The impact of the business cycle on ALMP effectiveness in the EU

(1)

Institute of Public Administration

2019

The impact of the

business cycle on ALMP

effectiveness in the EU

MASTER THESIS ECONOMICS & GOVERNANCE

(2)

Author:

Broek, van den, M.A. (Michel), MSc (s1510711)

University: Leiden University

Faculty:

Faculty of Governance and Global Affairs

Institute:

Institute of Public Administration

Study programme:

PA: Economics and Governance

Supervisor:

Prof.dr. Koning, P.W.C. (Pierre)

Second reader:

(3)

Abstract

A substantial amount of research on the question of how effective ALMPs are in minimizing unemployment already exists. A different question, that has been raised but not been substantially answered, is on how the use of different ALMP instruments is best varied over the business cycle in order to truly maximise the effectiveness of each.

Using quantitative research methods, in the form of both descriptive statistics and regression analysis, this thesis researches what ALMP instruments are best employed when. As a secondary objective this thesis also researches what the general effects are on unemployment for different ALMP instruments.

The theory from the existing body of knowledge suggests that ALMPs with relatively large lock-in effects and an emphasis on human capital accumulation are most effective at reducing unemployment during recessions. This is mainly due to opportunity costs and the scarring, or perceived scarring, of the individual inflicted by unemployment over the course of the spell. The contrary is suggested to hold true during booms. Generally, traditional ALMP instruments, such as training, are suggested to be most effective at reducing unemployment overall.

(4)

Overview of tables

§ Table 1 – Overview of variables (3.2)

§ Table 2 – Descriptive statistics, an overview (4.1) § Table 3 – Correlation matrix (4.2)

§ Table 4 – Regression analysis, total measure of ALMP intensity (4.2) § Table 5 – Regression analysis, separate measures of ALMP intensity (4.2) § Table 6 – Overview of results (5.0)

Overview of figures

§ Figure 1 – Unemployment rate, by country (4.1)

§ Figure 2 – Unemployment rate, by year / CLI, by year (4.1) § Figure 3 – ALMP intensity, by type (4.1)

§ Figure 4 – ALMP intensity, by type & year (4.1)

§ Figure 5 – Unemployment rate & ALMP intensity, by type (.2) § Figure 6 – Interaction plot, training & CLI (4.2)

§ Figure 7 – Interaction plot, employment incentives & CLI (4.2)

§ Figure 8 – Interaction plot, supported employment & rehabilitation & CLI (4.2) § Figure 9 – Interaction plot, direct job creation & CLI (4.2)

(6)

Chapter 1: Introduction

In recent decades the share of GDP devoted to active labour market policy (hereafter: ALMP) in EU countries has increased significantly (Forslund, Fredriksson & Vikström, 2010), thereby placing more interest and emphasis on these policies. ALMPs aim to improve the prospect of finding gainful employment or to otherwise increase the earnings capacity of those participating in these programmes (European Commission, 2016). A substantial amount of research on the question of how effective ALMPs are in minimizing unemployment already exists. A different question, that has been raised but not been substantially answered, is on how ALMPs are best varied over the business cycle in order to truly maximise their effectiveness (among others: Forslund, Fredriksson & Vikström, 2010; Nordlund, 2011 and Andersen & Svarer, 2012).

To understand why this might be important to do, a couple of points have to be considered. First, a recession might see a rapid and sudden inflow of individuals into ALMP programmes. The question is whether all ALMP programmes are equally capable of maintaining quality under such circumstances or whether some are more capable in doing so than others (Andersen & Svarer, 2012). Second, the composition of the pool of unemployed persons might be quite different during booms and recessions, and certain ALMP programmes might be better suited to deal with one over the other (Andersen & Svarer, 2012). Third, so-called ‘lock-in’ effects might make certain ALMP programmes more tempting during booms than during recessions (Andersen & Svarer, 2012). Finally, the political support for ALMPs could very well be business cycle dependent (Andersen & Svarer, 2012).

It goes without say that much greater elaboration on these mechanisms will be provided in chapter 2 (theory) than in this brief introductory comment. Also, note that moving forward this thesis will refer to ALMPs in a per participating idividual measurement (ALMP intensity). Expression as a mere share of GDP would fail to capture the nuance between a given sum of money being spent on a small group of individuals on the one hand, or a large group of individuals on the other hand.

There appears to be a lack of prior research dealing with the particular subject of this thesis, although it must be noted that some prior work does exist. For instance, Lechner & Wunsch (2009), Kluve (2010) and Forslund et al. (2011) are all mentioned by (Card, Kluve & Weber, 2017) as examples. However, the scarcity of this prior research is at the same time also

(7)

reiterated by the likes of (among others) Forslund, Fredriksson & Vikström (2010), Nordlund (2011) and Andersen & Svarer (2012). In their own words they all mention the same, mainly that there is not a lot of prioe work on how ALMPs should be adapted based on the business cycle situation. Furthermore, prior research has overwhelmingly approached this subject merely from a macroeconomic perspective (Nordlund, 2011). By addressing this research gap this thesis aims to be scientifically relevant.

The lack of knowledge surrounding this subject might come as a surprise, particularly if we stop to consider the societal relevance for just a moment. First of all, it is important to establish the negative societal effects of unemployment (generally). Unemployment introduces negative effects at both the micro and macro dimensions, stretching far beyond simply the loss of earnings (Nordlund, 2011; European Commission, 2016). This mainly is a consequence of the decay or perceived decay of human capital. Logically it would therefore be considered beneficial to society to minimize unemployment. Any insight and/or knowledge which could help guide policy in achieving this goal could therefore be considered beneficial to society and it is precisely this premise from which this thesis derives its societal relevance also.

Having briefly introduced the subject of this thesis, as well as having established its relevance (both scientifically and societally), we can now move on to the formulation of the main research question that this thesis will ultimately attempt to provide an answer to. It is formulated as follows:

‘To what extent does the business cycle influence the effectiveness of ALMPs on unemployment in the EU?’

As has been mentioned before already, even though some prior knowledge concerning this particular subject does exist – it is limited. Chapter 2 (theory) will provide some idea to the question of where the answer to this main research question is going to go. Nonetheless, it might be worthwhile to emphasize the inherent explanatory nature of this thesis. Some limitations because of this are inevitable, these will be discussed in chapter 3 (methodology).

The formulation of an answer to the aforementioned main research question might be considered the primary objective of this thesis, a secondary objective might however be considered as well also. As will become clear from chapter 2 (theory), ALMPs come in a variety

(8)

of different shapes and forms. Separately from the way in which they might interact in their effectiveness on unemployment with the business cycle, we might want to consider their effectiveness generally also. This can be considered as the secondary objectives of this thesis.

