The labour market outcomes of vocational education in Europe: evidence from the LFS survey

Transcription

1 The labour market outcomes of vocational education in Europe: evidence from the LFS survey Emilio Colombo and Luca Stanca University of Milano Bicocca This version: September 2012 Very preliminary and incomplete do not quote Department of Economics, University of Milan Bicocca, Piazza dell'ateneo Nuovo 1 (U6-367), Milan, Italy. emilio.colombo@unimib.it Corresponding author. Department of Economics, University of Milan Bicocca, Piazza dell'ateneo Nuovo 1 (U6-367), Milan, Italy. luca.stanca@unimib.it 1

2 1. Introduction One of the key policy issues in advanced economies is the ease with which young workers are able to make the transition from school to work. The youth unemployment rate is one side of the problem, entailing substantial social and economic costs for the entire society. Other issues are, however, equally important. Even if young workers do find a job after finishing school, this job may not have the characteristics they require, thus resulting in a mismatch that, in turn, can lead individuals to change several employers and occupations. Several approaches have been pursued by policymakers to smooth the transition from school to work and improve the match between the demand and supply sides of the labour market. One approach aims at increasing the education level of the labour force, under the assumption that unemployment and mismatches are inversely related to the level of education (i.e., the higher the general level of education, the lower the mismatches and the general unemployment rate). An alternative approach focuses more specifically on the orientation of educational programs, trying to link more effectively students with jobs through VET programs and apprenticeships with firms (e.g. Ryan, 2001). Until recently, data constraints have generally led the literature to neglect the second approach, while resulting in a focus on specific groups of countries, thus limiting the scope of crosscountry comparability. This paper is one of the first studies to analyse the transition from education to work of different education orientations for a large group of European countries in a comparative perspective. Such an approach has been made possible by the availability of the ad hoc module of the 2009 LFS survey, which allows comparing the labour market outcomes of general and vocational educational programs. The empirical analysis is structured in two parts. The first part aims at characterizing the relationship between education type and labour market outcomes in Europe overall. The second part aims at investigating how this relationship differs across countries and, more specifically, whether cross-country differences can be explained by national labour market conditions and institutions. The definition of labour market outcome adopted in this study is wide, encompassing several aspects of the labour market, such as the transition from education to work, the employment status, the characteristics of the current job, and the income level. All these features provide a comprehensive characterization of the returns to vocational education throughout Europe. The paper is organized as follows. Section 2 reviews the literature. Section 3 provides a description of the data and methods used in the empirical analysis. Section 4 presents the results. Section 5 concludes. 2

3 2. Related Literature This paper is related to two bodies of literature. The first comprises the large number of studies that estimate the returns to different education levels. The second comprises studies that aim at comparing labour market outcomes across education programs. 1 The first body of literature comprises a very large number of studies that aim at estimating the returns to education at both individual and social level. The findings indicate that, at the individual level, higher education levels carry a premium both in terms of improved employment opportunities and in terms of higher earnings. 2 These studies generally refer to individual countries, due to the lack of comparative data sets. There exist relatively few exceptions with cross-country analyses, such as Trostel et al. (2001). In a long-term perspective, and extending the analysis to the social returns to education, it is well known that improved education and schooling plays a key role for economic growth. As documented by Hanushek and Woessmann (2008), there is strong evidence that the cognitive skills of the population rather than mere school attainment are strongly related to individual earnings, the distribution of income, and economic growth. With reference to the EU, Hanushek and Woessmann (2012) estimate the potential economic impact of the achievement of the targets set out in the Lisbon 2020 strategy (measured by international student achievement tests). They show that the present value of the gains from educational reforms for the European Union on aggregate would be very large: for example, reaching the official EU benchmark of less than 15% low-achievers in basic skills by 2020 would generate gains of 211% of current GDP. Countries differ substantially not only in the level, but also in the organisation of schooling and education structures. Some authors 3 emphasize the importance of vocational education in developing specific job-related skills that can facilitate employment in specific occupations. Other authors 4 emphasize the importance of basic knowledge (mathematics, communication, literature etc.) through general education programs, under the assumption that specific skills may become obsolete quickly and that employability is therefore maximized by strengthening the foundations of basic knowledge. The first approach is followed by several European countries. The main example is the German dual system, that provides extensive vocational education and training at the secondary level, often involving firms directly in the development of programs through apprenticeships. The second approach has been followed recently by the United States, who has strongly reduced if not completely eliminated VET as a separate track in secondary education, while strengthening general education. 1 Due to the nature of the data set, this research focuses on individual returns to educational choices. It therefore neglects the returns at firm level which are the subject of a vast literature that concentrates particularly on the effect of workplace training on firm performance (see among others Colombo and Stanca (2008), Bassanini et. al. (2005), Dearden et al (2006)). 2 See e.g. Hanusheck et al. (2011), Dearden et al. (2002), McIntosch (2006), Blundell et al. (2005), Card (1999), Harmon and Walker (1995, 1999), among others. 3 Wolter and Ryan (2011), Hanushek and Woessmann (2008), Ryan (2012). 4 Hanushek and Woessmann (2012), Krueger and Kumar (2004a, 2004b). 3

