Sweden  Germany  Program category  4. Employment incentives  4. Employment incentives  Year started  2007  1998  Year ended  (ongoing as of 2013)  2010  Target groups   Registered unemployed  All; LTU; disabled; immigrants/ethnic min.  All; public priorities and other   Other registered jobseekers  All; disabled; immigrants/ethnic min.  –   Not registered  All; disabled; immigrants/ethnic min.  –   Employed  –  –  Core target group  Non-core  Core  Treatment of unemployment spell  Broken  Broken  Scope (Participant-years)  15 921  4655  Expenditure (M €, 2005 prices)  77.6  50.1  Scope (log)  ln(15,921) = 9.68  ln(4,655) = 8.45  Expenditure (log)  ln(77.6) = 4.35  ln(50.1) = 3.91  Source:European Commission (2015). Table A3. Comparison of two program-year observations from two ALMP programs Observation ID  SE84_2008  DE16_2008  Name in English  New start jobs  Recruitment subsidy for new businesses  Intervention ID (panel ID)  SE84  DE16  Year  2008  2008  Country  Sweden  Germany  Program category  4. Employment incentives  4. Employment incentives  Year started  2007  1998  Year ended  (ongoing as of 2013)  2010  Target groups   Registered unemployed  All; LTU; disabled; immigrants/ethnic min.  All; public priorities and other   Other registered jobseekers  All; disabled; immigrants/ethnic min.  –   Not registered  All; disabled; immigrants/ethnic min.  –   Employed  –  –  Core target group  Non-core  Core  Treatment of unemployment spell  Broken  Broken  Scope (Participant-years)  15 921  4655  Expenditure (M €, 2005 prices)  77.6  50.1  Scope (log)  ln(15,921) = 9.68  ln(4,655) = 8.45  Expenditure (log)  ln(77.6) = 4.35  ln(50.1) = 3.91  Observation ID  SE84_2008  DE16_2008  Name in English  New start jobs  Recruitment subsidy for new businesses  Intervention ID (panel ID)  SE84  DE16  Year  2008  2008  Country  Sweden  Germany  Program category  4. Employment incentives  4. Employment incentives  Year started  2007  1998  Year ended  (ongoing as of 2013)  2010  Target groups   Registered unemployed  All; LTU; disabled; immigrants/ethnic min.  All; public priorities and other   Other registered jobseekers  All; disabled; immigrants/ethnic min.  –   Not registered  All; disabled; immigrants/ethnic min.  –   Employed  –  –  Core target group  Non-core  Core  Treatment of unemployment spell  Broken  Broken  Scope (Participant-years)  15 921  4655  Expenditure (M €, 2005 prices)  77.6  50.1  Scope (log)  ln(15,921) = 9.68  ln(4,655) = 8.45  Expenditure (log)  ln(77.6) = 4.35  ln(50.1) = 3.91  Source:European Commission (2015). Table A4. Partial conditional effects of left-wing government on ALMP ‘effort’ (Scope) Program characteristic   Δy/Δx  Std. Dev.  P-value  e^(Δy/Δx)  Core  0.393  0.064  0.000  1.482  Non-core  0.006  0.078  0.934  1.006  Broken  0.376  0.053  0.000  1.457  Continuous  −0.097  0.087  0.264  0.907  Program characteristic   Δy/Δx  Std. Dev.  