To end this introductory paragraph, a short rundown of the structure of this thesis is now provided, a reading guide for what is to come. The chapter coming up next, chapter 2, will deal with the existing body of knowledge – the theory within which this thesis is positioned. A variety of elements deemed relevant for this particular thesis are covered, but in more clear terms the chapter will first deal with the general before it transitions to the specific. Labour market policy is covered generally, followed by active labour market policy and finally a link with the business cycle is established. The chapter ultimately culminates in the formulation of hypotheses. What follows thereafter is chapter 3, the methodology. It provides a detailed explanation on the question of how precisely this thesis is conducted. In short, quantitative research methods are used in the form of descriptive statistics and regression analysis. The chapter will elaborate on the exact thesis design, the variables used, the origin of the data and the considerations going into this particular design as well as its limitations. The results of this thesis and an analysis thereof are presented in chapter 4. The first section of the chapter deals with the descriptive statistics part of the thesis and builds towards the second part of the thesis. The second part of the chapter deals with the aforementioned regression analysis part of this thesis. Finally, conclusions are drawn in chapter 5. The hypotheses are accepted or rejected and an answer to the research question is formulated. A discussion and reflection are provided in the discussion paragraph, part of chapter 5.

(9)

Chapter 2: Theory

The primary function of this chapter is to position this thesis within the relevant body of knowledge. This body of knowledge could be characterised as being presented and structured as a funnel. Working from the general to the specific, the knowledge gap mentioned prior in the introduction will be addressed and become clearer in the process. Thereby the potentially added value of this thesis is clarified as well. In order to do so I conceptualise all relevant concepts and discuss theoretical explanations for the relationships between them. Ultimately this will result into the formulation of hypotheses at the end of this chapter.

As to how the aforementioned funnel will be structured, first a brief discussion of labour market policy will be presented. This should provide the broadest possible overview of the subject of research at hand. In this a distinction between passive labour market policy on the one hand and active labour market policy on the other hand, as well as their interplay, will be made. After that a transition is made to talking about the effects generated by ALMPs more elaborately. This is done by first talking about ALMP effectiveness on its own, both in general terms as well as in regard to specific measures. Next the connection to the business cycle is made. The differences between ALMP effectiveness during recessions and booms are discussed at length. A theoretical explanation for why certain ALMP measures might be more effective during one period than another is presented and discussed. Finally, in an acknowledgement of the heterogeneity inherent to this subject matter (i.e. in respect to institutional differences between countries) this chapter will go over various other factors signalled by the body of knowledge as being related and in need to be controlled for in order to obtain a proper result. Ultimately, deducted from the body of knowledge, hypotheses are formulated at the end of this chapter. These will help guide the formulation of the answer to the main research question of this thesis posed in chapter 1 (introduction).

2.1: Labour market policy

Labour market policy aims to optimize the promotion of labour while simultaneously combatting unemployment as well (Nordlund, 2011). According to Cazes, Verick & Heuer (2009) labour market policies include all regulatory policies that influence the interplay between supply and demand for labour. The main two means through which this optimization can be achieved is through passive labour market policy on the one hand and through active labour market policy on the other hand. In short, passive labour market policies are

(10)

unemployment benefits while searching for a suitable job (Nordlund, 2011) or income replacement (Cazes, Verick & Heuer, 2009). Examples on the passive labour market policy side of things include unemployment insurance, unemployment assistance and related welfare benefits (Martin, 2015). In short, active labour market policies are participation in specific programs and wage subsidies (Nordlund, 2011). These labour market integration measures are available to individuals that are unemployed or who run the risk of becoming unemployment (Cazes, Verick & Heuer, 2009). As active labour market policy specifically is prime subject of this thesis, it will be further elaborated upon in its own paragraph, 2.2, following this paragraph. Which means are most prevalent in any given mix might at least in part depend on labour market conditions and the business cycle (Nordlund, 2011). This will be further elaborated on in paragraph 2.4, specifically subparagraph 2.4.2.

Having established a general outline of labour market policy and its components, what logically follows is the question to why this is to be of interest in the first place. From an economic perspective the need for intervention can be seen as failure by the market, so what is going wrong? It is important to note that unemployment has several negative consequences, stretching beyond a simple reduction in earnings. Nordlund (2011) (see also: European Commission, 2016) points at multiple negative consequences, both at the micro as well as macro dimension. These negative consequences revolve mainly around the decay or, more importantly, perceived decay of human capital (European Commission, 2016). For these negative consequences to occur the unemployment spell does not even have to be necessarily long, indeed even short unemployment spells can have negative consequences too (Nordlund, 2011). Although it must be noted that long-term unemployment causes especially negative effects, while job search attempt also diminishes over time, making the situation problematic even more so and in need to be addressed (European Commission, 2016). Note that long-term unemployment as it is refereed to here is defined as an unemployment spell having a minimum duration of 12 months, which makes up 50% of all unemployment spell cases on average (European Commission, 2016).

2.2: Active labour market policy

According to Nordlund (2011) there are four separate aims of ALMPs that can be distinguished. These are prevention of labour market participation decline (1), prevention of earning losses (2), prevention of downward occupational mobility (3) and prevention of unstable work

(11)

arrangements (4). With the overarching function being to maintain and/or improve human capital. The European Commission (2016) says it differently but similarly. According to them the main goal of ALMPs is to increase the amount of opportunities for the unemployed to find employment by improving ‘matching’ between the unemployed and the available job opportunities. The effect of ALMPs therefore is a contribution to lowering unemployment rates, stimulating employment, lowering dependency on benefits and ultimately growing the GDP and the overall economy. However, ALMP effectiveness will be further discussed in paragraph 2.3.

The formal definition of ALMPs is provided by the OECD and is as follows: “Active labour market programmes include all social expenditure (other than education) which is aimed at the improvement of the beneficiaries’ prospect of finding gainful employment or to otherwise increase their earnings capacity. This category includes spending on public employment services and administration, labour market training, special programmes for youth when in transition from school to work, labour market programmes to provide or promote employment for unemployed and other persons (excluding young and disabled persons) and special programmes for the disabled” (European Commission, 2016, p. 1).

According to the European Commission (2016) ALMPs are, generally speaking, more specifically targeted at some groups, more so than others. These groups according to them primarily consist of young individuals, older workers, low-skilled and the long-term unemployed.

2.2.1: Specific measures

Before anything else it is important to note that the body of knowledge does not consistently divide ALMPs into set specific instruments. Instead different categorisations are used in prior research, resulting in some overlap and discrepancies in and between categories. Nonetheless the main categories of ALMP measures and their characteristics can be deducted from this overview and discussion. At the end of this paragraph, an overview of the European Commission (2006) its classification for labour market policy is provided in regard to the ALMP instruments, as this is the classification being adopted for this thesis its methodology.

According to Cazes, Verick & Heuer (2009) the main categories of ALMPs are job search assistance (JSA) and training. JSA involves a marginal investment in time and raises the

(12)

efficiency of search (Forslund, Fredriksson & Vikström, 2010). Training involves mainly ‘traditional’ education and skill development with the purpose of human capital accumulation or maintenance (Nordlund, 2011). Matrin (2015) also adds public sector employment and wage subsidies to these main categories of ALMPs. Employment involves employment distinctly different from regular labour market employment and that is being subsidised. Because it is different from regular labour market employment, the goal is not to learn or gain a skill but rather to network, stimulate abilities and motivate (Nordlund, 2011).

