1. The growing interest in ALMP in comparative social policy

The term ALMP covers a range of public programmes aimed at increasing employment, enabling people to move into jobs and achieving a better match between labour supply and demand. The OECD labour market policy database (see Martin, Reference Martin2014: 5–6) records measures targeted at specific groups of individuals at risk in the labour market that fall into one of six broad intervention categories: public employment services (PES) and administration, training, employment incentives, supported and sheltered employment, direct job creation and start-up incentives. These are contrasted with ‘passive’ labour market policies of out-of-work income maintenance and support. ALMPs thus defined are themselves far from new, and in some countries – most famously Sweden – have long been a core part of the ‘social model’. However, ALMP was until recently rarely an object of mainstream social policy scholarship. The few cross-national comparative studies of ALMP, whether quantitative (Janoski, Reference Janoski1990; Boix, Reference Boix1998) or qualitative (Weir and Skocpol, Reference Weir, Skocpol, Evans, Rueschemeyer and Skocpol1985; Therborn, Reference Therborn1986; Schmid et al., Reference Schmid, Reissert and Bruche1992), were generally framed by broader interests in macro-economic policy and combating unemployment, and stood at one remove from debates in social policy.

This changed with the ‘activation turn’ of the 1990s (Nelson, Reference Nelson2013), which saw governments in developed countries increasingly exhorted to complement their so-called ‘passive’ labour market policies, like unemployment benefits and social assistance, with supposedly more ‘active’ programmes. This more explicit association of ALMP instruments with the reform agenda around core pillars of the welfare state, as conventionally conceived, sparked a growing interest in ALMP among social policy scholars and led to a proliferation of comparative studies. Some of this work has been normative, pointing to links between enhanced benefit conditionality and ALMP as indicators of general developments towards ‘workfare’ (e.g., Lødemel and Trickey, Reference Lødemel and Trickey2001; Lødemel and Moreira, Reference Lødemel and Moreira2014). Other studies aimed to identify particular national styles of activation (e.g., Barbier, Reference Barbier2001), or sought to understand variation in the timing and pace of reforms across countries (e.g., Clasen and Clegg, Reference Clasen and Clegg2006; Daguerre, Reference Daguerre2007; Weishaupt, Reference Weishaupt2011). A common concern of this social policy oriented research into ALMP and activation has been the mapping of policy differences, as well as the ways in which policy features have been modified over time. Although some studies developed explanatory accounts, the objective of arriving at a better understanding of ALMP variation appeared to be as, if not more, relevant to this mainly qualitative scholarship than the testing of alternative explanations for cross-national differences.

From around 2005, a new wave of comparative research has however emerged which is primarily based on quantitative data analyses and centrally concerned with identifying the determinants of cross-national variations in ALMP, invariably expressed in expenditure terms. In this literature, public spending on ALMP is thus the sole object of empirical analysis, though it is seen as a proxy for a rather diverse range of underlying phenomena, including a renewed emphasis on ‘employment promotion’, a trend towards ‘workfare’ or an expression of ‘activation’. The analysis of change in ALMP across time and space serves as evidence to support competing theoretical arguments related to the changing role of partisanship in general and of social democracy specifically (for example, Rueda, Reference Rueda2005, Reference Rueda2006; Huo et al., Reference Huo, Nelson and Stephens2008; Vis, Reference Vis2011), the effects of trade union strength (Tepe and Vanhuysse, Reference Tepe and Vanhuysse2013), the role of the European Union in domestic policy (Armingeon, Reference Armingeon2007; van Vliet and Koster, Reference Van Vliet and Koster2011) and the implications for cross-national policy variation of domestic and international, or institutional and economic, factors (for example, Gaston and Rajaguru, Reference Gaston and Rajaguru2008; Swank, Reference Swank2011).

Invariably, this body of research has made use of OECD expenditure data, with analyses typically starting in 1985 and restricted to fifteen or sixteen countries for reasons of data availability. It might be expected that the use of the same data source, similarities in country selection and a common period of investigation should have generated some robust and increasingly consensual findings. Instead, the conformity in research interest and data application has, to date, created more rather than less ambiguity about, for example, the role of partisanship. In part, this lack of intellectual progress may reflect differences in the specification of independent and control variables, or the failure to disaggregate ALMP spending and focus on different types of programmes (Vlandas, Reference Vlandas2013; Bonoli, Reference Bonoli2013). However, as we suggest below, arguably a more relevant reason for the failure to arrive at a more common understanding of variation across countries and over time lies in the rather scant attention paid to the particular nature and theoretical substance of ALMP in most of this recent research. Before addressing these issues, however, we will first draw attention to some more strictly methodological challenges that arise in the use of expenditure data for comparative research on ALMP.

