Tying Up the Loose Ends : The Determinants of Active Labour Market Policy in the Western Countries 1985-2013

Knowledge on the determinants of Active Labour Market Policy (ALMP) spending accumulated during the 2000s. Despite these advances, the current research lacks a systematic approach to the relevant determinants. This article fills the research gaps by analysing simultaneously the 14 most frequently used determinants for the first time. In addition to these variables, this study introduces a new factor, namely the impact of economic crises. Through the analysis of the longest data period yet investigated of 20 Western countries and a comparison of methodological alternatives, this study both challenges and reinforces previous findings, as well as produces new ones. For example, it is the first investigation to reveal the positive effect of government indebtedness and economic crises on ALMP expenditure. However, the rivalry between the “usual suspects” continues, as the negative effects of budget deficits, foreign trade, and population ageing, and the positive effects of trade union density and GDP growth, were rediscovered in this analysis.


Introduction
Debate based on empirical studies on the determinants of social policy, or the factors behind policy development, gathered pace in the 2000s, along with developments in data and methods.One of the subcategories of this debate focuses on analysing the determinants of active labour market policy (hereafter ALMP).The role of ALMP is argued from the perspectives of maintaining the capacity for work of the unemployed (Barbier & Ludwig-Mayerhofer, 2004), providing encouragement to work (Kananen, 2014), and preserving the functionality of the labour markets (Bonoli, 2012), making it a particularly interesting subject for debate (Jessop, 2002).A main research strand investigates whether the development of ALMP is dependent on endogenous or exogenous determinants (cf., Schwartz, 2001).The latter dynamic predicts that ALMP contracts or expands as a direct result of globalisation.Researchers presenting counterarguments emphasise intrastate variables, such as national policy, labour market organisations, or the load-bearing capacity of the national economy (e.g., Genschel, 2004).
The present article provides an answer to the research question: What is the impact of the identified welfare state determinants on active labour market policy expenditures between 1985 and 2013?In order to test the explanatory power of several competing theories simultaneously, the study employs panel regression on extensive data from 20 countries.Panel regression studies formed the main stream of the relevant literature in the 2000s, as this approach can be used to test several theories simultaneously (De Boef & Keele, 2008).Since different panel regression methods may produce disparate results, the analysis described in the article applies both fixed effects and random effects methods.The research data were gathered from the statistical databases of the OECD (2016), IMF (2016), World Bank (2016), and European Central Bank (2016).To show ALMP dynamics, two ALMP variables were tested: ALMP expenditure in relation to gross domestic product; and, derived from this, a variable standardised by unemployment.The explanatory variables, selected on the basis of a review of the literature, are: foreign trade, foreign investments, the political left and right, unemployment, budget deficit, national debt, GDP growth, GDP per capita, proportion of the aged in the population, trade union movement, corporatism, share of the service sector in the labour force, and the European employment strategy.In addition to the 13 most used variables, this study tests the impact of the economic crises of the 1990s and 2000s on ALMP development.
In order to answer the research question and show the contribution of the article, a review of the previous literature and resulting empirical observations is required.This is undertaken in the next section.In the third section, I provide details of the data used in the analysis, the methodological choices, and the concomitant problems.Finally, the article compares the results to those in the existing literature and draws conclusions based on these findings.The article makes a contribution to the literature by identifying the principal determinants of ALMP and analysing them using panel data over a research period of 29 years.Through extensive theoretical testing, the article creates new perspectives on the debate by significantly changing the predominant current conclusions.

Determinants of Active Labour Market Policy in Theoretical and Empirical Literature
According to efficiency theory, the financial basis of the welfare state is jeopardised by globalisation, as countries are presumed to compete for international investments primarily through taxation (Strange, 1996).Therefore, competition for released investment capital leads to the systematic reduction of taxes, which are seen as hampering economic activity.Since particularly harmful taxes in this context are taxes on capital, globalisation forms a significant risk to the fiscal basis of the welfare state (Mishra, 1999).In terms of social policy, globalisation means a shift towards market-led services and income redistribution.This results in the responsibility for the actualisation of social risks shifting from the public sector towards the individual level, leading to a reduction in societal obligations, including the resources to fulfil such obligations (Shin, 2000).In addition to the above, Krugman (1999) writes about the indirect impact mechanisms of globalisation.He maintains that strengthening economic globalisation increases the political credibility of neoliberal arguments, which can also lead to decision-making that will contract the welfare state.Due to the tax-funded nature of ALMP, combined with its unclear short-term impacts on employment (Card, Kluve, & Weber, 2010) and tendency to interfere with the free labour market mechanism, its importance should decline along with the growth of globalisation (Gilbert, 2002).