P-value  e^(Δy/Δx)  Core  0.393  0.064  0.000  1.482  Non-core  0.006  0.078  0.934  1.006  Broken  0.376  0.053  0.000  1.457  Continuous  −0.097  0.087  0.264  0.907  Effects of a change from 0% to 100% left-wing cabinet seats, with all other variables at their means. Estimations produced from Model 5. Table A4. Partial conditional effects of left-wing government on ALMP ‘effort’ (Scope) Program characteristic   Δy/Δx  Std. Dev.  P-value  e^(Δy/Δx)  Core  0.393  0.064  0.000  1.482  Non-core  0.006  0.078  0.934  1.006  Broken  0.376  0.053  0.000  1.457  Continuous  −0.097  0.087  0.264  0.907  Program characteristic   Δy/Δx  Std. Dev.  P-value  e^(Δy/Δx)  Core  0.393  0.064  0.000  1.482  Non-core  0.006  0.078  0.934  1.006  Broken  0.376  0.053  0.000  1.457  Continuous  −0.097  0.087  0.264  0.907  Effects of a change from 0% to 100% left-wing cabinet seats, with all other variables at their means. Estimations produced from Model 5. Table A5. Regression results for models of Expenditure   (A1)  (A2)  (A3)  (A4)  (A5)  Lagged dependent variable        0.60***  0.70***          (0.03)  (0.01)  Left-wing government strength  0.58***  −0.06  0.03  −0.12  −0.15    (0.11)  (0.22)  (0.22)  (0.16)  (0.11)  Core target group × Left-wing gov’t    0.27  0.26  0.34***  0.34***      (0.18)  (0.20)  (0.11)  (0.08)  Broken unemployment × Left-wing gov’t    0.60***  0.65***  0.39**  0.37***      (0.20)  (0.19)  (0.15)  (0.10)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.01  0.01  0.01  0.02  0.02***    (0.01)  (0.01)  (0.01)  (0.01)  (0.01)   Unemployed and inactive (log)  1.57**  1.66**  1.59**  1.53***  1.59***    (0.67)  (0.68)  (0.66)  (0.38)  (0.23)   Government deficit  −0.02  −0.02  −0.02  −0.01  −0.01    (0.01)  (0.01)  (0.01)  (0.01)  (0.00)   Government debt  −0.01***  −0.01***  −0.01***  −0.01***  −0.01***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.01  −0.01  −0.01  −0.00  −0.00    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.12  −0.24*  −0.21  −0.23***  −0.23***    (0.14)  (0.14)  (0.13)  (0.07)  (0.09)   Broken unemployment  0.08  −0.05  −0.07  −0.03  −0.01    (0.12)  (0.12)  (0.12)  (0.07)  (0.10)   Employment incentives × Left-wing gov’t      −0.19  −0.10  −0.06        (0.14)  (0.11)  (0.09)   Sheltered empl. and rehab. × Left-wing gov’t      −0.44**  −0.31**  −0.29**        (0.18)  (0.12)  (0.13)   Direct job creation × Left-wing gov’t      0.42  0.34  0.32**        (0.48)  (0.29)  (0.14)   Start-up incentives × Left-wing gov’t      −0.36  −0.18  −0.16        (0.39)  (0.27)  (0.13)  Constant  −20.41**  −21.47**  −20.55**  −20.93***      (9.91)  (10.01)  (9.83)  (5.67)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  7025  7025  7025  7025  7025  Programs  1052  1052  1052  1052  1052    (A1)  (A2)  (A3)  (A4)  (A5)  Lagged dependent variable        0.60***  0.70***          (0.03)  (0.01)  Left-wing government strength  0.58***  −0.06  0.03  −0.12  −0.15    (0.11)  (0.22)  (0.22)  (0.16)  (0.11)  Core target group × Left-wing gov’t    0.27  0.26  0.34***  0.34***      (0.18)  (0.20)  (0.11)  (0.08)  Broken unemployment × Left-wing