Chapter 3. Latvian labour market policies for skills and employability

Training for the unemployed has remained a key component of active labour market policy in Latvia. This chapter assesses how effective such trainings have been in helping unemployed people access good jobs and considers how training provision could be improved using detailed, linked administrative data. The chapter finds that training for the unemployed has had positive effects on individuals’ chances of (re-)entering employment and on earnings among those who found a job. While these effects differed according to the gender, age, and social assistance receipts of training participants, virtually all types of participants benefited from taking part in training for the unemployed. In addition, combining training with other active labour market policy measures, especially measures to support regional mobility, appeared to boost effectiveness. On implementation, the chapter directly considers the implications of providing training for the unemployed through a voucher system.

Introduction

Despite changes in the landscape of active labour market policies (ALMPs) in Latvia in recent years, training has remained an important strategy for connecting people with jobs. Training has continued to be important because historical trends in the organisation of Vocational Education and Training (VET) provision, a weak tradition of lifelong learning, emigration, and changing employer demands have resulted in skills shortages in some occupations and sectors. Providing unemployed individuals with these skills may help them integrate into the labour market. This chapter assesses how effective those trainings provided under Latvia’s menu of ALMPs have been in helping unemployed individuals access good jobs and then considers how training provision could be improved.

The chapter shows that trainings have had positive impacts on unemployed Latvians’ labour market outcomes, but that the effects may differ (1) for certain sub-groups and (2) when training is combined with other ALMPs. At 18 months after the start of training, formal trainings increased participants’ likelihood of being employed by 7.7 percentage points while non-formal trainings increased participants’ likelihood of being employed by 5.3 percentage points.1 The sub-group analysis further suggests that: (1) women may have derived higher benefits from formal training while men may have derived higher benefits from non-formal training; (2) workers aged more than 30 years old derived higher benefits from training, especially formal training; and (3) social assistance recipients derived higher benefits from formal training, at least in the long run. Despite these differences, however, it is striking that formal and non-formal trainings appear to have positive effects on the labour market outcomes of virtually all sub-groups considered in the analysis. At the same time, combining training with mobility support and short “Measures to Improve Competitiveness” may boost its effectiveness.2

On implementation, the chapter discusses the delivery of training programmes through vouchers. While several key advantages to providing training through vouchers are outlined, three key risks are identified, which may warrant supplementary policy work. First, vouchers are less likely to be redeemed by (1) the young and (2) those with a weaker command of the Latvian language, suggesting caseworkers may need to provide additional support help certain groups use their vouchers. Second, there is substantial variation in the number of training providers across Latvia’s municipalities, underlining the importance of supporting regional mobility to foster the choice and competition on which the success of voucher systems rely. Third, individuals have to wait a long time to actually receive their vouchers: this may lead to periods when voucher recipients-to-be are unsure of their status, potentially compounding lock-in effects and prolonging unemployment spells. This latter point resonates with a trade-off policymakers typically face when implementing training for the unemployed, between building productivity among participants – which results in better long-term labour market outcomes but potentially takes time – and getting people back into work quickly.

The chapter proceeds as follows. The first main section sets the scene by describing skills shortages in Latvia, explaining the factors that have shaped the uptake of VET, and outlining reforms to the broader vocational and higher education systems that have followed in recent years. The second section uses detailed linked administrative data to assess the impact of formal and non-formal trainings for the unemployed – managed by Latvia’s State Employment Agency (SEA) – which form part of Latvia’s menu of ALMPs. The third section explores the implementation of these training measures in more detail, looking especially at the implications of delivering training through vouchers. The final section briefly concludes.

Developing skills in Latvia

This section examines where skills shortages in Latvia arise, documents trends in enrolment in VET and higher education, and considers how the VET and higher education systems have been reformed in recent years. This discussion of the broader trends in skills development in Latvia provides the background for the evaluation of training for the unemployed that comes in the following section.

Certain skills are in shortage in Latvia

Certain occupations in Latvia – many of which require some level of vocational education (either at the secondary or tertiary level) or some other type of tertiary education – are characterised by shortages of skilled labour, according to the OECD Skills for Jobs Database. This database calculates the extent of shortage in a particular occupation according to (1) wage growth, (2) employment growth, (3) growth in hours worked, (4) the unemployment rate, and (5) the growth in the proportion of workers who are underqualified. This information is presented in detail in Chapter 1. The database reports that there are key shortages in Latvia’s service sector, with shortages for workers with skills in customer and personal services and in sales and marketing being especially large. There are also shortages of workers with advanced quantitative and engineering skills (including in computers and electronics and telecommunications) that would typically rely on an education involving science, technology, engineering and mathematics (STEM), as well as shortages of workers with administrative and management skills.

Alongside the occupation-specific skills outlined above, employer surveys indicate that Latvian employers also need workers who possess cross-cutting skills, including Information and Communications Technology (ICT) skills and foreign languages. User-level computer skills and English language skills were required or preferable in approximately half of all vacancies published in 2018 (EURES, 2018[1]). In addition, Russian language skills were required or preferable in almost three-quarters of vacancies posted, with the demand for Russian being highest amongst employers in Riga and the Latgale region.