The compensation theory accepts efficiency theory's premise of the direct impact of globalisation on the welfare state, but reaches opposing conclusions regarding its consequences.Referring to the work of Cameron (1978), Ruggie (1982), andKatzestein (1985), Rodrik (1998) suggests that the expansion of globalisation leads to the expansion of the welfare state.The theory posits that from the foreign investors' viewpoint, the benefits of an extensive welfare state in the availability of labour, know-how, and the stability of the operating environment, outweigh the adverse effects of taxation and market turbulence.Garrett (1998) adds a political dimension to the compensation theory.In his view, the employment sector risks (unemployment, untypical employment relationships) that increase with globalisation boost the support of left-wing parties, as they promise voters protection from such risks.Thus, the compensation theory predicts the growth of ALMP, since global markets demand increasing investment in labour in more unstable labour markets (Auer, Umit, & Leschke, 2008;Torfing, 1999).
The diffusion theories maintain that the shortcomings of the efficiency and compensation theories are related to overemphasising the economic dimension of globalisation.They take the view that globalisation is indeed linked to social policy, but argue that for the main part, these links operate indirectly through internationally networked actors and institutions (Cerny & Evans, 2000).Diffusion theories may be divided into those stressing so-called hard and soft mechanisms.The first refer to the adoption of policies because of specific coercive mechanisms.These may be rivalries between neighbouring countries seeking a competitive edge (Stone, 2000) or the coercive directives and regulations of the EU (Dostal, 2004).Soft mechanisms include national policy recommendations and guidelines.Thus, soft mechanisms cover non-obligatory participation in international agreements within organisations like the OECD, IMF, UN, and EU (Evans, 2004).The most important international policy recommendation for ALMP development is the European employment strategy ratified in 1997 in the Amsterdam Treaty (Hemerijck & Visser, 2003).
The power resource theory focuses on analysing the preferences of traditional left-and right-wing party supporters in assessing the direction of social policy (Korpi, 1983).The theory makes the assumption that industrial and service sector workers strive for a comprehensive social policy in order to protect themselves from market failures (e.g., unemployment) and life risks (e.g., sickness, old age), and social democratic parties promise to implement policies in line with their interests (Esping-Andersen, 1985).More recent power resource theory interpretations still consider ALMP to be a form of social policy acting in the interests of workers.Thus, the theory presupposes that ALMP has held its central role in social democratic policy-making despite tighter economic constraints (Huber & Stephens, 2002).Because classic power resource theory assumes that the objective of right-wing policy-making is primarily to minimise government intervention, its effect on ALMP is theorised as negative (Boix, 1998, p. 77).
Corporatist theories highlight the importance of the Trade Union (TU) movement and employer organisations in the development of ALMP.Mainstream doctrines in economics maintain that the trade union movement works primarily for the benefit of its employed members or the so-called "inner circle".Consequently, such aims as improving the social position of the unemployed are of secondary importance to the TU movement and may even be counter to its interests (Blanchard & Summers, 1986;Lindbeck & Snower, 1989).Therefore, according to this theory, ALMP expenditure should fall with the increase in the power of the TU movement.The theory is based on the premise that the TU movement does not wish to further ALMP, which is important for the unemployed, as it wants to allocate the available resources to its employed membership.The alleged adverse effects of ALMP on the employed arise through taxation, as it is usually financed by tax revenue.ALMP may also improve the competitiveness of the unemployed compared with those in employment, which in this theory is counter to the interests of the "inner circle" (Rueda, 2007, pp. 26-28).An argument opposing this theory is that ALMP can also be in the interest of the TU movement, as preparedness for employment sector risks by furthering employment is a policy in the interest of the membership, in the face of increased labour market instability resulting from globalisation (Emmenegger, 2009).