4 Following this distinction, a second body of literature has developed in recent years, aiming at estimating the economic effects of different patterns of education, focusing in particular on the comparison between general education and VET. Overall, data limitations constrain crosscountry comparison and limit the analysis to national case studies (e.g. Dearden et al., 2002, Bonnal et al., 2002, Euwals and Winkelmann, 2002, Neuman and Ziderman, 2002, Ryan, 2002a,b, Hofer and Lietz, 2004, Bishop and Mane, 2005, McIntosh, 2006, Karmel and Nguyen, 2006, Jenkins et al., 2007, Meer, 2007, Lee and Coelli, 2010, Herault et al., 2011). In an influential paper, Ryan (2001) summarises the evidence across countries, emphasizing that vocational programs, and in particular apprenticeships, appear to increase the employment content of early working life and to carry a modest income premium. Using a natural experiment to correct for possible selection bias, Malamud and Pop-Eleches (2010) find that in Romania VET graduates are characterized by quite different occupational outcomes, namely are significantly more likely to be employed as manual workers and craftsmen. However there is no statistical difference between VET and general education in terms of more general outcomes such as participation rates, unemployment rates, periods of non employment and of family income. Using a similar approach, Fersterer et al. (2008) find no significant wage premium to apprenticeship in. This leads the authors to conclude that: apprenticeship training does not seem to be superior to other forms of school-based education, say, in colleges or vocational schools. Similar results are obtained by Wolter and Ryan (2011) and by Oosterbeek and Webbink (2007). Dearden et al (2002) find a significant premium for academic qualification over vocational courses in Britain, even if the gap is reduced when controlling for the amount of time spent to acquire different qualifications. Bonnal et al., (2002) find that in apprenticeship appears to have a significant premium (mainly short term) over alternative forms of VET. In the US, Bishop and Mane (2005) find that students who take a certain percentage of vocational subjects are more likely to earn higher wages and display higher participation rates with respect to purely academic students. These higher returns and payoffs are evident both in the short and in the long run. In a different setting Neuman et al. (2002) show that for individuals from important minorities and disadvantaged groups VET could be effective in VET in raising wages (although this effect varies strongly among groups) In addition to data constraints, one of the main limitations of the studies mentioned above is that, by focusing on the transition from education to work, they adopt a short-run perspective, while neglecting long-run effects. However, as stressed by Hanusheck et.al. (2011), there is likely to be a trade-off between short and long term costs and benefits of VET systems. On the one hand, vocational education may generate skills that facilitate the transition into the labour market. On the other hand, those same skills, being specifically job related, may become obsolete at a faster rate, entailing a long-term loss. Moreover, it is well known that individuals engaged in general programs are more likely to continue studying, relative to those following vocational studies, thus improving their long-run outcomes. The authors find support for this hypothesis, showing that the initial labour-market advantage of vocational education, relative to general education, tends to decrease with age. The distinction between education pattern could have profound long term implications affecting long run growth. In fact as argued by Krueger and Kumar (2004a, 2004b) vocational (i.e. skillbased ) as opposed to general (i.e. concept-based ) education could lead to slower adoption of 4

5 new technologies. The reason is that vocation education tends to be specialised and therefore favours learning within a current technological framework. On the opposite general education, while more costly to be acquired, allows a faster and better adaptation to new technologies that may complement or substitute the current one. Using this argument the authors explain the observed growth differentials between Europe and the United States during the last decades. European education policies, which generally favoured specialized, vocational education, were relatively successful during the 1960s and 1970s when the technological frontier was changing slowly. In fact, during these decades, Europe catched up with the US. However since the eighties when the rate of change of technology became faster the growth gap between US and Europe widened favouring the higher propensity in the United States to use general versus vocational education. Within the analysis of different orientation of education programs a particular emphasis has been devoted to apprenticeship. This is a type of vocation education which combines formal education with training at the workplace 5. The interest devoted by the literature to this form of vocational education is that by being characterised by elements of training conducted on the job and within the firm it provides a natural setting where it is possible to investigate the effect of different work organisation, firms characteristics and also industrial relations on the effect of educational programs in the labour market. In this framework Dustmann and Schonberg (2009) find that unionization increases participation in training and that a non-unionized firms are less likely to finance training of workers.6 Focussing on Kriechner et al (2012), using data with information on the costs and benefits of apprenticeship training at firm level, find that firms with works councils make a significantly higher net investment in training compared with firms without such an institution. Similarly, using data from and Switzerland Muehlemann (2010) find that in the presence of strong employment protection laws firms are more likely to invest more in apprenticeship training. With respect to the literature briefly surveyed in this paragraph this research improves over several dimensions. First it is the first study that undertakes a large cross-country comparison of labour market outcomes of different education pattern and levels. This allows to introduce in the analysis institutional features as possible explanations of differences of labour market outcomes across countries as done in section 5. Second the richness of the information contained in the LFS survey and in the ad hoc module allows to investigate several dimensions of the transition from education to work distinguishing between employment status, the first from the current job, the type the characteristics of the job, the associated income level. Notwithstanding these improvements our analysis suffers from some constraints: the first is that the ad hoc module of the 2009 LFS survey is focused on years old, not allowing to estimate the long run labour market outcomes of educational choices. The second constraint is that detailed information on education orientation are available only on a cross sectional basis limiting the possibility of controlling for the endogeneity of educational choices to labour market outcomes (see the next section for a more thorough analysis of this issue). 5 See Ryan (2012) and Wolters and Ryan (2011) for two recent surveys. 6 Dustmann et al. (2009) conclude from these findings that the recent process of de-unionization could be responsible for the increase in skilled-unskilled wage differential. 5