2. Reliability issues

All of the quantitative analyses mentioned above rely on the OECD, which offers publicly available data sets containing information on aggregate ALMP expenditure across countries and over time. As discussed further below (see Section 4), the acknowledgement that ALMP is a rather broad category has also seen some more recent quantitative studies (for example, Bonoli, Reference Bonoli2010, Reference Bonoli, Clasen and Clegg2011; Nelson, Reference Nelson2013; Vlandas, Reference Vlandas2013) exploit the disaggregation of this expenditure data by programme type, allowing separate analyses of spending on training, job placement, job creation and so on. However, the reliance on OECD data – whether aggregated or disaggregated – means that these studies are capturing much but not all expenditure on ALMP cross-nationally. The main reason for this is the tradition of responsibility for ALMP being substantially decentralised in many developed countries. An allied reason is the fact that the requirement, in principle, for national statistical agencies to report on both national and sub-national provision is, in fact, not followed through in practice. Many sub-national expenditures therefore go under-reported in comparative data. An additional factor that is gaining in importance relates to the implications of the marketisation of ALMP delivery systems for the availability of sufficiently detailed information on the types of services that are being provided in some countries. Each issue can be briefly illustrated in turn.

Local and regional authorities often play a major role in the funding and delivery of ALMP programmes. Particularly in countries where the provision of social assistance benefits and services is the responsibility of municipalities, significant parts of ALMP provision are the result of autonomous decisions by local government and are highly independent from central government directives. Sweden offers one good example, as does Germany before the labour market and social security reforms of 2005. While in these cases social assistance claimants are (or were) in principle entitled to access some ALMP provision from the offices of the national public employment service (PES), in practice co-ordination and co-operation deficits between different government levels resulted in these claimants facing serious barriers to accessing support. In this context, the municipalities stepped in to fund, design and provide their own employment services reserved for recipients of municipally funded benefits (see Angelin et al., Reference Angelin, Johansson, Koch and Panican2013; Kaps, Reference Kaps, Bothfeld, Sesselmeier and Bogedan2012).

Even in countries where benefit provision is wholly centralised, sub-national governments sometimes chose to complement existing provision from the central PES with their own provision, due to the perceived inadequacy of national policies in the context of the dynamics of political competition across different levels of governance. The current situation in the UK is illustrative of this. In Scotland in 2012, expenditure on so-called employability provision delivered by local authorities (around £70 million), financed mainly by the Scottish government and the European Social Fund, considerably surpassed the expenditure in Scotland from the UK government's flagship ALMP measure, the Work Programme (around £55 million) (Goerne and Clegg, Reference Goerne and Clegg2013). However, the most recently available qualitative report on UK ALMP statistics makes it clear that no Scottish sub-national provision is recorded in the overall UK data provided to Eurostat and the OECD (Eurostat, 2011).

The frequent omission of local ALMP spending from comparative datasets inevitably leads to a serious distortion of any kind of comparative analysis based on these sources. This is obvious for cross-sectional comparisons across countries with highly variable roles for local authorities in providing ALMP. However, the problem is equally relevant when making comparisons over time, as apparent trends in expenditure may simply be artefacts of shifts in central/local governance responsibility for ALMP. For example, far-reaching social security reforms in Germany in 2005 shifted the responsibility for providing employment services for social assistance claimants from the municipalities to institutions with the involvement of central level PES (Clasen and Goerne, Reference Clasen and Goerne2011). As a result, services to social assistance claimants are today recorded in the national (and, by extension, comparative) ALMP data, which was not the case before 2005. As such, the mid-decade decline in ALMP expenditure, which the OECD recorded for Germany, probably considerably understates the actual fall in expenditure in this policy area at this time.