From the corporatist viewpoint, there is a suggested link between a tripartite pay model and ALMP.The theory postulates that government participation links ALMP more firmly to centralised pay policy agreement systems.Based on this, Brandl and Traxler (2005) argued that the extension of ALMP may suit employees as compensation for slow wage development and that such a move would therefore not provoke significant resistance from employers.In other words, the centralisation of pay policy provides a platform for the government and labour market organisations to implement ALMP as compensation for stagnant pay levels, especially if the possibility of increasing unemployment security is excluded due to its alleged incentive and budgetary effects (Hassel, 2003).
The deindustrialisation theory also takes a critical view of the assumption of the functional theories as to the effects of globalisation on the welfare state (Iversen & Cusack, 2000).According to the theory, the effect of globalisation is in fact very slight, as the factor with most explanatory power pertaining to change in the welfare state is a shift in the production structure of the national economy from the industrial sector towards the service sector.A further crucial factor is the increased productivity of labour as a result of developments in knowledge and technology.The deindustrialisation theory maintains that a change in production structure causes increasing uncertainty.As a result, the welfare state expands as the citizens in democratic conditions demand collective security in the face of increased risks.Despite the different indicators (deindustrialisation vs. globalisation), the logic of the operational mechanism in Iversen and Cusack's theory resembles the compensation theory.As for ALMP, Swank's (2011) interpretation of Iversen and Cusack's theory is that it should expand with the decline of the relative proportion of industrial work.
The long tradition of the tension between economic constraints (Flora, 1986) and demand for public services (Wilensky, 1975) is also present in the debate on ALMP determinants.Based on theoretical tradition (Schwartz, 2001), the argument is made that the level of government debt position and budget deficit are probably linked to the contraction of ALMP, due to the reduction in allocable public economic resources (Rueda, 2005, p. 68).The public economic resource angle is also linked to the possible effect of an ageing population on ALMP development.Because an ageing population reduces the dependency ratio, it has been linked to the contraction of ALMP, as sparse public resources shift from services for the employed toward services for the aged (Gaston & Rajaguru, 2008, p. 363).There are two alternative effects of unemployment on ALMP.On the one hand, estimates show that growing unemployment automatically raises ALMP expenditure because of growing demand.On the other hand, the argument is made that unemployment increases the demand for other aspects of social security and reduces tax revenues, so that ALMP may fall as unemployment rises (Van Vliet & Koster, 2011, p. 226).
Next, we undertake a review of prior empirical literature (Note 1).Table 1 shows that most of the existing research took place in the 2000s.The number of countries included in the analyses is high, on average about 18 western countries per study.The studies have mostly measured ALMP relative to GDP, but a few have applied an ALMP variable standardised by the unemployment rate.The oldest time series cover periods from the 1970s to the beginning of 1990, the latest from the 1980s to 2008.Therefore, it should be noted that earlier research cannot, for example, have taken into account the 2007 financial crisis, which is included in the present analysis.Moreover, Table 1 shows that none of the previous studies analysed all of the most often-used variables simultaneously.This flaw is corrected in the present study.On the other hand, we find that left-wing support, foreign trade, GDP, and unemployment are very often tested in the relevant research.Conversely, EES (European Employment Strategy), government debt, and budget deficit appear considerably less often in the studies.The analysis carried out in the present article thus clearly contributes to the research literature.Note.The number of positive or negative signs denotes the statistical loadings and the direction of the regression analysis.For example "+++" implies a statistically highly significant positive result.An abbreviation "ns." implies a non-significant regression result between the ALMP expenditure and the independent variable.The dark shading indicates that the variable was not tested in a particular study.The diagonal lineation means that the study standardized ALMP by unemployment rate.
Table 1 indicates that it is difficult to explain conclusively the development of ALMP through globalisation variables like foreign trade and foreign capital.The reason is that according to most studies, there would seem to be no statistical link between these variables and ALMP expenditure.Nevertheless, prior research does not unequivocally reject the theories positing a link between globalisation and ALMP.In terms of theoretical debate, the explanatory power of efficiency theory in predicting a race to the bottom is poor, as a clear minority of studies using functional variables supported this theory.Of economic factors, the clearest effect on ALMP would seem to be determined by economic conditions.In a recession, when the government budget deficit increases, ALMP expenditure is reduced.Conversely, during boom conditions, when the gross domestic product per person rises, a number of studies would seem to indicate a stronger ALMP.The classic power resource theory, suggesting a positive link between ALMP and left-wing party support, was confirmed to a degree by the studies.A clear positive link was found between the trade union movement and ALMP development, and this finding was also partially supported by the studies employing an indicator of corporatism.At this juncture, it should also be noted that the effect of right-wing support on ALMP development is negative or mostly statistically insignificant.A reason for the small number of studies utilising diffusion theory is that it is cumbersome to measure the effect of international organisations on ALMP expenditure using panel regression.However, three studies would seem to show that the European employment strategy ratified in 1997 may have had an effect in increasing ALMP expenditure.