Hiring practices are also consistent with there being specific skills shortages in Latvia, although such practices may also simply reflect relatively high job quality in the country and employers’ optimistic perceptions of the workforce. The 2016 European Labour Force Survey indicates that just 35% of new hires were offered fixed-term contracts (the third lowest proportion in the OECD) meaning that the vast majority of new hires were offered permanent contracts. This may exemplify the additional incentives Latvian employers provide in order to attract workers with the right skills. However, it may also be that jobs are simply of higher quality, on average, in Latvia, perhaps due to the legal framework that governs hiring practices. Indeed, fixed-term contracts are only allowed in certain situations in Latvia, including seasonal work, replacement of absent employees, or casual work not normally performed within a particular firm (ILO, 2019[2]).3 At the same time, the high prevalence of hiring under-qualified candidates and the low prevalence of hiring over-qualified candidates in Latvia compared with other OECD countries (as per the OECD Skills for Jobs Database) is also consistent with there being skills shortages. Yet hiring of under-qualified candidates may also arise when employers have optimistic perceptions of the workforce and are willing to take on candidates whose skills they then build on-the-job.

Despite skills shortages, inter-occupation earnings premia are low compared with the OECD average

The earnings premia associated with attaining higher levels of formal education remain low in Latvia compared with the OECD average, as discussed in Chapter 1. Workers with tertiary education earn 44% more than those with upper secondary education in Latvia, but the differential is 53% for the OECD on average.4 Similarly, Latvian workers with less than upper secondary education earn 12% less than those with upper secondary education, but the equivalent difference is 19% across the OECD. Nevertheless, earnings premia are even lower in a number of other European countries than in Latvia. In Estonia, for example, workers with tertiary education earn just 24% more than those with upper secondary education, which is similar to the differential observed in Scandinavian countries.

Inter-occupation differences in earnings are also smaller in Latvia than in other OECD countries, especially in high-end occupations (Figure 3.1). The average hourly earnings of managers are almost double the average hourly earnings of all workers across the OECD, but in Latvia, the earnings premium for managers is 73%. Equally, professionals earn 47% more than average workers across the OECD as a whole, but the earnings premium for Latvian professionals is 34%. Again, however, inter-occupation earnings differences in Latvia are comparable to other countries in Europe: the earnings premium for professionals, for example, is remarkably similar across all three Baltic states. Interestingly, the size of the earnings penalty among service and sales workers is similar in Latvia to the rest of the OECD, despite the apparent skills shortages in these occupations.

Figure 3.1. Occupational earnings premia in the Baltic states and the OECD

The low returns to tertiary education and smaller inter-occupation earnings premia compared with other OECD countries, may partly explain why skills shortages persist in Latvia. There may be insufficient incentive for individuals to invest in skills that are in shortage. Nevertheless, the provision of education, training, and career guidance for young people – to which this chapter now turns – is also likely to affect skills shortages.

Relatively few Latvians attain Vocational Education and Training

The proportion of young people attaining some form of upper secondary education in Latvia is comparable to its Baltic neighbours and is higher than the EU average.5 In 2017, a little more than 87% of Latvians aged 20-24 had attained at least upper secondary education, compared to 83% for the EU as a whole. There has also been a substantial reduction in the number of young Latvians leaving full-time education early in recent years. Between 2007 and 2017, the proportion of 18-24 year-olds with at most lower secondary education, but who were no longer in education or training, fell from 16% to 9%: this constitutes the fourth largest improvement in the EU over that decade (Eurostat, 2015[3]).6

Nevertheless, while the proportion of young people with VET in Latvia is higher than in other Baltic States, it remains well below the EU average (Figure 3.2). Among those Latvians aged 20-24 who had attained upper secondary education but not tertiary education, just under 40% had focussed on vocational qualifications, according to the latest European Labour Force Survey.7 For the EU as a whole, this proportion was 49%. Dropout rates also appear to be higher for vocational education than for other educational pathways. Statistics from the Ministry of Education and Science (MoES) show that the average non-completion rate of students in general upper secondary programmes was 1.8% in the 2012/2013 academic year but ranged between 13% and 16% for equivalent vocational programmes (MoES, 2014[4]).

One possible reason for this relatively low take-up in the recent past is that the reputation of VET compared with other educational pathways in Latvia has not historically been strong. A 2011 Eurobarometer survey administered across the EU found that only 63% of respondents (second-lowest value; EU average of 75%) in Latvia considered learning in vocational schools to be of high quality and just 60% (sixth-lowest value; EU average of 73%) perceived VET to have a “positive” image (European Commission, 2011[5]). Moreover, Latvians were amongst the most pessimistic about the notion that VET professions were highly demanded in the labour market (third-lowest value; 60% in Latvia compared to EU average of 73%) with relatively few believing that VET graduates had good career opportunities (fourth-lowest value; 57% in Latvia compared to EU average of 72%). The recent wave of reforms to the VET system in Latvia have sought to improve the quality of VET in the country, building its reputation, and encouraging young people to pursue vocational pathways through the education system, but – as discussed below – results have so far been mixed.