The traditional limitations of the OECD data are furthermore being exacerbated by the growing delivery role of private agencies in the context of the development of ‘quasi markets’ for publicly funded employment programmes (Sol and Westerveld, Reference Sol and Westerveld2005; van Berkel et al., Reference van Berkel, de Graaf and Sirovátka2011). Though this trend has gone further in some places than others, most developed capitalist countries have at least experimented in recent years with the contracting-out of ALMP delivery to private providers, ostensibly to try and encourage service innovation and leverage effectiveness and efficiency gains. Thoroughgoing privatisation, however, further compromises the quality of national administrative, and by extension comparative, data in this policy field. An extreme illustration is the case of so-called ‘black box’ contracting, the use of which is most clearly seen in the current UK Work Programme (Finn, Reference Finn2012). Black box contracts are essentially non-prescriptive, as regards the specific types of service provision, with providers paid largely for achieving certain outcomes rather than for delivering a specific set of services. In other words, the principal (normally the PES) explicitly refrains from specifying the service content to be delivered by the agent (a private company), and the latter cannot be obliged to reveal this information ex post, not least for reasons of commercial confidentiality. The move towards performance-based rewards and away from remuneration based on service fees therefore makes accounting for the specific content of ALMP interventions much more difficult. Just as comparative analyses of ALMP are beginning to focus on expenditure patterns disaggregated by type of instrument or intervention, in many countries this information is, as a result, becoming much more problematic for national statistical services to obtain or record in any reliable way.

The general point illustrated by the foregoing discussion is that while the highly codified and centrally regulated nature of traditional welfare state programmes, such as pensions, sick pay or unemployment benefits, ensures that national expenditure data on them are fairly encompassing and offer a relatively high degree of reliability across developed countries, the same cannot be said with any degree of certainty for the comparative datasets based on national administrative sources that are used in most comparative analyses of ALMP.

3. Validity issues

Even if comparative datasets recorded ALMP expenditures in different countries in a comprehensive and consistent way, and at a sufficient level of programmatic detail, there would arguably remain considerable limits to their ability to measure the intensity of work-focused support for the unemployed across countries and over time. This is, however, what researchers believe them to be measuring, whether it is called ‘employment promotion’, ‘activation’ or something else. To put this another way, alongside issues with the reliability of comparative data on ALMP there are also serious problems with their validity. This can be illustrated here with respect to two rather different issues, namely the particularly acute inferential problems related to attempts to control for varying levels of demand for ALMP cross-nationally, on the one hand, and the questionable extent to which ALMP data genuinely measures work-focused support for the unemployed, on the other.

As almost all studies of ALMP acknowledge, expenditure dynamics are not only the result of policy settings, but also depend on factors such as economic development in general and unemployment trends in particular. Comparative analyses have therefore sought to take account of both, whether through their inclusion as control variables or by calculating the ratio of ALMP expenditure (relative to GDP) per percentage point of unemployment. Accordingly, several studies use the ratio of spending on both as an (additional) indicator of ‘ALMP effort’ and change (for example Huo et al., Reference Huo, Nelson and Stephens2008; Bonoli Reference Bonoli2010, Reference Bonoli2013), and some (Vis, Reference Vis2007) as a key indicator for the assessment of change within and across welfare state types.Footnote ¹ Of course, controlling for varying levels of demand or ‘problem pressure’ in this way is an established practice in comparative analyses of expenditure on traditional social policy programmes. It seeks to strip out the effect of differing shares of elderly people in the population on retirement expenditures or of varying rates of unemployment on unemployment benefit spending. Unemployment rates that, as mentioned, are used to capture demand in most analyses of ALMP expenditure, are notoriously problematic in comparative analysis as they are expressed as a ratio of the labour force (all the people subject to the risk of unemployment). Rates of labour force participation vary considerably cross-nationally, particularly at the extremes of the age distribution and amongst women. They are also affected by the very different ways that labour markets and social policy programmes in developed countries are structured, particularly with respect to the extent of part-time employment and the administrative categorisation of non-employment (Erlinghagen and Knuth, Reference Erlinghagen and Knuth2010; Sengenberger, Reference Sengenberger2011). Sharp declines in unemployment resulting, for example, from the expansion of early retirement or growing part-time work can thus generate a largely artificial image of increasing ALMP ‘effort’ (as well as of the effectiveness of ALMPs), which is arguably precisely what was seen in Denmark and the Netherlands in the 1990s (van Oorschot and Abrahamson, Reference van Oorschot and Abrahamson2003).