The European Commission (2006) classifies ALMP measures into six different categories, these and their definitions are as follows:

• Training: “measures that aim to improve the employability of LMP target groups through training, and which are financed by public bodies” (European Commission, 2006).

• Job rotation and job sharing: “measures that facilitate the insertion of an unemployed person or a person from another target group into a work placement by substituting hours worked by an existing employee” (European Commission, 2006).

• Employment incentives: “measures that facilitate the recruitment of unemployed persons and other target groups, or help to ensure the continued employment of persons at risk of involuntary job loss” (European Commission, 2006).

• Supported employment and rehabilitation: “measures that aim to promote the labour market integration of persons with reduced working capacity through supported employment and rehabilitation (European Commission, 2006).

• Direct job creation: “measures that create additional jobs, usually of community benefit or socially useful, in order to find employment for the long-term unemployed or persons otherwise difficult to place” (European Commission, 2006).

• Start-up incentives: “measures that promote entrepreneurship by encouraging the unemployed and other target groups to start their own business or to become self-employed (European Commission, 2006).

Note that job rotation and job sharing will not be covered further in this thesis, the reason for which will be covered in chapter 3 (methodology).

(13)

2.2.2: Some history

When comparing different European Union member states, noteworthy and important differences can be observed both now as well as in the past. This is very much true especially in regard to unemployment. This, and the need to address these differences, will be discussed at greater length in paragraph 2.5. However, when it comes to ALMPs, converging patterns are found in EU and OECD countries. Namely in the period between 1985 and 2008 the share of GDP devoted to ALMP measures within in the EU increased significantly and became more similar, especially in OECD countries (Forslund, Fredriksson & Vikström, 2010). According to Cazes, Verick & Heuer (2009) OECD countries in particular have made an increasing effort to emphasize the use of passive measures in order to stimulate the unemployed and underemployed. Note that because of the other factors differencing, convergence in ALMP spending does not necessarily equal convergence in ALMP intensity (a measure on the level of per participating). Kluve (2006) points mainly at the “European Employment Strategy” in order to explain the greater emphasis on ALMPs, as well as more general economic convergence in the EU. ALMPs play a central role in the “European Employment Strategy”, following the stimulation of employment being set as key objective of joint economic policy. In fact, attending some form of ALMP program is now a requirement for receiving benefits in most EU countries (European Commission, 2016).

2.3: ALMP effectiveness

This paragraph is divided into two separate parts. First the general effectiveness of ALMPs is discussed, before transitioning to a discussion on the effectiveness of different specific ALMP instruments.

Before anything else, when talking about ALMP effectiveness the first logical question that is in need of answering is how precisely that is to be assessed. According to the European Commission (2016), ALMP effectiveness should first of all be judged based on the number of long-term unemployed, because there are always some number of short-term unemployed entering and exiting the unemployment stock no matter the circumstances. Secondly, the correct timespan is to be considered. Some ALMP measures cannot be expected to be effective immediately, this is especially true with regard to human capital gain focused programs (Card, Kluve & Weber, 2017). Both short and longer time horizons should therefore be considered. Finally, group composition should be taken into account to some degree as well. Age matters

(14)

in regard to ALMP effectiveness especially in regard to youths (Kluve, 2006; Card, Kluve & Weber, 2017) and older individuals (Card, Kluve & Weber, 2017). They are harder to threat with ALMPs and gain less from them. Kluve (2006) found that it is the precise type of ALMP instrument that matters most in regard to effectiveness, target groups and also the business cycle are of secondary concern. That is not to say that these latter two aspects do not matter, but merely that they are secondary to the type of program when it comes to determining effectiveness. Target groups, in terms of controls, will be discussed further in paragraph 2.5. The business cycle and its interaction with ALMP effectiveness will be discussed further in paragraph 2.4.

2.3.1: General

When we turn to the body of knowledge in order to find out what the general and overall effectiveness of ALMPs might be, we come across different studies portraying mixed results. Nonetheless there seems to be an overarching consensus by the majority of prior research into this subject. First, in her review Nordlund (2011) found a range of studies indicating positive effects (Harkman, Jansson & Tamás, 1996; Harkman, 1997; Gerfin & Lechner, 2002; AMS, 2006; Strandh & Nordlund, 2008) as well as negative effects (Ackum, 1991; NUTEK Analys, 1994; Forslund & Krueger, 1995; OECD, 1996; AMS, 1997; Regnér, 1997; Larsson, 2000). Card, Kluve & Weber (2017) also reviewed prior research by performing a meta study in which they analysed over 200 studies dealing with ALMPs. They found that the effects of ALMPs are close to 0 in the short run, but most effects of instruments become more noteworthy after 2/3 years. They found this to be the in particular for ALMPs focussed on human capital accumulation.

The overarching narrative seems to be that as long as some side notes and caveats are taken into account the general and overall effectiveness of ALMPs are positive. Some side-effects do exist as well, i.e. search activity drops during ALMP participation. However, Nordlund (2011) finds the net result on employment to be nonetheless positive.

Kluve (2006) found ‘traditional’ ALMP programs (meaning training) to have a modest but positive impact on employment. Nordlund (2011) also found training to be the most effective ALMP measure generally, in comparison to employment types of ALMP programs. But note

(15)

that training produces relatively long-term effects (Nordlund, 2011). I.e. often training in an ALMP program leads to more regular education. This logically delays the effects of the investment by the ALMP. Martin (2015) also emphasizes training to indeed be effective, especially if specifically tied to labour market needs, but also emphasizes the need to evaluate its effectiveness using a window of several years.

Less clear of an effect are produced by hiring incentives for firms in the market sector. The European Commission (2016) does not find these to be very effective overall. Martin (2015) does find that these can work, but not always. He finds them to be subject often to significant deadweight and displacement effects. Therefore, he insists these programmes can only be effective if they are used sparingly and modestly. Finally, Kluve (2006) notes that private sector incentive programs show a significantly better performance than direct job creation programmes in the public sector, which will be discussed next.

These public sector job creation schemes are found to invariably not work (Martin, 2015). The European Commission (2016) finds the same, namely that the lowest effectiveness of all types of ALMP measures is typically found for direct employment creation in the public sector. Additionally, Card, Kluve & Weber (2017) make clear that public employment programs are ineffective and Kluve (2006) shows that direct job creation in the public sector often produces negative employment effects. Which would lead one to belief, that these types of instruments should merely be used for social or political reasons, not economic reasons. In employment types of ALMP programs human capital is accumulated through networking, ability and motivational stimulation instead of through training. The effectiveness of these programs is more immediate, but the long-term effects are also more doubtful (Nordlund, 2011).

2.4: Business cycle

In the past ALMP expenditures, taken as a whole, tended to be verily uncorrelated with the business cycle (Martin, 2015). Andersen & Svarer (2012) provide us with some reasoning as to why this might be undesirable. First, unemployment causes more inflow into ALMP programmes, which raises both a funding and a capacity concern (Andersen & Svarer, 2012). Quality and/or effectiveness might be at risk because of this. Second, the group composition is likely to change. Because of the larger inflows of participants, it is more likely the group might contain individuals with varying needs. According to Martin (2015), ALMP expenditures have

(16)

become more correlated with the business cycle since 2007.However, expressed in an intensity measure (per participating person), the increase in ALMP expenditures has not in all cases kept up fully with unemployment levels.