6 3. Data and methods The data set used for this study is the individual-level data of the 2009 Labour Force Survey, available for both the core and the Ad Hoc module, which has been designed specifically to provide information about young individual s transition from education to work. Eurostat implements some corrections and changes to the dataset in order to harmonise them and to guarantee anonymity. Overall, the data set provides detailed information on individual characteristics, socio-demographic background, level and pattern of education, and labour market outcomes. Table 1 reports descriptive statistics for the main variables used in analysis. Table 1. Descriptive statistics Variable Mean Std. Dev. Min. Max. N Secondary -GED Tertiary -GED Secondary -VET premium Secondary -GED Tertiary -VET premium Tertiary -GED VET34 -work based premium VET34 -school VET34 -no distinction First job, search duration First job, duration First job, full-time First job, permanent First job, nding method First job, type of occupation Employed Current job, full-time Current job, permanent Current job, self-employed Time in current employment Current employment, search duration Current job, long search duration Duration of unemployment Male dummy Age Age Age Age Separated Single Married Foreigner Parents with degree Children in the household (dummy) Number of Children in the household Source: LFS. 6

7 3.1 Defining VET Vocational education and training (VET) identifies educational programs designed to enable the participants to develop the practical skills, know-how and understanding needed to find employment in a particular occupation or trade or class of occupations or trades. 7 VET often involves a combination of workplace training with traditional formal school activities. as such VET differs from other more traditional types of education, which in this study are classified as "general education". Within Europe, national systems differ greatly in their practical definition of VET or non-vet programmes. 8 In the 2009 Labour Force Survey AHM, VET and general education refer to the orientation of the highest level of education attained (variable HATVOC) and the distinction is applied only to ISCED levels 2-4. More specifically 9 : General education: less than 25% of the programme content is vocational. Vocational (and pre-vocational) education and training: at least 25% of the programme content is vocational (oriented towards a specific category of occupations or trades and leading to a labour-market relevant qualification). Furthermore, in countries where it is possible to draw such distinctions 10, vocational education and training is divided into: mainly school-based vocational education and training: where at least 75% of the vocational education/training hours is spent in a school, college or training centre and the balance in a work environment (enterprise or other). mainly workplace-based vocational education and training: where at least 75% of the vocational education/training hours is spent in a working environment (enterprise or other) and the balance in a school, college or training centre. Combination of school and workplace-based vocational education and training (e.g. dual system, alternate programmes): where less than 75% of the vocational education/training hours is spent in a school, college or a training centre with the rest carried out in a work environment (enterprise or other). The 2009 LFS AHM applies the question on education orientation to solely those respondents who have attained a secondary or post-secondary non-tertiary level of education. However, in this paper, we will also distinguish between theoretical/generic programmes (ISCED 5A) and technical/occupation-oriented programmes (ISCED 5B) in tertiary education. Notwithstanding the methodological weaknesses, especially in terms of international comparability, ISCED 5B 7 See the International Standard Classification of Education, ISCED Detailed information can be found in the Country Reports published as part of Cedefop s VET in Europe initiative at 9 The categories proposed conform to the ISCED definitions applied in the UNESCO-OECD-Eurostat data collection and the discontinued Eurostat VET (Vocational Education and Training) data collection as agreed by Member States (EUROSTAT/E3/2000/VET02, includes a definition for VET). 10 Estonia, Italy, Netherlands, Norway,, Portugal and the Slovak Republic, did not report the distinction of different VET types. The following tables group these countries under the separate category "no distinction possible" for purposes of comparison. 7