While the above problem of controlling for demand in a valid way is common to analyses of both ALMP expenditure and unemployment benefit expenditure, the analyses of ALMP face an additional inferential difficulty that analyses of unemployment benefits or other cash transfer schemes do not. While the latter typically do not change their beneficiaries’ situation and administrative status, in many countries ALMP measures very deliberately do precisely that, as participants are no longer counted as unemployed (for a discussion, see Sengenberger, Reference Sengenberger2011: 84–85). Increased expenditure on certain categories of ALMP – such as training or direct job creation measures – will under these circumstances tend to reduce open unemployment as beneficiaries move to alternative labour market statuses (respectively inactive or employed in these two examples), albeit without necessarily finding regular employment. With reference to the Danish case, it is well known that the expansion of participation in ALMP was one of the major drivers behind declines in (open) unemployment in the 1990s (Bredgaard and Jørgensen, Reference Bredgaard and Jørgensen2000). For cross-national comparisons of ALMP, this kind of mechanism creates the obvious potential that indicators of ‘effort’ will be endogenously ratcheted upwards or downwards by increases or decreases in expenditure on measures that have a direct impact on the unemployment rate, itself typically used as the denominator in calculating ‘effort’.

This is just one reason why it is questionable how much evidence of higher or lower ALMP ‘effort’ across countries or over time actually tells us about the extent of underlying intensity of work-focused support for the unemployed. Another reason is that it is unclear that ALMP data are always mainly recording work-focused support for the unemployed. Oldervoll (Reference Oldervoll2014) has pointed out that the OECD data include information on numerous national programmes that are not actually designed to facilitate entry into the open labour market, but that serve as implicit or explicit alternatives to work for people who are not expected to be looking for employment. This can be shown, for example, by their very low participant inflow and outflow rates, and average individual participation periods of sometimes considerably more than ten years in some schemes in Sweden and Norway. Considerable shares of national ALMP spending in other countries may also focus on ‘sheltered employment’, as in the Netherlands, or be devoted to permanent wage subsidies for people with reduced work capacity, as in Denmark (ibid.). Of course, these types of programmes provide income to participants and other valuable benefits derived from being in a work activity. However, their inclusion in the ALMP portfolio suggests that (total) spending as reported by national governments, and thus included in international databases, should not necessarily be seen as a response to the problem of unemployment or as wholly (or even mainly) directed at the unemployed.Footnote ²

While ALMP might – as in this example – sometimes be regarded as an indicator that is too broad, from a different vantage point it might equally be considered as being too narrow. As ‘activation’ principles have gained ground, governments in many countries have converted measures traditionally targeted narrowly at the unemployed, and which are thus counted in ALMP statistics, into general measures that benefit all those active in the labour market, which are not counted. For example, between 1995 and 2005 expenditure in France on fiscal employment incentives targeted on the unemployed, as counted in OECD data, fell from 0.22 per cent of GDP to 0.12 per cent. Over the same period, expenditure on so-called allègements généraux – employment incentives targeted not on the basis of employment status but on the basis of salary level – were expanded massively, with expenditure increasing from 0.3 per cent of GDP in 1995 to 1 per cent in 2005. Similarly, the contemporary portrayal of the UK as a low-spender on ALMP compared with other OECD countries (see, for example, Berry, Reference Berry2014) is in large part due to the exclusion of expenditure on generalised wage subsidies in the form of tax credits, which in the early 2000s absorbed more than five times the expenditure devoted to the government's flagship ALMP measure, the New Deal (Clasen, Reference Clasen2005: 83). The omission of such non-targeted but arguably functionally equivalent programmes can lead to fairly major inferential errors in studies operationalising ‘activation’ or ‘employment promotion’ through ALMP expenditure data alone.

4. Conceptual and theoretical issues

This problem of ALMP expenditure being potentially either too narrow or too broad as an indicator of social policy change suggests that comparative research is challenged not only by issues of data reliability and validity, but also needs to pay closer attention to the theoretical substance of spending data, and the underlying concept of ALMP ‘effort’, for understanding the determinants and dynamics of state intervention. In other words, more consideration needs to be given to the conceptual specification of the ‘dependent variable’ in comparative ALMP research.