gov’t    0.60***  0.65***  0.39**  0.37***      (0.20)  (0.19)  (0.15)  (0.10)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.01  0.01  0.01  0.02  0.02***    (0.01)  (0.01)  (0.01)  (0.01)  (0.01)   Unemployed and inactive (log)  1.57**  1.66**  1.59**  1.53***  1.59***    (0.67)  (0.68)  (0.66)  (0.38)  (0.23)   Government deficit  −0.02  −0.02  −0.02  −0.01  −0.01    (0.01)  (0.01)  (0.01)  (0.01)  (0.00)   Government debt  −0.01***  −0.01***  −0.01***  −0.01***  −0.01***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.01  −0.01  −0.01  −0.00  −0.00    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.12  −0.24*  −0.21  −0.23***  −0.23***    (0.14)  (0.14)  (0.13)  (0.07)  (0.09)   Broken unemployment  0.08  −0.05  −0.07  −0.03  −0.01    (0.12)  (0.12)  (0.12)  (0.07)  (0.10)   Employment incentives × Left-wing gov’t      −0.19  −0.10  −0.06        (0.14)  (0.11)  (0.09)   Sheltered empl. and rehab. × Left-wing gov’t      −0.44**  −0.31**  −0.29**        (0.18)  (0.12)  (0.13)   Direct job creation × Left-wing gov’t      0.42  0.34  0.32**        (0.48)  (0.29)  (0.14)   Start-up incentives × Left-wing gov’t      −0.36  −0.18  −0.16        (0.39)  (0.27)  (0.13)  Constant  −20.41**  −21.47**  −20.55**  −20.93***      (9.91)  (10.01)  (9.83)  (5.67)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  7025  7025  7025  7025  7025  Programs  1052  1052  1052  1052  1052  Models A1–A4: Cluster-robust standard errors in parentheses (clustered at the country level). Model A5: Bootstrapped standard errors in parentheses. Bias correction initialized by Arellano and Bond estimator (Bruno 2005). *P < 0.10, **P < 0.05, ***P < 0.01. Table A5. Regression results for models of Expenditure   (A1)  (A2)  (A3)  (A4)  (A5)  Lagged dependent variable        0.60***  0.70***          (0.03)  (0.01)  Left-wing government strength  0.58***  −0.06  0.03  −0.12  −0.15    (0.11)  (0.22)  (0.22)  (0.16)  (0.11)  Core target group × Left-wing gov’t    0.27  0.26  0.34***  0.34***      (0.18)  (0.20)  (0.11)  (0.08)  Broken unemployment × Left-wing gov’t    0.60***  0.65***  0.39**  0.37***      (0.20)  (0.19)  (0.15)  (0.10)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.01  0.01  0.01  0.02  0.02***    (0.01)  (0.01)  (0.01)  (0.01)  (0.01)   Unemployed and inactive (log)  1.57**  1.66**  1.59**  1.53***  1.59***    (0.67)  (0.68)  (0.66)  (0.38)  (0.23)   Government deficit  −0.02  −0.02  −0.02  −0.01  −0.01    (0.01)  (0.01)  (0.01)  (0.01)  (0.00)   Government debt  −0.01***  −0.01***  −0.01***  −0.01***  −0.01***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.01  −0.01  −0.01  −0.00  −0.00    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.12  −0.24*  −0.21  −0.23***  −0.23***    (0.14)  (0.14)  (0.13)  (0.07)  (0.09)   Broken unemployment  0.08  −0.05  −0.07  −0.03  −0.01    (0.12)  (0.12)  (0.12)  (0.07)  (0.10)   Employment incentives × Left-wing gov’t      −0.19  −0.10  −0.06        (0.14)  (0.11)  (0.09)   Sheltered empl. and rehab. × Left-wing