2.4.1: Interaction with ALMPs

As has been discussed earlier, different prior research had found varying degrees of ALMP effectiveness. The OECD (1996) has suggested that this varying effectiveness of ALMPs might be ascribed to how well they were coordinated with the business cycle, even though that has historically not been the case very well often. What Forslund, Fredriksson & Vikström (2010) found is that varying ALMP expenditures with the business cycle has now improved in recent years significantly. In fact, they specifically found in these recent years that expenditures on ALMPs are relatively pro-cyclical compared to passive labour market policy expenditures. It might be interesting to counteract this with what Nordlund (2011) found, namely that ALMPs could counteract scarring of the unemployed during their unemployment spells, regardless of what point in the business cycle the economy is at. This of course fuels the notion that ALMP intensity is positively related to employment. This will be further specified coming up. Paragraph 2.4.2 will discuss the general notion on the usage of ALMPs during recessions and booms, while paragraph 2.4.3 will discuss the subject from the perspective of specific measures and their usage during recessions and booms.

2.4.2: Recessions & booms

Increasing numbers of unemployed individuals might lower the overall quality of ALMP programmes (Nordlund, 2011). One view might therefore be to not expand ALMP expenditures during a recession and also because the competition for a limited number of jobs is higher (Card, Kluve & Weber, 2017). However, ALMPs could simultaneously help those individuals who are having difficulties in finding jobs (Nordlund, 2011). Additionally, because employers become more selective during a recession, attending an ALMP program might become, relatively speaking, more valuable to individuals (Card, Kluve & Weber, 2017).

The European Commission (2016) is in favour of expanding ALMP expenditures during a recession. In particular programs with large potential lock‐in effects should be primarily focussed on during such a time. Forslund, Fredriksson & Vikström (2010) and Nordlund (2011) agree with this notion, emphasizing the use of longer programs with larger lock-in effects

(17)

during a recession. At the same time shorter programs with smaller lock-in effects should be emphasized during booms, especially those with direct returns (Forslund, Fredriksson & Vikström, 2010; Nordlund, 2011). These lock-in effects will be discussed more the next paragraph, 2.4.3.

According to Forslund, Fredriksson & Vikström (2010) the primary reason for relying more heavily on some specific measures during booms on the one hand and recessions on the other hand are the different lock-in effects of these measures. If measures with large lock-in effects should ever be used, they should be used in a recession. The reason behind this being simply that opportunity costs are lower in a recession. In their research Andersen & Svarer (2012) find training to be the prime example of such a measure. Contrary, JSA involves a marginal investment in time. Therefore, Forslund, Fredriksson & Vikström (2010) suggest that such an ALMP measure should be used more extensively during a boom. Based on the theoretical notion behind these examples, one should be able to classify other ALMP measures in a similar fashion. The European Commission (2016) provides us with two additional examples. First, they argue employment measures should only be used during recessions. As we’ve learned in paragraph 2.3.2, the effect of ALMP employment is to improve networks, ability and motivation. Because it does not entail employment in the regular labour market, a short return cannot be expected neither (Nordlund, 2011). Redistributive incentives should instead be used relatively more during booms (European Commission, 2016).

2.5: Controls

Some countries with below average ALMP spending have a relatively low-level unemployment rate, while other countries have a relatively high-level unemployment rate despite the fact that they have got an above average spending rate on ALMPs (Martin, 2015). Of course, correlation is not the same as causation and country differences might result in vastly different group compositions. In order to generate a as clean as possible result, it is therefore key that we take certain controls into account when attempting to estimate any kind of an effect in this regard. Luckily the body of knowledge provides quite a clear answer to the question of what other factors should be controlled for.

(18)

Nordlund (2011) provides an extensive overview of the prior research in this field and the corresponding use of control variables. Therefore, we might turn to her overview as a good and solid starting of point and building from there.

§ First, gender which is mentioned by Omarsson (2000) (in: Nordlund, 2011), but also by Forslund, Fredriksson & Vikström (2010) and the European Commission (2016). § Second, age is mentioned by Gregory & Jukes (2001), Rones (1983) (in: Nordlund,

2011), but also by Forslund, Fredriksson & Vikström (2010), Card, Kluve & Weber, 2017) and the European Commission (2016) also.

§ Third, country of birth is mentioned by the Swedish Integration Board (2003) (in: Nordlund, 2011) and by Forslund, Fredriksson & Vikström (2010).

§ Fourth, civic status is mentioned by Saraceno (1997) (in: Nordlund, 2011) and by Forslund, Fredriksson & Vikström (2010).

§ Fifth, having children is mentioned by Strandh & Nordlund (2008) (in: Nordlund, 2011). § Sixth, the level of education enjoyed is mentioned by Åberg (2001) (in: Nordlund,

2011), Forslund, Fredriksson & Vikström (2010) and by the European Commission (2016).

§ Seventh, labour market income is mentioned by Harkman (1999) (in: Nordlund, 2011). § Eight, place of residence (urban/rural) is mentioned by AMS (2006), Omarsson (2000)

(in: Nordlund, 2011) and by Forslund, Fredriksson & Vikström (2010).

§ Ninth, the employment sector is mentioned by Johansson et al. (1999), Lundborg (2001) and by the European Commission (2016).

§ Finally, the European Commission (2016) suggests that the overall state of the economy remains an important factor to be considered. This is of course proxied by the business cycle. Also, the level of passive measures is to be controlled for.

2.6: Hypotheses

Following the insights derived from the body of knowledge and a discussion of them, hypotheses in order to help guide the answer to the research question can now be formulated.

The first hypothesis in order help guide the answer to the research question is formulated as follows:

(19)

‘During recessions ALMPs with relatively large lock-in effects and an emphasis on human capital accumulation are most effective at reducing unemployment.’

Of course, the contrary of what has been stated in the above hypothesis, is hypothesised to be true during a boom.

The second and final hypothesis in order help guide the answer to the research question, and in regard to the secondary objective of this thesis (general effect of ALMP intensity), is formulated as follows:

‘Traditional ALMP instruments, such as training, are most effective at reducing unemployment.’

Next, I will move on to discuss the methodology of this thesis in chapter 3. The hypotheses formulated above will be put through their paces in chapter 4 (results & analysis).

(20)

Chapter 3: Methodology

The aim of this chapter is to present and explain the methodology of this thesis. I opt to do so in three distinct and consecutive parts. First, I will explain the research design of this thesis in general (paragraph 3.1). Different considerations and decisions had to be made in regard to the methodological design of this thesis based on the insights gained from the body of knowledge discussed in chapter 2. The availability of data (and in some cases the limitations that this induced) also played a key role in these considerations and decisions just mentioned. An elaboration on how precisely this is so, is presented and provided throughout the remainder of this chapter. Paragraph 3.2 discusses the different variables being used in this thesis. Specifically, an elaboration on how theoretical concepts are operationalized in this thesis is given. Finally, paragraph 3.3 covers very specifically the method of analysis used in this thesis and brings all of the former together concisely.