The concept of ALMP ‘effort’ recalls that of welfare ‘effort’, which featured heavily in the first generation of quantitative comparative analyses of traditional welfare state programmes (Wilensky, Reference Wilensky1975; Cameron, Reference Cameron1978; Stephens, Reference Stephens1979; Schmidt, Reference Schmidt1983). Esping-Andersen (Reference Esping-Andersen1989: 19) famously criticised this generation of scholarship for its reliance on social expenditure data, which he noted was ‘epiphenomenal to the theoretical substance of welfare states’. This critique has often been misunderstood as a concern about the use of expenditure data per se (for example, Jensen, Reference Jensen2011), when it was rather the underlying conceptualisation – or perhaps the absence of conceptualisation – of welfare state activity in this literature that was Esping-Andersen's true target. His argument was that, prior to embarking on empirical analysis of welfare state determinants, it is first necessary to identify criteria or concepts that could ‘capture the ideals or designs that historical actors sought to realise in the struggles over the welfare state’ (Esping-Andersen, Reference Esping-Andersen1989: 20). Building on the insights of Polanyi and others, Esping-Andersen advocated concepts such as ‘decommodification’ and ‘stratification’ as expressions of the outcomes of such struggles, and set about operationalising these for comparative empirical enquiry.

The important point here is that strong parallels can be drawn between the conceptual limitations of the early comparative welfare state scholarship and that of recent cross-national quantitative research on ALMP. The use of the concept of ALMP ‘effort’ has been justified by the argument that this is a ‘one size fits all’ policy area that appeals to conservatives, liberals and social democrats alike (Armingeon, Reference Armingeon2007: 906), or by stating that all different ALMP programmes have some common characteristics from the point of view of unemployed people (Huo, Reference Huo2009: 103). However, at a certain level of abstraction similar arguments could of course be advanced about traditional welfare state programmes, or indeed any public policy. But just as aggregate welfare state expenditure data mask policies of widely differing political inspiration and distributive effect, so ALMP ‘effort’ can be made up of expenditures on programmes that have markedly different effects on unemployed individuals and on the labour market, and presumably align in very different ways with the interests of employers and unions or the ideologies of conservatives, liberals and social democrats. While this has long been recognised in more qualitative comparative approaches to ‘activation’ or ‘workfare’ (e.g. Lødemel and Trickey, Reference Lødemel and Trickey2001; Barbier, Reference Barbier2001), in quantitative comparative research this aspect is often ignored. In view of this, it is hardly surprising that the controversies over the determinants of ALMP effort have proved rather difficult to resolve.

Recently, however, some comparative quantitative research on ALMP has engaged with and sought to address this problem. For example, Vlandas (Reference Vlandas2013: 4) has analysed the determinants of selected sub-types of ALMP, based on the programmatic categorisation in the OECD database. He argues that the determinants of ‘direct job creation’ measures can be assumed to differ from the determinants of ‘private sector employment incentives’, as the former are unambiguously in the interests of both the unemployed and employees, while the latter can be expected to place downward pressure on wages, and are thus unlikely to be popular with (employed) core social democratic voters.

This and similar approaches (e.g., Tepe and Vanhuysse, Reference Tepe and Vanhuysse2013) represent a substantial advance in sophistication on studies using aggregate expenditure as a proxy for ALMP ‘effort’. There are however a number of problems which arise from basing an analysis of discrete types of labour market policy intervention on categories established for administrative, rather than scientific or theoretical, purposes. For example, the OECD category of ‘direct job creation’ includes but does not differentiate between public employment schemes introduced in periods of high unemployment to the benefit of ‘core workers’, and ‘make work’ measures that target very long-term unemployed benefit claimants who are unlikely to find regular employment even when unemployment is low.

For example, the relatively well paid German job creation schemes introduced to cushion mass unemployment in East Germany after reunification (Arbeitsbeschaffungsmaßnahmen), are clearly qualitatively different from the more recent job creation schemes (Arbeitsgelegenheiten), which offer low remuneration and are mainly designed to ‘socially integrate’ claimants considered very distant from the labour market.

Though both classified as ‘direct job creation’ in comparative data, these programmes have very distinctive aims and presumably very different political support coalitions. Somewhat similarly, the extent to which ‘job subsidy’ schemes are likely to exert downward pressure on wages is crucially determined by their design features, including their target groups and extent to which they reach up the wage distribution

The statistical accounting of these measures is blind to such distinctions – all job subsidy schemes are similarly classified by Eurostat and the OECD as ‘employment incentives’. In practice however, such distinctions matter considerably, not least with respect to their acceptance by unions and social democratic parties across Europe.

In short, while disaggregating the concept of ALMP into policies which can be assumed to be determined differently seems sensible, relevant concepts as well as derived categories need to reflect theoretical considerations rather than administrative conventions, which is a point which we will revisit below (Section 5).