8 could be considered a proxy for tertiary level VET until further improvements have been incorporated into the revised ISCED Finally although the AHM distinguishes between VET and general education from the level of ISCED 2 and 3c short, these two educational levels are excluded from the analysis; education at this level focuses mainly on general programmes (88.4%) given that many countries do not include VET programmes at ISCED level Including ISCED 2 would create a bias in the results but not add much in terms of explanatory power. Therefore, in this paper VET will be restricted to upper secondary and post-secondary non-tertiary education (ISCED 3 and 4). 3.2 Defining Apprenticeship VET defines apprenticeship as systematic, longer term training in which the individual alternates his/her training hours between the workplace and the educational institution or training centre. The apprentice is contractually linked to the employer and is remunerated (wage or allowance). The employer is responsible for providing the trainee with specific occupational training to equip them with the skills needed to carry out a specific occupation 13. In Europe strong differences across countries in the definition of apprenticeship persist. Some countries, such as, and Switzerland, incorporate apprenticeship into the dual educational system as a VET system building block. Other countries (e.g., Italy) adopt a simple contractual form between the firm and the apprentice that has little to do with the education system. Given the information available from the LFS survey in this study we approximate apprenticeship with two categories of the variable HATVOC: "Combination of school and workplace" and "VET mainly workplace"; therefore the paper excludes from the analysis those countries that adopt some apprenticeship form but for which the EU LFS makes no distinction by type of VET. Despite providing a unique opportunity for comparing labour market outcomes of education across countries, the data set suffers from some important limitations. The first is that the data are available on a cross-sectional basis, thus not allowing the use of panel estimation methods. This limits the possibility of controlling for the endogeneity of educational choices (see below). The second limitation is that the 2009 AHM of the labour force survey is focused on young adults, i.e years old, and therefore does not allow estimating the long-term effects of education choices. Additional data limitations refer to specific countries and variables, as detailed below: No breakdown by orientation is available for Norway, which has been excluded from the analysis. and Switzerland reported difficulties in extracting precise and reliable information from the variable Jobstart (year or month of start of the first job held for more than three months after quitting formal education). In turn, Jobstart is used to construct some derived variables related to school-to-work transition. and Switzerland are therefore excluded from the analysis when these variables are used. 11 A difficulty arises also from the fact that Eurostat uses the ISCED classification instead of the European Qualification Framework. The latter, requiring a broader discussion about learning outcomes is more useful for analysing differences across tertiary programmes (see Cedefop 2011 for an analysis on the identification of VET at tertiary level). 12, Spain, Italy, Cyprus, Hungary,,, Romania,, Slovakia, Finland, Norway and Switzerland provided solely data on ISCED 2 - General orientation. 13 Terminology of European education and training policy. Cedefop,

9 Income data is not available for the following countries:, Ireland, Malta, Netherlands, Norway, Sweden, and Iceland. The distinction between different types of VET (e.g. school-based, work-placed based) is not available for the following countries: Estonia, Italy, Netherlands,, Portugal, Slovakia. These countries have therefore been excluded when VET types are used in the analysis. In addition to the restrictions above, we also implemented two additional restrictions: first we excluded from the analysis the age group 15-19, second we excluded students. Both choices are due to the fact that we want to focus on the transition from education to work and on the labour market outcomes of different education choices, therefore we exclude individuals for whom it is not possible to clearly separate education from working life. 14 Estimating the return to education is difficult because the analysis is generally plagued by problems of self-selection and omitted variables, which make the education choice endogenous. More able individuals are in fact more likely to have better labour market outcomes (both in terms of wages and employment status), and are also more likely to study longer than less able ones. This makes it difficult to disentangle the effect of education from that of unobservable ability. A similar issue arises also in comparing VET with general education programs. Students with stronger practical and manual abilities are more likely to enrol in VET-oriented education programmes (and benefit more from them) than in more academically oriented programmes (Meer, 2007). Several approaches have been proposed to solve the endogeneity problem, ranging from Instrumental Variable estimation to control function and matching techniques. In the absence of an experimental or quasi experimental setting 15 the endogeneity of educational choices is generally addressed by exploiting the time series dimension of the data set: as shown by Arellano and Bond (1991) and by Blundell and Bond (1998) in fact past lags of the level and/or difference of the variables that are suspected to be endogenous could be used as valid instruments in dynamic panel data models. Unfortunately the availability of a single cross section (year 2009) of our data does not allow us the use of these approaches. However the availability of detailed information of individual characteristics and on family background available in the LFS microdata allows to significantly attenuate the possible biases associated with the failure to properly account for individual ability The results are however robust to the inclusion of these categories in the regression. 15 This is for instance the approach followed by Malamud and Pop-Eleches (2010) who exploit the 1973 reform of the Romanian educational system that required students who entered vocational schools to receive an additional two years of general education. Therefore, following this reform, secondary school cohorts born after 1959, were treated with more general education and less vocational training than earlier cohorts. 16 The literature has also assessed the accuracy of cross section estimators and the direction and size of the biases associated with it. Overall, it emerges that failure to control for ability and family background characteristics that affect education choices tend to bias OLS estimators upwards. On the other hand, measurement error and sample composition bias tend to produce a downward bias. Dearden (1999b) and Dearden et al. (2002) show that, for most of the qualifications considered in their analysis, these two opposite sources of biases roughly offset each other, so that, overall, OLS tends to provide reasonable estimates of the returns to education. These results are important for our purposes because they downplay the possible limitations of the data set used in the present research for controlling the endogeneity of education choices. 9