3.1: Research design

At its most rudimentary form, this thesis takes the form of a deductive quantitative analysis. It uses both descriptive statistics and regression modelling techniques and is primarily based on an existing microlevel dataset, which is subsequently enriched with data from additional sources at the macrolevel. The additional sources at the macrolevel include the European Commission (2019) and the OECD (2019). Further elaboration on the precise data gathered from these sources will be provided in the next paragraph (3.2). The main microlevel dataset referred to is the European Social Survey (European Social Survey, 2019). This dataset contains extensive data gathered through face-to-face interviews in over thirty countries on a large variety of topics. It covers the time period 2002-2016 and includes all but a few of the EU member states, the EU being the case of this thesis. In essence two reasons for selecting the EU as case for this thesis can be given. The first is in regard to the availability of data (i.e. the European Social Survey) and the second is in regard to external validity. This thesis aims to maximise its external validity through the ability to generalize results to the best of its ability and the extensiveness of the European Social Survey allows for a comprehensive comparison between countries that are relatively comparable (EU member states). The unit of observations in this thesis is there for established as EU citizens in the time period 2002-2016. The European Social Survey roughly totals around 800 respondents on average in each partaking country in each year covered. Again, this relatively large number of observations will also increase the external validity of this thesis through better generalizability. According to Stegmueller (2013),

(21)

there is always a minimum sampling quantity that should be met to avoid possible bias. This level of extensive sampling should minimise possible bias and, in providing a good representation of EU citizens in the time period 2002-2016, enhance the external validity and reliability of this thesis also. The time period 2002-2016 is selected because of the availability of data, as well as because it covers both periods of boom and recession (to be elaborated on later in this chapter). Unfortunately, as mentioned already, the European Social Survey does not include all 28 current EU member states and there for three countries are omitted from this thesis. These are the following countries: Latvia, Malta and Romania.

3.2: Variables

Note that a table containing an overview of all variables used in this thesis is provided below (table 1). The variables are divided into different categories (i.e. dependent, independent, etc), a brief description of each variable is also provided in the table and sources for each of the variables are given in the right-hand column. This chapter will now move on to elaborate on the different variables and will do so by category.

The dependent variable in this thesis is ‘unemployed’. Which is a binomially distributed dummy variable. This variable only covers the active population, i.e. individuals that have different main activities other than working or being unemployed such as education, being retired, etc are not included in the data. As the higher value of this variable indicates unemployment a positive effect of any other variable on this variable should be interpreted as ‘an increase to unemployment’.

The independent variables in this thesis correspond to the European Commission (2006) its classification of ALMP measures into six different categories. Job rotation and job sharing is omitted because of a lack of data, but the other five categories are each operationalized into their own independent variable. The variables are primarily constructed using data from the European Commission (2019), although they are partly constructed using data from the OECD (2019). The design of the intensity measurement is primarily based on the previous research by Martin (2015). Expenditures per category were first converted to constant 2010 prices, they were then converted to Euros (if they were not already) and finally adjusted for purchasing power parity (PPP). This makes the expenditures comparable between countries and over time. The expenditure per category was then divided by the number of participants resulting in the final measurement of ALMP intensity.

(22)

The insights of the body of knowledge provide a list of additional variables that are to be taken into account and controlled for. These variables were already named in chapter 2 and are here operationalized. The descriptions provided in table 1 should provide ample explanation in regard to these variables. Gender, age < 30, age > 55, country of birth, civic status and children are binomially distributed dummy variables. Age, education, labour market income and place of residence are scale variables. Finally, employment sector is a categorical variable and central/local government is used as base category.

On the macrolevel, two additional variables are added. The composite leading indicator (CLI) is a business cycle indicator provided by the OECD (2019). It is an index and 100 represents the long-term trend. Passive labour market policy on the other hand includes out-of-work income maintenance and support as well as early retirement and is constructed similarly to the independent variables.

Finally, year dummies are included as they might control for some time sensitive variability. More importantly however, interaction effects between each of the independent variables and the business cycle indicator (the CLI) are used. These are essentially multiplications of the two variables that they concern. If significant, it can help indicate the variability of the corresponding independent variable over the business cycle, while the independent variable by itself can indicate a general effect on the dependent variable.

(23)

Table 1 – Overview of variables

Name Description Source

Dependent variable

Unemployed Main activity, last 7 days. Dummy variable. ‘0’ = paid work. ‘1’ = unemployed. European Social Survey (2019)

Independent variables

Training ‘Intensity’, expenditure per participant at constant 2010 prices (EUR) adjusted for PPP. European Commission (2019);

OECD (2019)

Employment incentives ‘Intensity’, expenditure per participant at constant 2010 prices (EUR) adjusted for PPP. European Commission (2019); OECD (2019)

Supported employment &

rehabilitation ‘Intensity’, expenditure per participant at constant 2010 prices (EUR) adjusted for PPP.

European Commission (2019); OECD (2019)

Direct job creation ‘Intensity’, expenditure per participant at constant 2010 prices (EUR) adjusted for PPP. European Commission (2019); OECD (2019)

Start-up incentives ‘Intensity’, expenditure per participant at constant 2010 prices (EUR) adjusted for PPP. European Commission (2019); OECD (2019)

Micro control variables

Gender Dummy variable. ‘0’ = female. ‘1’ = male. European Social Survey (2019)

Age Measurement in years. European Social Survey (2019)

Age < 30 Dummy variable. ‘0’ => 30. ‘1’ =< 30. European Social Survey (2019)

Age > 55 Dummy variable. ‘0’ => 30. ‘1’ =< 30. European Social Survey (2019)

Country of birth Dummy variable. ‘0’ = born in current country. ‘1’ = not born in current country. European Social Survey (2019)

Civic status Dummy variable. ‘0’ = not married. ‘1’ = married. European Social Survey (2019)

Children Dummy variable. ‘0’ = does not have children. ‘1’ = has children. European Social Survey (2019)

Education ‘1’ = less than secondary. ‘2’ = lower secondary. ‘3’ = higher secondary. ‘4’ = post-secondary, not

tertiary. ‘5’ = tertiary. European Social Survey (2019)

(24)

Table 1 – Overview of variables (continued)

Name Description Source

Micro control variables (continued)

Place of residence ‘1’ = countryside. ‘2’ = country village. ‘3’ = town/small city. ‘4’ = suburbs/outskirts of

big city. ‘5’ = big city. European Social Survey (2019)

Employment sector Categorical variable. ‘1’ = central/local government. ‘2’ = public sector (other). ‘3’ =

state-owned enterprise. ‘4’ = private firm. ‘5’ = self-employed. ‘6’ = other. European Social Survey (2019) Macro control variables

Composite leading indicator Index. Business cycle indicator. ‘100’ = long-term average. OECD (2019)

Passive labour market policy ‘Intensity’, expenditure per participant at constant 2010 prices (EUR) adjusted for PPP. European Commission (2019); OECD (2019) Other variables

Year dummies Dummy variable for each year in cumulative dataset. 2002 – 2016. ‘0’ = not ‘X’. ‘1’ = ‘X’.