Research Design
The previous section outlined the principal theories relevant to the ALMP determinant debate and the empirical research making use of them, with their main findings.This review shows that the present study fills a gap in the research literature: it is the first to test the 13 most frequently used variables simultaneously using panel regression.This article aims to fill the gaps in previous research by answering the research question: What is the impact of the most common welfare state determinants on the active labour market policy expenditure between 1985 and 2013?The data, covering a period of 29 years, includes 20 countries, namely Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, The Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, Great Britain, and the United States.Next, I will illuminate the methodological choices necessary to answer the research question and the variables used in the analysis.
The dependent variable problem is one of the main methodological challenges in welfare state research (Clasen & Siegel, 2007).In the case of the present study, the problem is crystallised in the issue of using expenditure statistics in the time series panel data, where linking the expenditure to demand is theoretically well founded.Thus, ALMP development would follow the natural demand factor (here unemployment), even if the study was also interested in the effect of other factors on the expenditure level (Jensen, 2011).The challenge of the demand factor is met in this article by adding Model 2 to the analysis, dividing the GDP variable of ALMP by the unemployment rate.The addition of Model 2 to the study is justified, as dividing ALMP expenditure per one percent of unemployment is one way of including the demand for ALMP in the analysis (Wenzelburger, Zohlnhöfer, & Wolf, 2013).An ALMP variable formed this way was, for example, used by Swank (2011) and Vliet and Koster (2011).The sources of the dependent and independent variables are listed in Table 2 below.The operationalisation of variables was modelled on previous research (cf., Table 1).In addition to the problem of the dependent variable discussed above, the problems related to the choice of the independent variables require elucidation.Briefly, this is a question of the number of theoretical factors impacting on the phenomenon studied in the analysis and how well they succeed in explaining it (Clasen & Siegel, 2007).The present article responds to this problem by taking previous peer-reviewed international research as its starting point and increasing the main identified explanatory variables to 13 in total.Moreover, the theoretical review in the previous section explains the operational mechanism of each variable tested here in relation to ALMP.The first column of Table 2 shows the variable unit being tested, the second the theoretical background on which operationalisation is based, and the third the data source.In common with all prior studies, the indicator of ALMP here is obtained from the OECD statistical database.The sources of the political variables are from the ParlGov material.Following the example of Huo et al. (2008) in the operationalisation of the power resource theory, the extreme left and right, as classified by the ParlGov material, were excluded from the analysis.Following the method of Alemán (2012), corporatism was measured by the centralisation of pay policy, assigning scores from 1 (extremely fragmented) and 5 (extremely centralised) according to the scale created by ITCTWSS (2016).Deindustrialisation is measured by the number of service sector employees relative to the total labour force, as shown by Iversen and Cusack (2001).The economic crises variable captures the three main global market disturbances of the 1990s and 2000s: the early 1990s recession, the ICT recession in the early 2000s, and the global finance crisis of 2007-2008.
In addition to the above, particular attention is required as to the technique of performing the panel regression, as different methods may produce results that are either divergent or convergent (de Boef & Keele, 2008).In recent times, two main analytical methods have been employed in ALMP research: the fixed effects model and the random effects model.In previous studies, the former was applied by Armingeon (2007) and Tepe and Vanhuyse (2013), the latter by Rueda (2007) and Kim and Zurlo (2008).The base assumptions of the two models are slightly different.The fixed effects model makes the assumption that the dependent variable is partly dependent on its previous values.This dependency is also called the unobserved factor (Halaby, 2004).The model makes the further assumption that this dependence on previous values varies between countries.Conversely, the random effects model makes the assumption that the dependent variable does not correlate strongly with itself.In statistical terms, the difference between the methods is that in the fixed effects model, the unobserved factor is defined as part of the error term (Wooldridge, 2009, p. 493).