10 (1) where the subscript I identifies the individual, j the country, the variable OUT identifies the labour market outcome which will be defined by several different variable explained below. The terms VET34, GED34, VET56, GED56 are the ones we are most interested in; these are dummy identifying respectively whether an individual has a secondary education of VET type, secondary education of GED type, tertiary education of VET type and tertiary education of GED type. The value associated to each of those coefficients measures the premium (in terms of outcome) for reaching a secondary (tertiary) level of education of VET (GED) with respect to having a level of education up to secondary. This specification allows to disentangle two effects of education choice: the first effect derives from the level of education captured by the size of the coefficients secondary (VET/GED) and tertiary (VET/GED) with respect to the reference group ( Not having completed secondary education ). The second effect derives from the orientation of the education and is captured by the difference between VET and GED coefficients for each level of education. The tables present both the value of the coefficients to and the value of the difference and with the associated statistical significance. The matrix X denotes individual controls (age sex marital status, parents education, family characteristics) and D is a full set of country dummies. The analysis has been conducted for the overall sample of individuals, and by splitting the sample by gender and by age group. The estimation method for equation (1) depends on the nature of the variable. In case of a dichotomous dependent variable (job characteristics, job mismatch) logit models are used, in case of multiple non ordered outcomes (employment status, job search methods, job type) a multinomial logit, in case of multiple ordered outcomes (income deciles) an ordered logit or a poisson (job duration) model are used. 4. The labour market outcomes of VET In this section, we provide a comprehensive analysis of the labour market outcomes of education level and orientation at aggregate level (EU-wide). In Section 4.1, we examine the transition from education to work (first significant job), with the aim of assessing the initial steps of young adults when leaving formal education. In section 4.2, we examine longer-term outcomes, focusing on the characteristics of workers current employment. The results reported are the parameter estimates for the effects of education levels and orientations on the relevant labour market outcome. The full specification and the estimators use for different dependent variables are discussed in the technical appendix Transition to work Advocates of vocational education emphasize its merits in speeding up the transition process from education to work by providing early and effective matches for secondary education graduates. However, particularly for the case of apprenticeship, this process could ultimately result in inefficient outcomes, to the extent that it ties young people to a particular employer, thus hindering workers mobility (Heckman et al. 1994; Acemoglu and Pischke, 1998). 10

11 The AHM module of the 2009 LFS survey contains several variables that provide valuable information on the transition process from school to work. In particular, this paper analyses the characteristics of the first job, 17 focusing search duration, job duration (tenure), type of contract (full-time vs part-time, temporary vs permanent, etc.) and type of occupation. We start by considering the effects of secondary education on search duration (in months) for the first job. OLS estimates are reported in Figure 1. Ceteris paribus, the transition to the first job is relatively faster for VET graduates than for general education graduates. Relative to individuals with no secondary education, search duration is 1.56 months shorter for a VET secondary degree and 1.28 months shorter for a General Education secondary degree. The same pattern applies irrespective of age group and gender. Compared to graduates of lower secondary education, male VET graduates find a job 1.51 months faster and female VET graduates 1.63 months faster on average. The return to VET is also increasing with age (1.07, 1.61 and 1.83 months for age-groups 20-24, 25-29, and 30-34, respectively). Figure 1. First job, search duration VET GED Table 2 presents estimates of the VET premium (the difference between the returns to VET and General Education degrees, both secondary and tertiary), overall and by individual gender and age groups. Relative to general education, the search duration VET premium for secondary education is statistically significant, both overall and by sub-sample (with the only exception of the age group). Overall, these findings highlight one of the main benefits of VET programs: being oriented to specific jobs, VET secondary education results in a faster transition to work. On the other hand, job search duration is significantly longer for VET tertiary education graduates. 17 More precisely, the first job lasting more than 3 months. 11

12 Table 2. VET premium, job search duration Secondary - VET premium -0.27** -0.29** -0.26** -0.14** ** (-5.23) (-3.98) (-3.51) (-2.58) (-1.88) (-3.51) Secondary - GED -1.28** -1.22** -1.37** -0.93** -1.47** -1.47** (-19.20) (-13.68) (-13.44) (-11.98) (-14.14) (-11.41) Tertiary - VET premium 0.50** 0.56** 0.43** 0.27** 0.48** 0.55** (8.91) (7.08) (5.35) (3.98) (7.32) (5.31) Tertiary - GED -2.40** -2.29** -2.51** -1.42** -2.37** -2.77** (-41.01) (-29.48) (-27.74) (-20.13) (-26.68) (-26.93) R N Notes: Dependent variable: first job search duration, in months, as described in column heading. OLS estimates, t-statistics reported in brackets, heteroskedasticity-robust standard errors. The set of regressors also includes country dummy variables and individual-level controls. * p<0.05, ** p<0.01. Next, we turn to the duration of the first job (in months). OLS estimates are displayed in Figure 2. Overall, job duration is relatively shorter for VET (0.43) than for general education (0.72), although the difference is relatively small. The effect of secondary education on job tenure is large and positive for females, while small and negative for males. It is also positively related to age for general education, while the relationship with age is less clear-cut for VET. Figure 2. First job duration VET GED As shown in Table 3, the VET premium for job tenure is negative but not statistically significant overall. The pattern observed for secondary education is reversed for tertiary education. The duration of the first job is significantly higher for graduates from VET tertiary education programs. A possible explanation for this pattern might be the different orientation of education programs. Tertiary VET programs tend to develop job-specific skills, which could result in 12