European Commission (2019); European Social Survey (2019); OECD (2019)

Interaction effects Interaction effect variable for each independent variable and composite leading indicator.

Multiplication of each independent variable and composite leading indicator. European Commission (2019); OECD (2019) Note: Whenever OECD (2019) is given as a source the European Union/Europe indicator is used for non-OECD countries.

(25)

3.3: Method of analysis

As was mentioned briefly at the beginning of this chapter, the method of analysis of this thesis consist of two parts. The first being descriptive statistics and the second being regression analysis. Many quantitative analyses include a good bit of descriptive statistics, which in done in order to thoroughly dissect the dataset being used (Healey, 2012). Through this, anomalies, outliers and patterns can be identified. This thesis aims to do so using a combination of graphs and figures, which can be both cross-sectional as well as longitudinal (depending on which is more relevance). Any findings will be explicated on to the best of this thesis its ability. Of course, I should emphasize that the descriptive statistics are merely a stepping stone towards the regression analysis. The regression analysis being the one that ultimately determines the outcome of this thesis.

The second part of the method of analysis of this thesis is regression analysis. Quite a few control variables are being used and before anything else the appropriateness of their inclusion in this thesis its regression equation must be considered. Blindly including all of them would negatively impact the internal validity of this thesis. There would simply be no way of knowing for sure whether any of the control variables being used are actually in line with the insights of the body of knowledge, from which they are derived. To this end individual correlation coefficients (A correlation coefficient being the statistical measure that indicates the strength of the relationship between the relative movements of two variables ranging between 1 for perfect positive correlation and -1 for perfect negative correlation (Bryman, 2012, p. 348).) between the dependent variable ‘unemployed’ and the given control variable are studied in a correlation matrix, after which the appropriateness for the inclusion of each control variable into the main regression equation is evaluated and considered.

The actual main regression equation between the dependent variable ‘unemployed’ and the independent ALMP measures variables will be executed using all control variables deemed suitable for inclusion into the regression equation as mentioned earlier. Based on the dependent variable ‘unemployed’ being a binomially distributed dummy variable in this thesis, applying the ordinary least squares (OLS) mathematical model will in fact result in what is known as a linear probability model (LPM) (Allison, 1999, p. 153). Because standard errors are assumed to be normally distributed in order to calculate significance in OLS models, but in fact are binomially distributed in an LPM, significance of a correlation coefficient cannot be confidently assumed merely based on an LPM (Allison, 1999, pp. 122-123). Therefore, this thesis uses a

(26)

Probit model in conjunction with an LPM, as a Probit model estimates significance more accurately compared to an LPM (Allison, 1999, p. 184). Using multiple statistical modelling techniques can be considered beneficial to this thesis its internal validity. Both of these statistical modelling techniques will be estimated once with robust standard errors and once with clustered standard errors, which again can be considered beneficial to this thesis its internal validity. Robust standard errors target possible heteroscedasticity (Allison, 1999, p. 127) and clustered standard errors adjust for the nature of hierarchal data (individuals within countries) and possible serial correlation (Serricchio, Tsakatika & Quaglia, 2013). However, because every country is different and unobserved heterogeneity cannot be ruled out completely the final statistical model used in this thesis will be a country fixed effects (FE) model to control for time constant unobserved effects. This kind of model allows us to measure the within country variation instead of the overall variation and should, because of the adjustment for unobserved heterogeneity, provide the most accurate results out of all of the different models being used. Its results can be considered leading, while the results of the other models can be considered as additional support of validity if in alignment with this final model. This model is estimated using clustered standard errors, which because of the fixed effects are simultaneously robust. No statistical modelling technique is without its drawbacks. However, by employing multiple specifications we can at least adjust and control for some of their flaws.

The definitive regression equation is determined by which control variables are deemed appropriate for inclusion by the correlation matrix. However, in principle it is constructed based on all of the variables that were listed in table 1. In regard to the interaction effects, it must be noted that the ‘CLI’ variable (business cycle variable) has been cantered around its mean. This way they can be included in the same regression model used to determine the general effects of the independent variables on the dependent variable, without causing distortions. I will first estimate a regression model using a cumulative independent variable, constructed by adding the separate ALMP intensity measures together. This will establish a baseline/reference point to compare the results of the separate estimates of ALMP intensity measures against thereafter.

(27)

Chapter 4: Results & analysis

This chapter reports the results of the statistical analysis performed on the dataset being used by this thesis, as was previously outlined and described in the former chapter on the methodology of this thesis already. This is attempted to be done in the most accurate fashion possible, i.e. meaning unbiased, objective, verifiable, systematic and complete. Keeping that in mind, results are presented in several tables, and figures are occasionally used as well. Each will be thoroughly elaborated on and interpreted as part of this analysis. Do note that actual conclusions will only be drawn in the next chapter (chapter 5, conclusion) following this one, part of which will also be a discussion of the results.

This chapter is structured into two distinct parts. Paragraph 4.1 covers the descriptive statistics, which will ultimately build-up towards paragraph 4.2 and covers the regressions analysis part of this thesis. Paragraph 4.1 is very much structured as a funnel, the table at the beginning of the paragraph proving a general overview and the subsequent figures highlighting more specific elements of that. Paragraph 4.2 first covers the correlation matrix being used to test theoretical assumptions and to evaluate the appropriateness of the inclusion of certain control variables. From this it moves to some additional visualisations using figures (scatterplots) to further our understanding of certain specific relationships. The actual results of the regression analysis are presented in thereafter. Additionally, R2_{values are provided with the results of the regression} analysis to appropriate the goodness of fit of the statistical model used by this thesis. Finally, interaction plots will be used to further analyse and explain the results in regard to the interaction effects.

4.1: Descriptive statistics

Table 2 (descriptive statistics), on the page hereafter, presents a global overview of the complete dataset being used in this thesis. The variables are presented on the x-axis of the table and the countries are presented on the y-axis. From left to right the following variables can be found; the dependent variable (converted to a country/macro level measurement), the independent variables, the micro control variables and finally the macro control variables. The year dummies and interaction effects are not included in table 2. However, do note once again that the dataset covers the period 2002 – 2016. ‘Employment sector’ is also not included in table 2 as it is a categorical variable.