Figure 3.2. Type of educational attainment among individuals with upper-secondary but without tertiary education among EU countries

With the youth population in decline, the number of young people enrolled in tertiary education has fallen

The number of young people in Latvia has declined in recent years, leading to a corresponding drop in the number of young people enrolled in tertiary education. Between 2010 and the 2015, the total population of 20-24 year-olds in Latvia fell by 24.4%, from 165 000 to 125 000 (UN DESA, 2017[6]). Over the same period, the number of 20-24 year-olds enrolled in tertiary education in Latvia fell by 25.3%, from 59 000 to 44 000 (OECD, 2018[7]). Some young people are choosing to enrol in universities abroad, in part because this option has become easier following Latvia’s accession to the EU. Between 2007 and 2011, the number of Latvians enrolled in tertiary education in other OECD countries almost doubled, reaching 6 650 students, nearly half of whom were based in the United Kingdom (OECD, 2016[8]).8 The United Kingdom’s exit from the EU may affect this trend or divert prospective Latvian students to other EU countries (see Box 1.1 in Chapter 1).

The decline in university enrolment has not been even, with the drop being especially sharp among students who are not enrolled in state-subsidised programmes. Access to these state-subsidised programmes is highly competitive and determined by exam scores, with subject- and institution-specific quotas being set by the MoES. Students in these state-subsidised programmes receive their tuition free of charge, although they still have to compete for a limited number of government-funded monthly stipends, which are awarded to the students with the highest prior academic achievement. The fees for those individuals who do not participate in state-subsidised programmes vary substantially. In 2013/14, the yearly academic fees for bachelor’s degree students ranged from EUR 882 to EUR 5 208, depending on the subject and the institution (MoES, 2014[9]; World Bank, 2014[10]).9

Notwithstanding the ongoing improvements discussed above, tertiary education may still need further realignment with labour demand, in order to encourage young people to enrol in universities in Latvia. The quotas for state-subsidised programmes have recently sought to bolster the number of graduates in STEM-related fields, leading to moderate increases in the proportion of graduates completing STEM degrees between 2004 and 2014 (Central Statistical Bureau of Latvia, 2015[11]). This also means that the decline in the number of students (described above) has been larger in non-STEM subjects. Nevertheless, continued reform may be needed if Latvia is to achieve its target of 27% of all graduates completing degrees in STEM-related fields by 2020. For one, pursuing STEM subjects remains relatively rare among women, a problem experienced by many OECD countries (OECD, 2016[12]). Additionally, private tertiary education providers remain focussed on social sciences, business, and law – skills that are currently less in shortage according to the OECD Skills for Jobs Database – rather than STEM-related fields. Tackling these issues may help ensure that tertiary education is valued by employers and, in turn, that young people are more willing to enrol in tertiary education institutions in Latvia.

Regardless of whether tertiary education itself is completed in Latvia or abroad, it appears that Latvians holding tertiary education have typically been more likely to emigrate. As Hazans (2013[13]) shows, in 2010/11 approximately two-thirds of Latvian students aged 18-65 – most of whom are likely to be in tertiary education – indicated that they intended to live and work abroad, although it should be borne in mind that these data come from shortly after the lowest point of the financial crisis. These intentions to move were also borne out by trends in emigration: around 24% of all emigrants who left Latvia since 2000 were either a student or trainee before departing (Hazans, 2015[14]).

Nevertheless, despite the decline in the number of young people enrolled in tertiary education and the draw of opportunities abroad, the share of the population holding tertiary education in Latvia has actually risen in recent years. In the decade to 2017, the proportion of Latvians aged 25-64 years old holding tertiary education increased by more than half, rising from 22% to 34% (OECD, 2018[7]). The analogous increase has been even larger for 30-34 year-olds, amongst whom the share holding tertiary education rose from 26% to 44% between 2007 and 2017, bringing Latvia above the EU average (Eurostat, 2018[15]). The rising proportion of individuals holding tertiary education has occurred in part because the decline in enrolment has been commensurate with the decline in the youth population. Young cohorts continue to add a disproportionately higher share of tertiary educated individuals to the population as a whole, especially given the low levels of enrolment in tertiary education prior to 2000: in 1995 the gross enrolment rate for tertiary education was just 23% compared to 81% in 2016 (World Bank, 2019[16]).10 Additionally, while relatively few highly-educated Latvians indicated that they intended to return when the most recent migration intentions survey data were collected 2010/11 (see (Hazans, 2015[14])) migration patterns have started to change in the last decade. While still negative, 2017 saw the highest level of net migration in Latvia since the year 2000, suggesting that at least some Latvians – potentially with tertiary education – may be returning (CSB, 2019[17]).