The theoretical limitations of recent social policy scholarship on ALMP are not only confined to challenges of conceptualisation of the dependent variable, but also extend to the range of independent or explanatory variables. As mentioned above (Section 1), aside from a few studies on the role of the European Union and other international organisations, the explanations for cross-nationally variable patterns of ALMP development have been sought among the ‘usual suspects’, i.e. political variables (partisanship, political institutions) and economic factors (economic openness, corporatism) which have featured heavily in comparative research on classic welfare state programmes. The underlying but unarticulated assumption here is that ALMPs are essentially comparable to cash transfer programmes as dependent variables for comparative analysis (for a discussion, see Clasen and Clegg, Reference Clasen, Clegg, Clasen and Siegel2007), and their development should therefore be explained in a broadly similar way.

However, in at least three important respects, ALMPs are actually quite unlike cash tranfers, such as pensions or unemployment benefit. First, while cash transfers are – almost by definition – entitlement programmes, most ALMPs are a form of discretionary spending, and the politics of discretionary spending and entitlement programmes are known to differ in significant ways (e.g., see Breunig and Busemeyer, Reference Breunig and Busemeyer2012). Governments do not normally need to embark on the arduous process of changing legislation to modify their ALMP ‘effort’, but can instead make ongoing adjustments through regular budgetary procedures. An implication of this is that the possibility frontier for ALMP expenditure will be shaped by cross-national variations in systems of public expenditure planning and control that are of limited explanatory relevance with respect to entitlement programmes. As a major pioneering qualitative comparative analysis of the determinants of ALMP underlined, a significant role in explaining cross-national variation in this policy field is played by different financing systems (Schmid et al., Reference Schmid, Reissert and Bruche1992), an institutional variable which is not usually foregrounded in comparative analyses of cash transfers.

Second, ALMP is a policy that does not have the same unambiguous impact as classical welfare state programmes. When proposing modifications to the legal rules governing entitlement programmes, social and political actors can be relatively certain of the impact that the changes they advocate will have on the incomes of all concerned populations.Footnote ³ At both the individual and aggregate levels, the effect of increased expenditure on one or other type of ALMP is by contrast far more difficult to predict, as the ample evaluation literature on this subject has shown (for a review, see Card et al., Reference Card, Kluve and Weber2010). While participating in a training programme may positively improve the employment prospects of an unemployed individual, equally it could have no effect at all or even a negative one. A targeted job subsidy programme may lead to additional job creation, or might equally crowd out unsubsidised work through deadweight losses and substitution effects. At a minimum, such uncertainty around the individual and aggregate effects of ALMPs makes it very difficult to unambiguously ascribe clear and stable preferences for and against them to social and political actors, as standard quantitative approaches require.Footnote ⁴ More fundamentally, insofar as the evaluation evidence is relatively sober regarding the effectiveness of ALMP, it may even be necessary to consider whether the traditional mechanism of actors seeking credit by advocating policies that will benefit particular constituencies really has much traction (Bonoli, Reference Bonoli, Bonoli and Natali2012). As far back as the mid-1980s, when the Conservative government in the UK was spending massively on ALMP – particularly through direct job creation schemes such as the ‘Community Programme’ – while simultaneously pursuing policies that were driving up unemployment and doing battle with the unions, Richardson and Henning (Reference Richardson, Henning, Richardson and Henning1984) suggested that these policies were essentially ‘symbolic’ or ‘placebo’ measures. In other words, they were intended to signal concern about unemployment to massage open unemployment downwards for electoral reasons rather than to really offer anything of value to the unemployed themselves.

Finally, allied to its discretionary character, the multi-level and multi-actor nature of this policy field adds complexity not only to its measurement, but also to its politics. In this policy environment, political authorities in many countries have invested in ALMP for reasons that have very little (directly) to do with the broad economic or political factors whose influence is explored in recent literature. For example, through so-called ‘shifting yard’ policies, German local authorities for many years spent money on job creation schemes essentially to re-entitle long-term unemployed people to insurance benefits, thereby transferring the burden of supporting the jobless from (locally financed) social assistance to the social insurance system (Clasen, Reference Clasen2005: 67). Somewhat differently, the fact that many ALMP measures are provided through non-governmental third parties has always meant that provider interests have been an important part of the political story of expenditure dynamics in this field (Moon and Richardson, Reference Moon and Richardson1984), and one that is likely to become more important with the recent entrance of large multinational corporations into expanding welfare-to-work markets. In these respects, the politics of social services provision may offer an analogy that is as, or even more, useful for explanatory accounts of ALMP than the core entitlement programmes of the ‘old welfare state’.