13 longer search of the correct match (Table 1). 18 However, once the match is found, the probability of separation is lower. On the contrary, tertiary general education programs tend to develop more generic skills that increase the probability of finding a match in the labour market but can result in a sub-optimal match followed by higher mobility. Table 3. VET premium, first job duration Secondary - VET premium ** * (-0.45) (-0.35) (-0.06) (2.66) (-0.18) (-2.06) Secondary - GED * * (0.99) (-0.03) (1.57) (-2.28) (0.69) (2.20) Tertiary - VET premium 3.48** 3.99** 3.36* 7.62** 4.36** 2.12 (3.62) (2.80) (2.56) (3.98) (2.65) (1.51) Tertiary - GED -3.60** -5.84** ** -4.54** (-4.85) (-5.82) (-1.42) (-7.03) (-3.28) (-1.44) R N Notes: Dependent variable: duration of first job, in months, as described in column heading. OLS estimates, z-statistics reported in brackets, heteroskedasticity-robust standard errors. The set of regressors also includes country dummy variables and individual-level controls. * p<0.05, ** p<0.01. Turning to the characteristics of the first job, logit estimation results indicate that VET graduates are relatively more likely to find a full-time first job than general education graduates (Figure 3). The full-time returns to secondary education are negatively related to age. The VET premium is more pronounced for males (0.41) than for females (0.07). The disaggregation by age-group indicates that the full-time VET premium is largest for the group (0.32). 18 Differently from secondary programs, tertiary programs do not have features, such as apprenticeship, facilitating the transition to work. 13

14 Figure 3. First job, full-time VET 0.45 GED As shown in Table 4, the VET full-time premium is statistically significant overall, for male workers, and for workers aged 25 to 29. General secondary education has a non-significant fulltime return. Tertiary education has a large and significant full-time return, but the difference between VET and general education is not statistically significant. Overall, these findings indicate that, relative to general education, vocational secondary education has a positive, albeit small, relative effect on the probability of finding a full-time first job. Table 4. VET premium, full-time first job Secondary - VET premium 0.16* 0.41** ** 0.01 (2.47) (3.28) (0.83) (0.31) (2.61) (0.05) Secondary - GED ** (1.20) (-0.71) (1.80) (2.94) (0.26) (0.53) Tertiary - VET premium * (1.22) (2.03) (0.44) (0.99) (-0.13) (1.37) Tertiary - GED 0.32** ** 0.48** 0.39** 0.2 (4.40) (-1.22) (5.96) (2.81) (3.03) (1.82) R N Notes: Dependent variable: first job, full-time vs part-time. Logit estimates, z-statistics reported in brackets, heteroskedasticity-robust standard errors. The set of regressors also includes country dummy variables and individual-level controls. * p<0.05, ** p<0.01. Logit estimation results also indicate that, ceteris paribus, individuals with a secondary education degree are relatively more likely to find a permanent first job than individuals with no secondary education (Figure 4). The return to VET secondary education is higher than for general secondary education and, as for full-time jobs, the premium is more pronounced for males (0.11) than for females (0.03). The disaggregation by age-group indicates that the VET permanent-job premium is negative for younger workers (20-24), while positive and relatively large for the group (0.15). 14

15 Figure 4. First job, permanent 0.23 VET GED Point estimates reported in Table 5 indicate that the return to secondary education and the corresponding VET premium, however, are not statistically significant, irrespective of the subsample considered. Tertiary education carries a positive and significant permanent-job return, without any significant difference between VET and general education. Overall, these results indicate that vocational secondary education plays a small role in increasing the probability of finding a permanent first job. Table 5. VET premium, permanent first job Secondary - VET premium (1.82) (1.27) (0.38) (0.95) (1.47) (0.03) Secondary - GED (1.84) (-0.12) (1.24) (-1.45) (0.55) (1.25) Tertiary - VET premium (1.23) (-0.11) (0.20) (1.03) (-0.32) (0.40) Tertiary - GED 0.16* 0.24** * ** (2.47) (2.67) (1.22) (-2.08) (1.14) (3.68) R N Notes: Dependent variable: first job, permanent vs temporary. Logit estimates, z-statistics reported in brackets, heteroskedasticity-robust standard errors. The set of regressors also includes country dummy variables and individual-level controls. * p<0.05, ** p<0.01. By exploiting the information about the occupational level it is possible to analyze the likelihood that individuals with a given education level and orientation perform a job of a certain skill profile. Occupations at one-digit ISCO levels have been grouped into four skill groups: high skill, skilled manual, skilled non-manual and low-skill. A multinomial logit analysis has been performed using low-skill as reference group. Figures 5 to 7 display the 15