Learning on the job has historically been rare in Latvia, although its coverage has broadened recently

Opportunities for learning on the job have traditionally been limited in Latvia, fostering a weak culture of lifelong learning, but new evidence suggests that employers are now becoming more involved in building skills among their employees. In 2010, only 40% of Latvian enterprises provided any sort of Continuing Vocational Training (CVT), but virtually all Latvian enterprises did so by 2015 (Eurostat, 2015[18]; World Bank, 2015[19]). However, these statistics use a very broad definition of what counts as CVT, including participation in conferences and trade fairs for the purposes of learning. Latvia still lags behind European countries, if a strict definition of CVT – which includes only structured courses conducted in locations away from the active workplace – is adopted. Only 31% of Latvian enterprises provided CVT (under the stricter definition) in 2015, which is approximately half the EU average and significantly less than Estonia (64%) and Lithuania (44%).

The proportion of adults in some form of education has risen in Latvia, but this education has become shorter and less frequent. Between 2011 and 2016, the proportion of employed Latvians aged 25-64 who participated in some form of formal or non-formal education over the last 12 months increased from 40% to 56% (Figure 3.3). However, the time actually spent in adult education decreased substantially. Between 2011 and 2016, the estimated average time of instruction among employed adults who participated in formal or non-formal education fell from 148 hours to 92 hours for Latvia and from 103 hours to 94 hours for the EU as a whole (Eurostat, 2016[20]).11 In addition, Latvian adults are far less likely than adults in other EU countries to have received education in the previous four weeks. In 2017, 7.5% of Latvians aged 25-64 had participated in formal or non-formal education in the previous four weeks, compared with 10.9% for the EU as a whole (Eurostat, 2017[21]). That said, while the rate of adult participation in education (in the last four weeks) remains low in Latvia, the proportion has in fact increased in recent years, up from 5.6% in 2014 (an increase of just over one-third).

Figure 3.3. Proportion of the workforce participating in education

Vocational Education and Training and tertiary education have been reformed to promote uptake

Since 2009, the government of Latvia has embarked on a menu of deep and wide-ranging reforms of the vocational education system, which seek to meet three key objectives (Cabinet of Ministers, 2009[22]). First, the reforms aim to promote the quality of vocational education in Latvia. Second, the reforms aim to align vocational education with the needs of the labour market. Third, the reforms aim to ensure resources are used efficiently, raising the attractiveness of vocational education to potential learners and to employers (MoES, 2015[23]).

One of the main components of these reforms has been to consolidate and restructure the network of institutions that provide VET in Latvia. This restructuring aims to: (1) improve access to vocational education; (2) ensure co-operation between key stakeholders, including education institutions and employers; and (3) promote efficient use of resources (Cabinet of Ministers, 2010[24]). Since 2009, larger vocational schools – those with more than 500 students outside of Riga and more than 800 students in Riga – have gradually been transformed into Vocational Education Competence Centres (VECCs), which act as regional hubs to develop closer links between vocational education and employers, to improve quality, and to provide pedagogical support for other vocational schools. To ensure VECCs boost the quality of VET, they must meet several specific criteria. Not only must VECCs meet certain standards in terms of students’ results, but they must also ensure that they work with the latest technologies, provide career guidance, and create and publish educational and methodological materials for learners and educators online. VECCs are also tasked with providing part-time learning, which is vital for adults wishing to participate in education and training activities while remaining in employment. At the same time, VET schools with fewer than 300 students have been merged with VECCs or, more rarely, combined with general education schools. As a result of these reforms, the number of vocational education schools for which the MoES is responsible dropped from 59 to 21 between 2010 and 2018 (OECD, 2016[12]; MoES, 2019[25]). While consolidating the provision of VET in this way offers a clear strategy for improving efficiency and involving stakeholders, there is a risk that reducing the number of VET providers also reduces (physical) access. Certain steps have been taken to combat this issue, including supporting learners’ mobility (discussed with reference to training for the unemployed in the section on vouchers) and building or renovating dormitories on site. Nevertheless, uptake of VET will only increase if the reputation and perception of the quality of VET provision also improves.

The VET curricula has also been reformed to ensure that vocational education is of high quality and is relevant to labour market needs in Latvia. Since 2011, special Sector Expert Councils (SECs) have been established to give key stakeholders a voice to shape the content of vocational education. The SECs include employer representatives (from industrial associations), government representatives (from the relevant ministries), and employee representatives (from the Free Trade Union Confederation). Latvia is also modularising vocational education, dividing programmes into discrete components (the “modules”) that have specific learning outcomes, teaching methods, and indicators of achievement. By 2013, 56 VET programmes had been modularised and a further 68 are currently in the process of being modularised. Modularising VET programmes enables vocational education to adapt to changes in the nature of work brought about by technological advance and globalisation, giving students greater labour market flexibility (Pilz, 2012[26]). In line with the restructuring described in the previous paragraph, VECCs were given responsibility for approving newly-developed modules in the 2016/17 academic year (CEDEFOP, 2018[27]). Finally, Latvia has developed a set of consistent qualification standards for VET, which increasingly align with the European Qualifications Framework (EQF) to help potential participants and employers understand the level and content of vocational training.