Since there is no clear agreement as to the superiority of either method in social sciences (Halaby, 2004, p. 514), the present article tests the data using both fixed and random effects analyses.In terms of research, this may produce added value, as similar methods may produce convergent or divergent results.The main value of Stock and Watson's (2003, pp. 289-290) fixed effects model lies in the fact that by treating country-specific fixed effects as separate from the error term, the chances of finding differences between the explanatory variables are improved.Thus, in accordance with the assumptions inherent in the fixed effects model, the error terms between countries should differ from each other.This may be examined using the Hausman (1978) test.In models 1 and 2, the error terms do not intercorrelate, therefore the fixed effects model should produce more accurate results in both models.In order to understand the methodological choices, it is important to note that the Hausman test result does not preclude using the random effects model.Following Hamilton's (2012) example, the analysis was carried out using the xtreg command of the STATA (ver.13MP) program, in which one of the available criteria of panel regression is multicollinearity and its testing.In addition, heteroscedasticity was tested separately.Both tests supported executing the analysis.
As I have endeavoured to show above, the study design responds to a number of challenges relating to methodology and use of the data.One of the problems external to the present research question concerns the criticism of using aggregate variables in ALMP panel regressions, addressed mainly by Vlandas (2013).He maintains that ALMP research should use disaggregated variables, because the internal development of ALMP in some cases is a different matter from external development measured on aggregate levels.In practice, there are two impediments to applying the disaggregated view, which Vlandas also acknowledges.First, the time series data used are not sufficiently long to carry out a panel regression in accordance with the criteria and several independent variables.The expediency of analysing disaggregated variables is also questionable due to the fact that in theoretical terms, the mechanisms through which forces of change affect the subcategories of ALMP are largely outside the scope of current knowledge.Thus, a justification for using the aggregate level is that the present article focuses on quantifiable long-term change factors affecting ALMP as a whole, which can be interpreted in theoretical terms.

Results
This section analyses the data as described above, answering the research question: What is the impact of the identified welfare state determinants on active labour market policy expenditure between 1985 and 2013?Before setting out the results of the panel regression, it is useful to review the development of dependent and explanatory variables during the research period.Table 3 below shows the descriptive data or the values of the variables used in the study as the average of 20 countries.The economic recession of the early 1990s reduced GDP in most countries, resulting in a substantial rise in the GDP variable.Conversely, the variable taking unemployment into account shows that the expenditure fell considerably as the result of the recession."The golden age of active labour market policy" in the early 2000s is shown in the figure by the high level of expenditure at the time according to both indicators.However, the trend soon turns, and expenditure falls rapidly up to the economic crisis starting in 2007-2008.After this, the indicators again diverge, as was seen before at the start of the 1990s.An interesting observation in theoretical terms is the downward trend of the left wing during the research period.Over this time, foreign trade is an indicator of economic globalisation that has undergone steady growth.Foreign capital, on the other hand, increased up to the 2007-2008 financial crisis.GDP per person reflects the rapid growth of national economies across the 20 countries.Other clearer trends include the drop in influence of the trade union movement, but also the growth of service sector employment and the proportion of the aged during the study period.Despite a small drop compared with the 1980s, right-wing support has remained constant in the 2000s.The value of corporatism would indicate that pay policy agreements have shifted to some degree towards more local levels.Table 4 below sets out the results of the study.In common with the majority of previous studies, the present study found no direct effect of left-or right-wing political influence on ALMP expenditure development.Foreign trade, however, contradicts Aleman's (2012) and and Swank's (2011) studies by making an opposite finding, thus lending support to the efficiency thesis.The second variable linked to economic globalization, foreign investment, was not found to affect ALMPs.Interestingly, GDP growth and the trade union movement were found to have a strong positive effect on ALMP almost regardless of the analytical method.Similarly, a convergence with previous research is shown by Model 1 on the effect of a budget deficit, which is strongly negative.Iversen and Cusack's (2001) argument is challenged by the statistically non-significant result in both models, showing that the service-sectorisation of labour may not have a effect on ALMP.As was the case in most of the previous research, corporatism does not appear to affect ALMP expenditure.