(28)

Table 2 – Descriptive statistics, an overview Un em pl oy m en t r at e Tr ai ni ng Em pl oy m en t i nc en tiv es Suppor te d e & r Di re ct jo b cr ea tio n St ar t-up incent ives Ge nd er Ag e Ag e < 3 0 Ag e > 5 5 Co un tr y of b ir th Ci vi c st at us Ch ild re n Ed uc at io n Pl ac e of r es ide nc e Co m po si te le ad in g in di ca to r Pa ss ive la bour m ar ke t pol ic y AT 5% € 12,476 € 5,227 € 6,252 € 20,579 € 5,307 51% 41.2 19% 11% 9% 49% 61% 3.2 3.1 98.2 € 14,145 BE 8% € 6,463 € 5,313 € 13,614 € 10,126 € 8,400 54% 40.8 19% 10% 12% 53% 69% 3.7 2.8 100.1 € 9,805 BG 11% € 1,040 € 745 N/A € 1,123 € 118 46% 44.9 12% 21% 0% 59% 79% 3.3 3.6 100.4 € 628

HR 11% N/A N/A N/A N/A N/A 51% 40.9 20% 11% 8% 66% 66% 3.5 3.2 99.3 N/A

CY 8% € 5,224 € 6,054 € 3,897 N/A € 5,624 52% 41.2 21% 14% 11% 65% 69% 3.5 3.4 100.2 € 9,238 CZ 6% € 3,182 € 4,981 € 2,984 € 3,008 € 862 53% 42.8 15% 16% 2% 54% 69% 3.3 3.3 99.1 € 1,873 DK 6% € 23,851 € 27,110 € 28,479 € 22,823 N/A 54% 44.5 12% 20% 7% 61% 74% 3.9 3.1 99.9 € 29,104 EE 9% € 4,137 € 1,528 € 1,011 € 652 € 3,289 47% 44 18% 22% 13% 32% 77% 3.8 3.3 98.7 € 3,298 FI 8% € 17,133 € 13,476 € 21,938 € 12,480 € 8,505 53% 43.6 16% 19% 4% 47% 70% 3.7 2.9 100.4 € 12,729 FR 9% € 11,089 € 2,942 € 13,127 € 13,733 € 3,940 50% 41.8 16% 13% 10% 38% 72% 3.5 3.1 99.9 € 12,494 DE 7% € 9,022 € 9,600 € 21,070 € 7,675 € 11,922 57% 43.9 14% 17% 9% 56% 67% 3.7 3.1 99.1 € 10,327 GR 12% € 20,519 € 3,575 € 789 N/A € 8,173 55% 40.2 19% 11% 14% 57% 58% 3.2 3.9 97.2 € 5,580 HU 9% € 1,917 € 986 N/A € 2,823 € 1,893 50% 41.2 18% 12% 2% 53% 68% 3.3 3.2 99.7 € 1,164 IE 11% € 14,397 € 17,759 € 5,515 € 16,652 N/A 55% 41.9 18% 17% 16% 50% 61% 3.5 2.7 99.9 € 10,683 IT 10% € 3,444 € 4,858 € 3,769 € 5,684 N/A 55% 42.3 18% 14% 5% 57% 59% 3 2.7 94.9 € 14,843 LT 12% € 3,529 € 1,945 € 3,408 € 1,863 € 348 43% 44.3 14% 19% 3% 57% 79% 3.8 3.5 100.1 € 1,324 LU 4% € 20,857 € 9,137 N/A € 22,945 N/A 64% 39.2 21% 7% 40% 59% 62% 3 2.9 96.3 € 28,359

(29)

Table 2 – Descriptive statistics, an overview (continued) Un em pl oy m en t r at e Tr ai ni ng Em pl oy m en t i nc en tiv es Suppor te d e & r Di re ct jo b cr ea tio n St ar t-up incent ives Ge nd er Ag e Ag e < 3 0 Ag e > 5 5 Co un tr y of b ir th Ci vi c st at us Ch ild re n Ed uc at io n Pl ac e of r es ide nc e Co m po si te le ad in g in di ca to r Pa ss ive la bour m ar ke t pol ic y NL 5% € 4,633 € 12,585 € 20,789 € 19,377 N/A 53% 43.2 14% 18% 9% 49% 62% 3.5 3.1 100.1 € 15,412 PL 12% € 3,589 € 1,621 € 1,022 € 2,927 € 2,616 56% 39.8 25% 10% 0% 66% 71% 3.3 3.2 100.2 € 1,533 PT 10% € 6,376 € 2,106 € 4,637 € 1,622 € 420 44% 42.1 17% 16% 9% 60% 71% 2.3 3.4 99.7 € 4,837 SK 15% € 1,645 € 984 € 1,961 € 240 € 1,076 51% 41.4 18% 12% 2% 62% 73% 3.3 3.0 99.5 € 2,201 SI 7% € 2,927 € 4,449 N/A € 6,446 € 4,492 52% 40.5 20% 8% 9% 55% 72% 3.5 2.8 100.0 € 6,099 ES 17% € 5,263 € 1,004 € 8,564 € 4,585 € 3,155 57% 40.4 19% 11% 12% 56% 62% 3.0 3.1 99.9 € 9,706 SE 7% € 23,348 € 17,069 € 21,474 N/A € 20,255 53% 43.6 16% 21% 11% 47% 74% 3.6 3.1 99.5 € 11,721 GB 6% € 10,744 € 6,053 € 26,258 € 10,511 N/A 51% 42.5 17% 17% 12% 49% 67% 3.5 3.1 100.0 € 4,225 Min 3% € 1,040 € 745 € 789 € 240 € 118 43% 39.2 12% 7% 0% 32% 58% 2.3 2.7 94.9 € 628 Max 17% € 23,851 € 27,110 € 28,479 € 22,945 € 20,255 64% 44.9 25% 22% 40% 66% 79% 3.9 3.9 100.4 € 29,104 Average 10% € 7,729 € 8,427 € 9,237 € 6,725 € 6,362 53% 42.3 17% 15% 9% 53% 69% 3.5 3.1 99.6 € 6,706

(30)

The reported values are per country averages for the entire time period covered. Minimum, maximum and average values for each variable are also presented at the bottom of the y-axis. Do note that the minimum and maximum values are not the absolute minimum and maximum values, but instead are the minimum and maximum values of the per country averages reported in the main part of the table.

The different variables will now be covered more in depth, any noteworthy observations will be pointed out and described. Meaning primarily that significant outliers will be pointed out and generalization will be attempted to be made whenever possible. Also, when there is little or nothing noteworthy about a given variable this will also be pointed out briefly. The primary focus of this thesis goes out to the dependent and independent variables; therefore, these will be covered relatively more extensively. As mentioned earlier in the introductory part of this chapter, the descriptions of the variables will be aided by figures throughout this paragraph, whenever visualisation are thought to be informative.

First up is the dependent variable in this thesis, the unemployment rate. Figure 1 (below) is a geographical representation of the by country average unemployment rate values, as presented in table 2.

(31)

Tones of red represent low unemployment rates; neutral tones represent average unemployment rates and tones of blue represent high unemployment rates. What stands out above anything else, and becomes abundantly clear from figure 1 especially, is the geographical divide when it comes to unemployment rate between northern and western EU countries on the one hand and southern and eastern EU countries on the other. The former having on average relatively low unemployment rates and the latter having on average relatively high unemployment rates. Additionally, there are no noteworthy outliers in terms of this geographical divide. Luxembourg has the lowest average unemployment rate at 3% and Spain has the highest average unemployment rate at 17%. The average comes in at 10%.

Figure 2 (below) depicts the average overall EU unemployment rate by year (left hand side) and the OECD composite leading indicator (‘CLI’), or business cycle indicator, for Europe is presented next to it (right hand side) for reference purposes (OECD, 2019).