The latest evidence on whether perceptions of and enrolment in vocational education have improved in response to the reforms described above is mixed. A CEDEFOP survey from 2016 reveals that the results from the 2011 Eurobarometer have changed relatively little insofar as perceptions of VET in Latvia are less positive than for the EU as a whole (although making direct comparisons between these two surveys is difficult given differences in the sample and the questionnaire).12 The 2016 CEDEFOP survey suggests that just 61% of Latvians have a positive image of VET compared with 68% for the EU as a whole, while 76% of Latvians agreed with the statement that “people in vocational education learn skills that are needed by employers [in our country]” compared with 86% for the EU as a whole (Daija, Krastina and Rutkovska, 2018[28]). At the same time, enrolment in vocational upper secondary education relative to general upper secondary education has risen very slightly in the last two years, climbing from 38.2% of all those in upper secondary education in 2016 to 38.9% in 2018 (CSB, 2019[29]). However, these changes are relatively small especially when placed in recent historical context: indeed, in 2013, 39.1% of upper secondary students were in vocational education (OECD, 2018[7]).

New arrangements for learning in the workplace have also sought to promote adults’ participation in VET, which may partly explain some of the rise in the proportion of adults participating in on-the-job training discussed above. Following successful pilot projects in the 2013/14 and 2014/15 academic years, the vocational education law was amended in 2015 to define clear roles for learners, for SECs (described above), and for enterprises when providing learning for their employees (European Commission, 2015[30]). Learners now receive both theoretical and practical training at both a vocational school and at the company, with the latter comprising at least 25% of the training time. SECs help to promote and evaluate arrangements for learning in the workplace, ensuring co-operation between employers and education institutions. Given SECs’ links to the formal education system, learning in the workplace now leads to nationally recognised qualifications. As of January 2017, enterprises are required to assign workers appropriate mentors, who must have a master of crafts certificate, vocational education, or at least three years of relevant work experience as well as certified teaching competence. Additionally, there are now tax exemptions for scholarships for learning at work, not exceeding EUR 280 per month, to further incentivise participants and enterprises (CEDEFOP, 2018[27]).

Nevertheless, the fact that the new arrangements for learning in the workplace (described above) operate largely in isolation from the existing apprenticeship system in Latvia may limit their effectiveness (OECD, 2016[12]). Although the existing apprenticeship system – organised through the Chamber of Crafts – has very few participants, it may be a source of expertise on learning on-the-job that is currently going untapped.13 At the same time, those individuals working through the existing apprenticeship system (rather than the new arrangements for learning in the workplace) fall outside the formal education system. The qualifications that result from the existing apprenticeship system do not provide access to regulated professions nor the formal education system, and there are also no mechanisms in place for reintegrating those individuals who drop out (Daija, Kinta and Ramiņa, 2014[31]).

The tertiary education system has also been reformed to boost quality assurance and strengthen finances, complementing the reforms to VET described above. In July 2015, the Academic Information Centre (AIC) become the institution responsible for quality assurance in higher education (including accreditation and licensing), operating in accordance with EU standards and regulations. Within AIC, a separate department known as the Quality Agency for Higher Education or Augstākās izglītības kvalitātes aģentūra (AIKA), which focusses solely on quality assurance, is currently aiming to align with the European Quality Assurance Register for Higher Education (EQAR) (ENQA, 2018[32]). Aligning with EQAR in this way would be a vital step for promoting the quality, visibility, and international recognition of Latvia’s tertiary education system (OECD, 2016[12]). In 2015, Latvia also adopted a “three-pillar” funding model, designed to balance stability, performance, and innovation. The funding model was developed with support from the World Bank, involving also representatives from the higher education sector and other social partners. The three pillars of the funding model comprise: (1) base financing (institutional financing to ensure the functioning of education and research); (2) performance-based financing (financing that is allocated to reaching set study outcomes and research results); and (3) innovation financing (future development-oriented financing that promotes specialisation of institutions and profile development) (World Bank, 2017[33]).

Given how recent the reforms to the tertiary education system have been, it is unlikely that drastic changes in quality and hence uptake would already be observed. There is some suggestive evidence that the decline in absolute enrolment described above may be starting to stabilise. Between 2015 and 2016, the number of people of all ages enrolled in tertiary education fell by just 1.9%, the smallest year-on-year decline observed since consistent data collection began in 2005 (OECD, 2018[7]). Additionally, the rate of gross enrolment in tertiary education is now high in Latvia, having grown substantially over the previous two decades. Even in recent years there has been an uptick in the gross enrolment rate in tertiary education, rising from 67% to 81% between 2012 and 2016 (World Bank, 2019[16]). However, it remains to be seen whether these recent trends will persist and whether they can really be attributed to the reforms to the tertiary education system.

A specialised Training Commission also helps align education and training with future skills shortages

There are two main types of labour market forecast in Latvia, each with very different aims. First, the SEA has its own short-term forecasting model, which produces sector – and region-specific predictions about skills shortages, by combining: (1) macroeconomic data from Eurostat and the Ministry of Economics; (2) labour market data from the Labour Force Survey; and (3) employers surveys (European Commission, 2016[34]). Second, the Ministry of Economics publishes an annual report with medium- and long-term forecasts for broader measures of labour supply and demand. Labour market forecasting can be challenging in Latvia, for two main reasons. First, given Latvia’s population, sample sizes for sector-specific data may be small, making it harder to produce sector-specific forecasts, especially over long time horizons. Second, Latvia is a very open economy, such that large, unpredictable sectoral shifts can occur in the face of external shocks. Despite these challenges, both the SEA’s short-run forecasts and the Ministry of Economics’ long-term forecasts provide crucial insights into future skills shortages in Latvia. Indeed, since 2016, both the SEA and the Ministry of Economics have been working to improve the quality of Latvia’s labour market forecasts, under a 5-year European Social Fund project.