Model 2 produces a statistically significant positive link on the effect of the European Employment Strategy (EES) on ALMP.It would seem that EES, ratified in 1997, increased ALMP in relation to unemployment, but not relative to GDP.It could be that EES had the effect of momentarily increasing ALMP in fiscal terms, but the growth in GDP reduces its effect in Model 1.In any event, the effect of EES was now tested for the first time in five years, and its positive effect is still evident in the data.The 2007 economic crisis is discernible as a statistically significant result only in Model 2. In previous literature, the effect of ageing on ALMP has been found as both positive (Kim & Zurlo, 2009) and negative (Huo et al., 2008;Iversen & Stephens, 2008;Gaston & Rajaguru, 2008).This study suggests that the growth of the senior population actually reduces ALMP spending.Through the economic mechanism, population ageing seems to pose a threat to the ALMP-like welfare services targeted at the working age population.
The effect of government debt on ALMP offers the next contrary result compared with the existing research.
Prior studies of the link between national indebtedness and ALMP have been carried out by Gaston and Rajaguru (2008) and Rueda (2005Rueda ( , 2007)).Their findings are that government debt negatively affects ALMP.The present study is the first to make an opposite observation.The result of Model 1 may be considered robust, as the fixed effects model is more suited for the analysis than the random effects model.Furthermore, it should be noted that Model 2 quantifies ALMP in relation to unemployment, thus differing significantly from Model 1 in terms of its focus.The divergence of the finding from previous research may be explained by the fact that the data for the present study cover the years 1985-2013, whereas the periods of previous studies ended in 1999 (Gaston & Rajaguru), 1998 (Rueda, 2007), and 1995(Rueda, 2005).It is possible that the longer period by 14 years produces different results from those obtained earlier, as the western countries have become considerably more indebted in this period (cf., Table 3).Thus, it seems likely that debt is used for ALMP expenditure, which alters the view of the economic policy role of ALMP offered by the previous literature.This interesting observation is worthy of more detailed attention in the next section.

Conclusions
The present article analysed the ALMP determinants most frequently used by 12 previous studies.The panel data covered 20 western countries for a record period of 29 years.The study answered the research question: What is the impact of the welfare state determinants on the active labour market policy expenditure between 1985 and 2013?The article fills gaps in the research by analysing the 14 most often used systematically obtained determinants for the first time.Through extensive theoretical testing, the study draws new conclusions, as well as reinforcing existing ones.
When comparing these results with the findings of previous studies, it is clear that the ALMPs are affected negatively by the budget deficit and population ageing, and that GDP growth is the fuel for ALMP expansion.Related to the welfare state's economic capacity, the negative effect of globalization cannot be neglected at this point.The crux of the matter is that all of these four determinants seem to develop negatively from the perspective of ALMP development in post-industrial society.Deviating from previous research, the present study found a positive link between national indebtedness and ALMP.It would seem that ALMP is a viable option for policy makers, if they are willing to ease the budgetary framework by increasing the government debt.This finding, nevertheless, reinforces the conclusion that ALMP is partially subordinated to the economic framework within which it operates.Related to this, the the impacts of economic crises of the 1990s and 2000s were found positive to the ALMP spending for the first time.These novel findings, however, do not mean that ALMPs function countercyclically, because unemployment was again found uncorrelated with ALMP spending.
Future debate on the determinants of ALMP might discuss the positive political factors counterbalancing the negative effects, which are mainly economic.Attention could be given to the opinions of the trade union movement regarding the development of ALMP.Its positive impact has been confirmed by the present and previous studies, and we can assume that a quantitative assessment would also show substantive effects.The positive impact on international levels is also discernible in the case of the European employment strategy ratified in 1997.As for the political determinants, it is worth noting that the political left or right did not seem to affect ALMP.Speaking against power resource theory in general, it seems that research will be quite adequate if focused on factors other than party programmes.To put this differently, ALMPs may not be at the focal point of the two sides of the political spectrum, as commonly held.In the era of slow growth, globalization, and ageing populations, rising uncertainty characterizes the future of ALMP in the western countries.

Table 1 .
Determinants of active labour market policy in previous studies

Table 2 .
The sources and theoretical background of independent and dependent variables

Table 4 .
The determinants of active labour market policy in the western countries