16 results. Skilled manual jobs are more likely to be performed by VET secondary graduates, irrespective of gender and age. On the contrary, high-skill first jobs are more likely to be found by general education secondary graduates, irrespective of gender and age, although the difference is less pronounced. No clear-cut differences between education orientations can be observed for skilled non-manual jobs. Figure 5. First job, occupation: skilled manual VET GED Figure 6. First job, occupation: skilled non-manual VET GED

17 Figure 7. First job, occupation: high skill VET GED The estimates reported in Table 6 indicate that the skill returns to education are large and significant. The VET premium, for both secondary and tertiary education, is large and significant for skilled manual jobs, while it is negative and significant for high-skill jobs. Highskill jobs are significantly more likely to be performed by individuals with a general level of education (both among tertiary and secondary education holders). Skilled manual jobs are significantly more likely to be performed by VET graduates. 17

18 Table 6. VET premium, first job occupation level Skilled manual Secondary -VET premium 0.88** 0.87** 0.41** 0.86** 0.70** 0.81** (9.95) (8.24) (2.66) (4.52) (4.75) (6.24) Secondary -GED (0.41) (-0.28) (-0.44) (-0.86) (-0.61) (0.47) Tertiary -VET premium 1.07** 0.85** 1.09** 1.39** 0.97** 0.90** (6.56) (4.31) (3.71) (2.96) (3.41) (4.15) Tertiary -GED (-1.19) (-1.37) (-1.33) (-1.72) (-1.54) (-0.31) Skilled non manual Secondary -VET premium ** (-0.40) (-3.03) (-0.19) (-1.84) (0.16) (-1.44) Secondary -GED 1.02** 1.10** 0.92** 0.70** 0.89** 1.23** (12.98) (9.87) (8.76) (4.51) (6.71) (10.38) Tertiary -VET premium ** (-0.44) (-2.76) (1.82) (1.37) (-0.02) (-1.02) Tertiary -GED 1.70** 1.80** 1.56** 0.72** 1.68** 1.87** (16.72) (12.97) (10.57) (2.70) (10.15) (13.14) High-skilled Secondary -VET premium -0.29** -0.32* -0.35** * (-2.96) (-2.43) (-2.64) (-1.74) (-1.92) (-2.38) Secondary -GED 1.72** 1.70** 1.79** 1.28** 1.67** 2.08** (13.86) (10.36) (9.45) (5.32) (7.31) (11.87) Tertiary -VET premium -0.83** -0.96** -0.54** ** -0.96** (-5.97) (-5.27) (-2.59) (0.10) (-3.39) (-4.98) Tertiary -GED 4.32** 4.26** 4.36** 2.86** 4.37** 4.71** (33.20) (25.40) (21.09) (9.51) (18.77) (25.47) R Observations Notes: Dependent variable: first job, occupation level. Multinomial logit estimates, z-statistics reported in brackets, heteroskedasticity-robust standard errors. The set of regressors also includes country dummy variables and individual-level controls. * p<0.05, ** p<0.01. Finally, Figures 8 and 9 examine the search methods for the first job. The dependent variable comprises five different items: Formal, Informal, Previous experience, Private, and Other (reference group). Multinomial logit estimates, reported in Table 7, indicate that VET graduates are more likely to use formal than informal methods for job search. The use of past experience is also more widespread, indicating that VET degrees are more successful in providing a clear direction to secondary education graduates and, as a consequence, prompt the use of more formal job search methods. General education programs, on the contrary, do not have a clear connection with the labour market and informal methods are needed to compensate for these deficiencies. 18