Since the two sets of forecasts are so useful for guiding training-relevant policies, Latvia has a specialised “Training Commission” (established in 2003), which seeks to integrate the SEA’s short-term forecasts and the Ministry of Economics’ long-term forecasts (Zvīdriņa, 2015[35]; Bratti et al., 2018[36]). The fields of study for training for the unemployed – on which the next section focusses – are decided by meetings of the Training Commission, which take place at least once a year.14 The Training Commission brings together representatives from key ministries – including the Ministry of Welfare, Ministry of Economics, and MoES – but also includes members from the SEA, local government associations, and employers’ associations to ensure the voices of all key stakeholders are heard. Combining the short- and long-term forecasts is not an easy task, so meetings of the Training Commission adopt a specific structure to facilitate co-ordination (EACEA, 2018[37]). The Ministry of Economics first presents the long-term forecasts. The SEA then presents the implementation results of ongoing training measures as well as the results of the short-term forecasts. All members of the Training Commission then review the full list of fields of study to determine which should be retained, which should be suspended, and whether any types of training should be added.

Training is a tenable strategy for activating the unemployed in Latvia

The setting in Latvia means that providing training to unemployed individuals is a tenable strategy for helping them to connect with good jobs. Firstly, there are skills shortages in certain occupations and sectors in the Latvian economy. While providing training for the unemployed should not necessarily be seen as the main way to address these skills shortages at the macro level, the fact that skills shortages exist means that building skills among the unemployed may provide them with a tenable pathway back into work. While the recent and ongoing reforms to the VET and tertiary education systems support the job prospects of future cohorts, they may come too late for those who have already entered the labour force. Equally, the expanding coverage of on-the-job training only helps those who are actually in work.

Description and evaluation of Latvia’s main training programmes for the unemployed

This section evaluates the effectiveness of selected training programmes for the unemployed, which fall under the menu of ALMP measures implemented in Latvia. The analysis focusses principally on formal vocational trainings (henceforth “formal trainings”) and non-formal trainings, which typically last several weeks and build concrete and substantive skills. These are distinct from shorter “Measures to Improve Competitiveness” (MICs), which typically last one or two days and try to develop individuals’ approach to engaging with the labour market (for example by improving CV writing or interview technique). The section begins by outlining the formal and non-formal training programmes that are evaluated and describing the detailed linked administrative data on which the evaluation draws. The section continues by explaining the challenge of evaluating training programmes that begin at different times throughout individuals’ unemployment spells, putting forward an econometric approach to deal with this challenge. Finally, the section reports the main evaluation results, exploring the sub-groups for which trainings are most effective and testing the implications of combining trainings with other ALMP measures. Where possible, the analysis builds on the previous evaluation of training for the unemployed in Latvia undertaken by Hazans and Dmitrijeva (2013[38]) to better contextualise the estimated effects.

The evaluation focuses on substantive formal and non-formal trainings

As discussed in Chapter 2, several ALMP measures involve some form of training or workshop. Alongside the formal and non-formal trainings on which this chapter focusses, in 2017, there were also approximately 80 000 participations in MICs.15^,16 These MICs seek to equip participants with the competencies required to engage successfully in the labour market. They often comprise very short courses, focussing on how to write CVs, how to succeed during interviews, and how to network effectively. There were also small numbers of trainings that cannot be classified as formal training nor non-formal training nor MICs. In 2017, there were approximately 600 participations in “workshops for young people” as part of the Youth Guarantee and 300 participations in “trainings at the employer”.

This chapter mainly analyses formal and non-formal trainings for three main reasons:

First, the content of formal and non-formal trainings tries to build concrete skills that are in demand in the Latvian labour market. Formal trainings build a specific new skill such as social care, project management, or welding, with participants working towards a professional qualification. Non-formal trainings, which do not necessarily result in a formal qualification, cover cross-cutting skills, such as languages and ICTs, which are in demand among employers. Figure 3.4 shows the number of formal and non-formal trainings or the unemployed that took place between January 2012 and October 2017, falling under different fields of study.
Second, formal and non-formal trainings last longer and require more contact hours than MICs (see Figure 3.5). Formal trainings take between 22 and 202 days to complete, lasting 91 days on average. They require at least 160 hours of contact time and require approximately 500 hours of contact time on average. Non-formal trainings take between nine and 134 days to complete, lasting 42 days on average. They require at least 40 hours of contact time and require approximately 125 hours of contact time on average. MICs, by contrast, typically last around one day and involve just seven hours of contact time. Given that MICs are so short, it may be difficult to capture their effects in a quantitative evaluation of the type undertaken in this chapter.
Third, the sample sizes of MICs and of other trainings are not amenable to reliable statistical analysis. Taking all individuals that received any ALMP measures from the SEA between January 2012 and October 2017, around three-quarters received a MIC at some point. In some sense, MICs – like regular caseworker meetings or career consultations – are part of the regular ongoing services provided by the SEA, rather than a discrete programme. This makes it difficult to make meaningful comparisons between individuals that did and did not receive MICs. At the other extreme, since participations in training at the employer and workshops for young people are relatively rare, the sample size is insufficient to evaluate their impacts rigorously.