gov’t      −0.44**  −0.31**  −0.29**        (0.18)  (0.12)  (0.13)   Direct job creation × Left-wing gov’t      0.42  0.34  0.32**        (0.48)  (0.29)  (0.14)   Start-up incentives × Left-wing gov’t      −0.36  −0.18  −0.16        (0.39)  (0.27)  (0.13)  Constant  −20.41**  −21.47**  −20.55**  −20.93***      (9.91)  (10.01)  (9.83)  (5.67)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  7025  7025  7025  7025  7025  Programs  1052  1052  1052  1052  1052    (A1)  (A2)  (A3)  (A4)  (A5)  Lagged dependent variable        0.60***  0.70***          (0.03)  (0.01)  Left-wing government strength  0.58***  −0.06  0.03  −0.12  −0.15    (0.11)  (0.22)  (0.22)  (0.16)  (0.11)  Core target group × Left-wing gov’t    0.27  0.26  0.34***  0.34***      (0.18)  (0.20)  (0.11)  (0.08)  Broken unemployment × Left-wing gov’t    0.60***  0.65***  0.39**  0.37***      (0.20)  (0.19)  (0.15)  (0.10)  Country-level controls             Real GDP  −0.00  −0.00  −0.00  −0.00  −0.00***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.01  0.01  0.01  0.02  0.02***    (0.01)  (0.01)  (0.01)  (0.01)  (0.01)   Unemployed and inactive (log)  1.57**  1.66**  1.59**  1.53***  1.59***    (0.67)  (0.68)  (0.66)  (0.38)  (0.23)   Government deficit  −0.02  −0.02  −0.02  −0.01  −0.01    (0.01)  (0.01)  (0.01)  (0.01)  (0.00)   Government debt  −0.01***  −0.01***  −0.01***  −0.01***  −0.01***    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Number of programs in operation  −0.01  −0.01  −0.01  −0.00  −0.00    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)  Program-level controls             Core target group  −0.12  −0.24*  −0.21  −0.23***  −0.23***    (0.14)  (0.14)  (0.13)  (0.07)  (0.09)   Broken unemployment  0.08  −0.05  −0.07  −0.03  −0.01    (0.12)  (0.12)  (0.12)  (0.07)  (0.10)   Employment incentives × Left-wing gov’t      −0.19  −0.10  −0.06        (0.14)  (0.11)  (0.09)   Sheltered empl. and rehab. × Left-wing gov’t      −0.44**  −0.31**  −0.29**        (0.18)  (0.12)  (0.13)   Direct job creation × Left-wing gov’t      0.42  0.34  0.32**        (0.48)  (0.29)  (0.14)   Start-up incentives × Left-wing gov’t      −0.36  −0.18  −0.16        (0.39)  (0.27)  (0.13)  Constant  −20.41**  −21.47**  −20.55**  −20.93***      (9.91)  (10.01)  (9.83)  (5.67)    Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  7025  7025  7025  7025  7025  Programs  1052  1052  1052  1052  1052  Models A1–A4: Cluster-robust standard errors in parentheses (clustered at the country level). Model A5: Bootstrapped standard errors in parentheses. Bias correction initialized by Arellano and Bond estimator (Bruno 2005). *P < 0.10, **P < 0.05, ***P < 0.01. Table A6. Regression results for models of Scope, by welfare regime cluster   (A6)  (A7)  (A8)  (A9)  (A10)  Lagged dependent variable  0.69***  0.65***  0.59***  0.65***  0.35***    (0.07)  (0.08)  (0.11)  (0.09)  (0.04)  Left-wing government strength  −0.58  −0.20  −0.60  −0.54*  −0.85**    (0.40)  (0.27)  (1.68)  (0.30)  (0.40)  