19 Figure 8. First job, finding method: formal VET GED Figure 9. First job, finding method: informal 2.37 VET GED

20 Table 7. VET and job finding method Formal Secondary -VET premium 0.31** 0.27** 0.48** * 0.42** (4.02) (2.62) (4.09) (1.29) (2.13) (3.52) Secondary -GED * (-0.90) (0.08) (-2.49) (-0.77) (-0.21) (-0.80) Informal Secondary -VET premium ** * (1.95) (1.10) (3.19) (1.16) (1.39) (2.36) Secondary -GED -0.48** -0.36** -0.75** -0.57** -0.41* -0.59** (-5.44) (-3.11) (-5.70) (-3.13) (-2.55) (-4.21) Previous Secondary -VET premium 0.89** 0.89** 0.84** 0.72** 0.79** 0.88** (7.53) (5.24) (5.06) (3.11) (3.73) (4.81) Secondary -GED (0.51) (0.05) (-1.26) (1.51) (0.28) (-1.48) Private Secondary -VET premium (-0.71) (-1.19) (1.61) (-0.03) (0.72) (-0.10) Secondary -GED * (-0.72) (0.67) (-2.43) (-1.14) (-0.34) (-0.85) R Observations Notes: Dependent variable: job finding method. Multinomial logit estimates, z-statistics reported in brackets, heteroskedasticity-robust standard errors. The reference group is Other method. The set of regressors also includes country dummy variables and individual-level controls. * p<0.05, ** p< Current employment Employment status Consider a young person who is approaching the end of compulsory schooling. She faces a number of choices or outcomes: continue studying, try to conjugate study with working activity, enter the labour market with or without success (i.e. becoming employed or unemployed) or remain inactive. These choices have been modelled as separate categories in a multinomial logit model. The categories considered are Employed, Unemployed, Student and Inactive, with the latter being used as the reference group. Figures report the results. Figure 10 shows that completing upper-secondary education increases the likelihood of being employed. More specifically, the log-odds for being employed, as opposed to being inactive, are expected to increase by 0.81 for VET secondary education, relative to below secondary, and by 0.80 for general secondary education. The effect of secondary education on the probability of being employed is larger for females than for males. 20

21 Figure 10. Current employment status: employed VET GED As reported in Table 8, the difference between the two coefficients is small (0.01) and not statistically significant. Interestingly the gender split reveals that the secondary education employment premium is higher for females than for males, while the differential premium for VET is positive and significant only for males. Thus, while the overall secondary education premium is stronger for females, the secondary VET premium appears to be relevant only for males. The employment premium seems fairly constant at different age level. Considering the unemployed status (Figure 11), secondary education also increases the likelihood of being unemployed, albeit the size of the coefficients is less than half of those for employment. The apparent contradiction is explained by considering that the estimates reported refer to a double benchmark. The first is the group of individuals who have not completed secondary education, while the second is the inactive status used as a reference group. The estimates suggest that, with respect individuals who have not completed secondary education those who have completed secondary education are less likely to be inactive. Being less likely to be inactive, they are more likely to participate to the labour market and therefore also to be unemployed. 21

22 Figure 11. Current employment status: unemployed 0.43 VET GED These results are confirmed by Figure 12, reporting the log odds of the student employment status. In this case the difference between VET and general education is large and strongly significant. While having a secondary level of education reduces the likelihood of being inactive, within this group the choices appear to be different: individuals with VET (particularly males) are less likely to be student and more likely to participate in the labour market, while individuals with general education are more likely to continue studying. Figure 12. Current employment status: student VET GED

23 Table 8. VET premium, employment status Employed Secondary -VET premium * (0.27) (2.04) (-0.57) (0.78) (-0.22) (1.64) Secondary -GED 0.80** 0.56** 0.93** 0.83** 0.96** 0.80** (15.81) (5.70) (15.22) (8.72) (10.31) (9.61) Tertiary -VET premium * (0.88) (1.80) (0.08) (2.56) (0.12) (0.13) Tertiary -GED 1.53** 0.95** 1.77** 1.26** 1.73** 1.68** (26.42) (7.65) (26.98) (11.04) (16.60) (19.03) Unemployed Secondary -VET premium * (1.09) (2.39) (0.42) (0.92) (-0.30) (1.39) Secondary -GED 0.29** ** 0.29* 0.29* 0.21 (4.55) (0.37) (4.80) (2.53) (2.36) (1.62) Tertiary -VET premium (0.15) (1.40) (-0.34) (1.00) (-0.78) (0.94) Tertiary -GED 0.54** ** 0.69** 0.61** 0.32* (7.32) (-0.04) (7.75) (5.05) (4.71) (2.56) Student Secondary -VET premium -1.90** -1.88** -1.79** -2.05** -2.08** -2.07** (-31.99) (-17.71) (-24.23) (-21.60) (-16.20) (-9.08) Secondary -GED 1.58** 1.64** 1.39** 2.54** 3.56** 2.62** (27.10) (15.42) (19.18) (24.10) (17.03) (10.09) Tertiary -VET premium -0.38** ** (-2.97) (-0.25) (-2.64) (-0.97) (-1.19) (-1.00) Tertiary -GED 1.19** 0.99** 1.05** 2.01** 2.80** 1.96** (16.59) (7.25) (11.97) (15.53) (12.95) (7.19) R N Notes: Dependent variable: current employment status. Multinomial logit estimates, z-statistics reported in brackets, heteroskedasticity-robust standard errors. The reference group is inactive individuals. The set of regressors also includes country dummy variables and individual-level controls. * p<0.05, ** p<0.01. In order to investigate these issues in more detail, we also considered analysing separately students and non-students for the employed-unemployed and inactive categories. In this case the reference category becomes inactive non-students. Table 9 reports the results. While individuals with general education are more likely to be employed while studying, individuals with VET are more likely to be employed and non-student. A similar result is obtained for unemployment. Analogously, the pure education choice (inactive student) is more likely for individuals with general education with respect to individuals with VET. 23