Figure 3.4. Main types of formal and non-formal training

Figure 3.5. Length and duration of trainings and Measures to Improve Competitiveness

As shown in Chapter 2, there was some year-to-year variation in the number of participations in formal and non-formal trainings between 2012 and 2017, yet there was a clear decline in the number of MICs. In 2012, there were almost 150 000 participations in MICs, but this dropped to just under 80 000 in 2017.

Currently, the SEA – with guidance from the Training Commission – is reforming the provision of MICs by bundling them in with non-formal trainings, which may reduce the number of MIC participations. This is to consolidate the programme of support provided to the registered unemployed and improve the quality of trainings and MICs. These new consolidated courses will focus on topics including basic communication skills, state language, and ICT skills, while supplementary e-learning courses on topics such as financial literacy and preparing for interviews may also be offered. While it is too early to tell the effects of these reforms, the SEA anticipates a resulting increase in both the quality of non-formal trainings and the number of individuals participating in such trainings.

Both formal and non-formal trainings are provided through a system of vouchers. The vouchers specify the field of study for which they are valid, and their face value reflects the length of the training to be provided. Further details of the voucher system and its implementation are discussed in in the penultimate section of this chapter.

Linked administrative data paint a detailed picture of individuals’ participation in ALMPs and their labour market outcomes

An econometric evaluation of the effectiveness of training programmes requires rich data, which track people’s outcomes after their participation in training and contain sufficient information about their personal and household characteristics and situation. The data on which this evaluation draws come from four main sources, outlined in Table 3.1. The data cover the period January 2012 to October 2017. Unique individual identifiers allow the data to be combined, providing a rich understanding of individuals’ participation in ALMPs (from the SEA), their background characteristics (from the population registry), and their labour market outcomes and social security outcomes (from the Social Insurance Agency as well as the Social Assistance Database, which comes municipalities).

Table 3.1. Data sources used in the evaluation
Data source	Information available	Periodicity	Sample
State Employment Agency (SEA)	Participation in ALMPs, interactions with SEA, and detailed background characteristics of registered unemployed.	Start and end dates of ALMPs recorded.	Registered unemployed.
Social Insurance Agency	Employment outcomes and receipts of various benefits, including unemployment benefit, disability benefit, state family benefit, sickness benefit, and pensions.	Monthly.	All working-age individuals.
Population Registry	Individual background characteristics, including gender, age, ethnicity, citizenship status, and marital status.	Monthly.	All working-age individuals.
Social Assistance Database (from municipalities)	Receipts of social assistance and Guaranteed Minimum Income.	Monthly.	All working-age individuals.
Note: ALMP: Active labour market policy.

Despite the richness of the data on which this evaluation draws, two key limitations should be borne in mind. First, while detailed employment outcomes are known for the observation period (January 2012-October 2017), full individual employment histories are not available.17 This implies that (1) only partial employment histories can be used to control for differences between individuals when trying to estimate the effects of trainings and (2) it is not possible to know whether the first recorded unemployment spell (occurring between January 2012 and October 2017) corresponds to an individual’s first true unemployment spell. Second, it is difficult to know the precise content of the formal and non-formal trainings. This is because the SEA’s main role is to provide training participants with a voucher, aligned with a broad field of study. The precise content of each training is the responsibility of the educational institution that the voucher recipient chooses, and programme descriptions may be very long and detailed (especially for non-formal trainings). As such, the SEA does not collect information on the precise content of each specific training programme.

Looking to the data on employment outcomes from the Social Insurance Agency, it emerges that many individuals experienced more than one spell of registered unemployment between January 2012 and October 2017 (Figure 3.6).18 Approximately one third of those individuals that became unemployed at least once between January 2012 and October 2017 experienced more than one spell of registered unemployment. The typical durations of first, second, third, and subsequent unemployment spells, are discussed in Box 3.1.

Figure 3.6. Number of unemployment spells experienced

Box 3.1. Patterns in terms of unemployment spell length

Unemployment spells lasted 6 months on average but, given the possibility that individuals experience multiple unemployment spells between January 2012 and October 2017, subsequent unemployment spells were found to be shorter than initial spells (see Figure 3.7). The first recorded unemployment spells lasted seven months on average, whereas the fifth recorded unemployment spells lasted four months on average. Approximately 10% of the first recorded unemployment spells lasted more than 12 months, whereas just 4% of the fifth recorded unemployment spells lasted that long. These patterns are consistent with the unemployment benefits system giving individuals additional incentive to exit unemployment quicker after the first spell. As discussed in Chapter 2, levels of unemployment benefit depend on how long an individual was in employment before their job ended: this period of accumulation is likely to be shorter for second, third, fourth (and so on) unemployment spells. Additionally, individuals need to have been in employment for at least 12 months out of the previous 16 months to qualify for any unemployment benefits (nine months out of the previous 12 months before December 2016). As such, individuals in their second, third, fourth (and so on) unemployment spells may potentially be ineligible for unemployment benefit receipts, providing a particularly strong incentive to find a way back into work.