Figure 2 – Unemployment rate, by year / CLI, by year

What is derived from figure 2 is that with in regard to the years covered by our dataset, the average EU unemployment rate peaked twice. The first instance being around 2002 – 2004 and the second instance being around 2009 – 2013. The former peak being a more moderate one and the latter peak being a more extreme one. The average EU unemployment rate was at its lowest at around 2007 – 2008. As figure 2 also shows, these fluctuations in the average EU unemployment rate coincide quite accurately with business cycle fluctuations as depicted by the ‘CLI’. 7 8 9 10 11 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 Unemployment rate (%) 95 96 97 98 99 100 101 102 103 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 CLI

(32)

Next is a discussion of the independent variables in this thesis, the different ALMP intensity measures, the discussion of which will also be related back to the prior discussion of the independent variable (unemployment rate) towards the end.

Figure 3 (below) visualises the difference in EU averages of ALMP intensity by type. To stress again and reiterate, these are intensity measures at the participant level. Therefore, higher and/or lower values do not necessarily represent higher or lower values in total spending. Total spending is not the subject and/or concern of this thesis and will therefore not be covered. Supported employment & rehabilitation has the highest measure of ALMP intensity in terms of EU average levels (€ 9,237), with Greece coming in lowest (€ 789) and Denmark coming in highest (€ 28,479). Intuitively this should not be too surprising as this type of ALMP covers those persons with ‘reduced working capacity’ (European Commission, 2006). Or in other words, juxtapose other persons they require and/or could really benefit from the additional assistance. In contrast, start-up incentives have the lowest EU average level of intensity (€ 6,362), with Bulgaria coming in lowest (€ 118) and Sweden coming in highest (€ 20,255).

Figure 3 – ALMP intensity, by type

Bulgaria has the lowest average level of training intensity at € 1,040, whereas Denmark has the highest average level of training intensity at € 23,851. When it comes to employment incentives Bulgaria again has the lowest average level (€ 745) and also Denmark again has the highest

Training (€

7,729) Employment incentives (€ 8,427)

Supported e & r

(€ 9,237) creation (€ Direct job 6,725)

Start-up incentives (€

(33)

(€27,110). The lowest average value for direct job creation can be found in Slovakia (€ 240) and the highest average value for direct job creation can be found in Luxembourg (€ 22,945).

We can also plot average EU ALMP intensity values by type, as well as over the time period covered by our dataset. Figure 4 (below) shows this. Figure 4 should be viewed keeping in mind that the sample of countries changes somewhat for different years, for different measures. This is due to some missing data for given years, for certain measures and for certain countries. Nonetheless, figure 4 should be able to give a general impression of the way in which the intensity of different ALMP measures developed over the years.

Figure 4 – ALMP intensity, by type & year

What is especially interesting is to compare these graphs to the graph of the development of the unemployment rate, as depicted in figure 2 before. Based on figure 4, what all types of ALMP intensity measures seem to have in common in a downward trend over time. They differ in that this seems to be more linearly the case in regard to direct job creation and start-up incentives, whereas this seems to occur with more shocks in between in regard to employment incentives and supported employment & rehabilitation, with training fitting somewhere in between these

(34)

two categories. An explanation for differences in development over time does not become immediately clear from these figures, but as said what is especially interesting here is the comparison to figure 2. As discussed, the unemployment rate peaked twice over the course of the time period covered by our dataset. This first peak coincides with the early highs of all types of ALMP intensity measures. What followed in the development of the unemployment rate was the low point in the time period covered by our dataset, again coinciding with the overall downward trend of all types of ALMP intensity measures. However, where the similarities largely end are at the second peak of the unemployment rate. It is true that varying degrees of peaks can be observed for the different types of ALMP intensity measures in figure 4, that seem to roughly coincide with the second peak in figure 2, but these never reach the original levels of the first peak. By contrast, the second time the unemployment rate peaked in figure 2 was more extreme then the first. The reason this is important is of course because this is one the main relationships being delft into by this thesis. It is however important to note that the values depicted in both figure 2 and 4 are the EU average values. The question subsequently being whether or not the countries with higher levels of ALMP intensity were better able to retain lower levels of unemployment then others. We will return to that question in paragraph 4.2, in which the regression analysis part of this thesis is covered.

That leaves us with the description of the control variables used in this thesis, as mentioned before these will be described relatively briefer juxtapose the dependent and independent variables discussed before. First up are the micro control variables. Both genders seem to be equally well represented, males making up 53% of the dataset. Most countries seem to follow this distribution quite well. The same holds true for age, with the average age in this dataset coming in at 42 years old. Again, the country averages only diverge a couple of years at most from this global average. About 17% of individuals fit into the younger than 30 years old category and about 15% fit into the older than 55 years old category. Poland has the most individuals in the younger than 30 years old category (25%) and Estonia has the most individuals in the older than 55 years old category (22%). On average, 9% of individuals were born in a different country. Quite a few outliers here, this percentage in lower than 1% in Bulgaria and Poland and it is 40% in Luxembourg. In regard to having a partner (average of 53%), having children (average of 69%), level of education (average of 3.5 on our 5-point scale) and the degree of the place of residence being in a rural one (average of 3.1 on our 5-point scale); there are very few real noteworthy outliers in terms of country averages juxtapose the

(35)

Finally, the macro control variables used in this thesis. ‘100 represent’ the long-term OECD trend of their composite leading indicator. The countries in our dataset come in just below that at 99.6. In terms of the individual country averages in our dataset, their deviations are very minimal from this EU average – which is probably an illustration of how far EU integration has come. In regard to passive LMP, the average intensity comes in at € 6,706. It is highest in Denmark at € 29,104 and lowest in Bulgaria at € 628. From this we may derive that Bulgaria and Denmark represent the two most extremes, the most frequently amount of times in regard to active and passive LMP.

4.2: Regression analysis

We will now move on to the regression analysis paragraph of the results and analysis chapter. To reiterate some that has already been mentioned, this paragraph is made up out of several consecutive subparts. First, we will briefly take a closer look at a correlation matrix between all the variables used in this thesis and examine its results. We will then briefly provide additional elaboration on the matrix by ways of taking a look at some scatter plots between the dependent and independent variables used in this thesis. After that the tables containing the main results of the regressions are presented and discussed, focusing mainly on the general effects of the ALMP measures at first. Although, the results pertaining the control as well as interaction effects variables are also presented in these tables. We first briefly look at the overarching measure of ALMP intensity. This will provide us with a baseline and a reference point to compare separate measures of ALMP intensity against, which we will do subsequently thereafter when we move on to the separate measures for the different types of ALMPs. Finally, we will look at the interaction effects of the different types of ALMPs and the business cycle, using both the results obtained from the main regression tables as well as by plotting these into interaction plots to provide additional inside and elaboration.

The existing body of knowledge identified a wide variety of control variables we ought to be taking into account when making our analysis. We presented these first in chapter 2 (theory), before subsequently further discussing their inclusion in this thesis in chapter 3 (methodology). As a first step in the regression analysis I opted to compile a correlation matrix between all these variables. The results of which are presented in table 3 (next page). Do note that only the first column of the correlation matrix is presented here, as it is the only part of the matrix of