Core target group × Left-wing gov’t  0.31  0.48**  0.83  0.73***  0.33    (0.23)  (0.23)  (1.23)  (0.28)  (0.29)  Broken unemployment × Left-wing gov’t  0.63*  0.13  2.91  0.38  0.89***    (0.33)  (0.26)  (1.75)  (0.28)  (0.31)  Country-level controls             Real GDP  0.00  −0.00***  −0.00  −0.00  0.00*    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.04**  −0.02  −0.12  0.08**  0.05***    (0.02)  (0.02)  (0.11)  (0.03)  (0.01)   Unemployed and inactive (log)  −1.05  0.28  3.85  0.19  −0.06    (0.73)  (0.78)  (4.30)  (1.56)  (0.98)   Government deficit  −0.00  −0.03**  0.16  −0.01  0.04*    (0.02)  (0.02)  (0.11)  (0.02)  (0.02)   Government debt  0.01  −0.01  0.26  −0.00  0.02**    (0.00)  (0.00)  (0.17)  (0.01)  (0.01)   Number of programs in operation  −0.02  −0.00  −0.00  −0.01*  −0.00    (0.02)  (0.01)  (0.06)  (0.01)  (0.01)  Program-level controls             Core target group  −0.53**  −0.23  0.50  −0.20  −1.16***    (0.25)  (0.19)  (0.65)  (0.24)  (0.30)   Broken unemployment  −0.16  0.66*  −0.02  −0.84***  −0.65    (0.13)  (0.34)  (0.18)  (0.18)  (0.47)   Employment incentives × Left-wing gov’t  0.24  0.09  0.95  −0.22  −0.53    (0.35)  (0.25)  (0.97)  (0.27)  (0.41)   Sheltered empl. and rehab. × Left-wing gov’t  −0.14  −0.46**  0.29  −0.21  −0.50    (0.25)  (0.22)  (1.02)  (0.26)  (0.39)   Direct job creation × Left-wing gov’t  0.11  0.72**  0.04  −0.45  −0.09    (0.36)  (0.32)  (0.74)  (0.30)  (0.52)   Start-up incentives × Left-wing gov’t  0.05  0.18  1.02  0.07  −0.40    (0.25)  (0.60)  (1.24)  (0.34)  (0.45)  Constant  17.56*  0.06  −61.98  1.04  6.43    (10.08)  (11.97)  (62.39)  (23.82)  (14.17)  Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  929  2095  333  1143  1168  Programs  112  276  41  222  245    (A6)  (A7)  (A8)  (A9)  (A10)  Lagged dependent variable  0.69***  0.65***  0.59***  0.65***  0.35***    (0.07)  (0.08)  (0.11)  (0.09)  (0.04)  Left-wing government strength  −0.58  −0.20  −0.60  −0.54*  −0.85**    (0.40)  (0.27)  (1.68)  (0.30)  (0.40)  Core target group × Left-wing gov’t  0.31  0.48**  0.83  0.73***  0.33    (0.23)  (0.23)  (1.23)  (0.28)  (0.29)  Broken unemployment × Left-wing gov’t  0.63*  0.13  2.91  0.38  0.89***    (0.33)  (0.26)  (1.75)  (0.28)  (0.31)  Country-level controls             Real GDP  0.00  −0.00***  −0.00  −0.00  0.00*    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.04**  −0.02  −0.12  0.08**  0.05***    (0.02)  (0.02)  (0.11)  (0.03)  (0.01)   Unemployed and inactive (log)  −1.05  0.28  3.85  0.19  −0.06    (0.73)  (0.78)  (4.30)  (1.56)  (0.98)   Government deficit  −0.00  −0.03**  0.16  −0.01  0.04*    (0.02)  (0.02)  (0.11)  (0.02)  (0.02)   Government debt  0.01  −0.01  0.26  −0.00  0.02**    (0.00)  (0.00)  (0.17)  (0.01)  (0.01)   Number of programs in operation  −0.02  −0.00  −0.00  −0.01*  −0.00    (0.02)  (0.01)  (0.06)  (0.01)  (0.01)  Program-level controls             Core target group  −0.53**  −0.23  0.50  −0.20  −1.16***    (0.25)  (0.19)  (0.65)  (0.24)  (0.30)   Broken unemployment  −0.16  0.66*  −0.02  −0.84***  −0.65    (0.13)  (0.34)  (0.18)  (0.18)  (0.47)   Employment incentives × Left-wing gov’t  0.24  0.09  0.95  −0.22  −0.53    (0.35)  (0.25)  (0.97)  (0.27)  (0.41)   Sheltered empl. and rehab. × Left-wing gov’t  −0.14  −0.46**  0.29  −0.21  −0.50    (0.25)  (0.22)  (1.02)  (0.26)  (0.39)   Direct job creation × Left-wing gov’t  0.11  0.72**  0.04  −0.45  −0.09    (0.36)  (0.32)  (0.74)  (0.30)  (0.52)   Start-up incentives × Left-wing gov’t  0.05  0.18  1.02  0.07  −0.40    (0.25)  (0.60)  (1.24)  (0.34)  (0.45)  Constant  17.56*  0.06  −61.98  1.04  6.43    (10.08)  (11.97)  (62.39)  (23.82)  (14.17)  Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  929  2095  333  1143  1168  Programs  112  276  41  222  245  Cluster-robust standard errors in parentheses (clustered at the program level). *P < 0.10, **P < 0.05, ***P < 0.01. Model A6: Scandinavian (DK, FI, NO, SE). Model A7: Continental (AT, BE, DE, FR, NL, LU). Model A8: Anglo-Saxon (IE, UK). Model A9: Southern (CY, EL, ES, IT, MT, PT). Model A10: Central and Eastern European (BG, CZ, EE, HU, LT, LV, PL, RO, SI, SK). Table A6. Regression results for models of Scope, by welfare regime cluster   (A6)  (A7)  (A8)  (A9)  (A10)  Lagged dependent variable  0.69***  0.65***  0.59***  0.65***  0.35***    (0.07)  (0.08)  (0.11)  (0.09)  (0.04)  Left-wing government strength  −0.58  −0.20  −0.60  −0.54*  −0.85**    (0.40)  (0.27)  (1.68)  (0.30)  (0.40)  Core target group × Left-wing gov’t  0.31  0.48**  0.83  0.73***  0.33    (0.23)  (0.23)  (1.23)  (0.28)  (0.29)  Broken unemployment × Left-wing gov’t  0.63*  0.13  2.91  0.38  0.89***    (0.33)  (0.26)  (1.75)  (0.28)  (0.31)  Country-level controls             Real GDP  0.00  −0.00***  −0.00  −0.00  0.00*    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.04**  −0.02  −0.12  0.08**  0.05***    (0.02)  (0.02)  (0.11)  (0.03)  (0.01)   Unemployed and inactive (log)  −1.05  0.28  3.85  0.19  −0.06    (0.73)  (0.78)  (4.30)  (1.56)  (0.98)   Government deficit  −0.00  −0.03**  0.16  −0.01  0.04*    (0.02)  (0.02)  (0.11)  (0.02)  (0.02)   Government debt  0.01  −0.01  0.26  −0.00  0.02**    (0.00)  (0.00)  (0.17)  (0.01)  (0.01)   Number of programs in operation  −0.02  −0.00  −0.00  −0.01*  −0.00    (0.02)  (0.01)  (0.06)  (0.01)  (0.01)  Program-level controls             Core target group  −0.53**  −0.23  0.50  −0.20  −1.16***    (0.25)  (0.19)  (0.65)  (0.24)  (0.30)   Broken unemployment  −0.16  0.66*  −0.02  −0.84***  −0.65    (0.13)  (0.34)  (0.18)  (0.18)  (0.47)   Employment incentives × Left-wing gov’t  0.24  0.09  0.95  −0.22  −0.53    (0.35)  (0.25)  (0.97)  (0.27)  (0.41)   Sheltered empl. and rehab. × Left-wing gov’t  −0.14  −0.46**  0.29  −0.21  −0.50    (0.25)  (0.22)  (1.02)  (0.26)  (0.39)   Direct job creation × Left-wing gov’t  0.11  0.72**  0.04  −0.45  −0.09    (0.36)  (0.32)  (0.74)  (0.30)  (0.52)   Start-up incentives × Left-wing gov’t  0.05  0.18  1.02  0.07  −0.40    (0.25)  (0.60)  (1.24)  (0.34)  (0.45)  Constant  17.56*  0.06  −61.98  1.04  6.43    (10.08)  (11.97)  (62.39)  (23.82)  (14.17)  Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  929  2095  333  1143  1168  Programs  112  276  41  222  245    (A6)  (A7)  (A8)  (A9)  (A10)  Lagged dependent variable  0.69***  0.65***  0.59***  0.65***  0.35***    (0.07)  (0.08)  (0.11)  (0.09)  (0.04)  Left-wing government strength  −0.58  −0.20  −0.60  −0.54*  −0.85**    (0.40)  (0.27)  (1.68)  (0.30)  (0.40)  Core target group × Left-wing gov’t  0.31  0.48**  0.83  0.73***  0.33    (0.23)  (0.23)  (1.23)  (0.28)  (0.29)  Broken unemployment × Left-wing gov’t  0.63*  0.13  2.91  0.38  0.89***    (0.33)  (0.26)  (1.75)  (0.28)  (0.31)  Country-level controls             Real GDP  0.00  −0.00***  −0.00  −0.00  0.00*    (0.00)  (0.00)  (0.00)  (0.00)  (0.00)   Real GDP growth  0.04**  −0.02  −0.12  0.08**  0.05***    (0.02)  (0.02)  (0.11)  (0.03)  (0.01)   Unemployed and inactive (log)  −1.05  0.28  3.85  0.19  −0.06    (0.73)  (0.78)  (4.30)  (1.56)  (0.98)   Government deficit  −0.00  −0.03**  0.16  −0.01  0.04*    (0.02)  (0.02)  (0.11)  (0.02)  (0.02)   Government debt  0.01  −0.01  0.26  −0.00  0.02**    (0.00)  (0.00)  (0.17)  (0.01)  (0.01)   Number of programs in operation  −0.02  −0.00  −0.00  −0.01*  −0.00    (0.02)  (0.01)  (0.06)  (0.01)  (0.01)  Program-level controls             Core target group  −0.53**  −0.23  0.50  −0.20  −1.16***    (0.25)  (0.19)  (0.65)  (0.24)  (0.30)   Broken unemployment  −0.16  0.66*  −0.02  −0.84***  −0.65    (0.13)  (0.34)  (0.18)  (0.18)  (0.47)   Employment incentives × Left-wing gov’t  0.24  0.09  0.95  −0.22  −0.53    (0.35)  (0.25)  (0.97)  (0.27)  (0.41)   Sheltered empl. and rehab. × Left-wing gov’t  −0.14  −0.46**  0.29  −0.21  −0.50    (0.25)  (0.22)  (1.02)  (0.26)  (0.39)   Direct job creation × Left-wing gov’t  0.11  0.72**  0.04  −0.45  −0.09    (0.36)  (0.32)  (0.74)  (0.30)  (0.52)   Start-up incentives × Left-wing gov’t  0.05  0.18  1.02  0.07  −0.40    (0.25)  (0.60)  (1.24)  (0.34)  (0.45)  Constant  17.56*  0.06  −61.98  1.04  6.43    (10.08)  (11.97)  (62.39)  (23.82)  (14.17)  Program-fixed effects  Yes  Yes  Yes  Yes  Yes  Year-fixed effects  Yes  Yes  Yes  Yes  Yes  Observations  929  2095  333  1143  1168  Programs  112  276  41  222  245  Cluster-robust standard errors in parentheses (clustered at the program level). *P < 0.10, **P < 0.05, ***P < 0.01. Model A6: Scandinavian (DK, FI, NO, SE). Model A7: Continental (AT, BE, DE, FR, NL, LU). Model A8: Anglo-Saxon (IE, UK). Model A9: Southern (CY, EL, ES, IT, MT, PT). Model A10: Central and Eastern European (BG, CZ, EE, HU, LT, LV, PL, RO, SI, SK). © The Author 2017. Published by Oxford University Press and the Society for the Advancement of Socio-Economics. All rights reserved. For Permissions, please email: journals.permissions@oup.com

Journal

Socio-Economic ReviewOxford University Press

Published: Apr 26, 2017

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off