Figure 3.7. Distribution of spell lengths by number of recorded unemployment spells

Nevertheless, there are two important caveats on the finding that subsequent unemployment spells are shorter than initial ones. Firstly, the finding partly stems from censoring and sample selection. Individuals’ second, third, fourth (and so on) unemployment spells are more likely to be right-censored than their first unemployment spells. Subsequent spells are more likely to continue past the end of the observation period (October 2017), especially if such spells are long, so that their length cannot be observed. Indeed, if the average spell lengths are recalculated focussing only those individuals that experienced five complete unemployment spells within the observation period, the average differences between the first spell and subsequent spells are far more modest (Annex Figure 3.A.1). Of course, this recalculation introduces a different type of sample selection, by simply excluding anyone who experienced less than five unemployment spells. Secondly, individuals’ second, third, fourth (and so on) unemployment spells happen later by construction, meaning that they are more likely to coincide with the stronger period of the economic upturn (see Chapter 1). Again, differences between initial and subsequent spells are smaller, but not eliminated, if this “cohort” effect is isolated and extracted (Annex Figure 3.A.1).

This evaluation focusses primarily on those individuals that entered registered unemployment between January 2012 and October 2017: the analysis uses an “inflow” sample. It is important to know when the unemployment spell started so the analysis can correctly ascertain how long an individual had been unemployed before they received training and how long their unemployment spell actually lasted. As such, individuals that were unemployed in December 2011 and remained so in January 2012 will only be included in the analysis if they exit and then re-enter unemployment during the observation period.

The analysis also focusses specifically on individuals’ first recorded unemployment spells for two main reasons. First, the first recorded unemployment spells happen earlier in the observation period, making it easier to observe the long-term impacts of training on individuals’ labour market outcomes. Second, as Box 3.1 shows, the first recorded unemployment spells typically last longer than other spells, making it easier to consider the outcomes of those who spend longer periods in unemployment before training starts.

Individuals often participate in multiple Active Labour Market Policy measures

Training clearly remains a sizeable component of Latvia’s ALMP strategy, even after adjusting the data for the needs of this evaluation. After restricting the data to focus only on ALMP measures received during individuals’ first recorded unemployment spells, approximately 15 000 individuals participated in at least one formal training and a further 33 000 participated in at least one non-formal training (Figure 3.8). These figures include trainings provided as part of the Youth Guarantee.19 Participation in non-formal trainings was therefore wider than all employment measures – including public works and employment subsidies – taken together. However, as discussed in Chapter 2, participation in MICs far exceeded participation in any other category of ALMP measure in Latvia.

Figure 3.8. Participation in ALMP measures

Many individuals participate in more than one ALMP measure within their first recorded unemployment spell (Figure 3.9). Even if MICs, career consultations, and “short other measures” (i.e. other measures that comprise only a short consultation or session) are excluded, 30% of those individuals that participated in any of the remaining substantive ALMP measures end up participating in more than one substantive ALMP measure. Of those individuals that participated in formal training in their first recorded unemployment spell, 47% had participated in more than one substantive ALMP measure. Similarly, of those individuals that had participated in non-formal training in their first recorded unemployment spell, 36% had participated in more than one substantive ALMP measure. Estimating the effects of training despite this overlap between different ALMP measures is one of the key challenges faced by this evaluation.

Participation in formal and non-formal trainings also overlapped substantially. Approximately one third of formal training participants also participated in non-formal training during their first recorded unemployment spell, while around 15% of non-formal training participants also participated in formal training.

Nevertheless, there are certain restrictions over the number of formal and non-formal trainings in which individuals can participate. Individuals can only participate in one formal training every two years. This explains why the number of participations and participants for formal training in individuals’ first recorded unemployment spells are virtually identical in Figure 3.8. By contrast, individuals may be involved in non-formal training focussed on Latvian language up to three times per year, and any other type of non-formal training up to twice per year. This explains why the number of participations exceeds the number of participants for non-formal training in Figure 3.8.

Figure 3.9. Multiple participations in ALMP measures

The timing of ALMP measures – in terms of when they take place within individuals’ unemployment spells – varies substantially (Figure 3.10). MICs and career consultations typically happen within two months of becoming unemployed: the median time taken to receive a MIC after registration was 46 days while for career consultations it was 58 days. Participation in employment measures also occurred relatively early on in individuals’ unemployment spells (median of 124 days). This may reflect the fact that enrolment into public works schemes – which remain by far the largest component among the employment measures – happens on a predictable rotating basis, especially in rural areas where some work is seasonal. Measures to support regional mobility typically begin approximately 6 months into individuals’ unemployment spells: such measures may be explicitly linked to training to help individuals reach training providers, or they may simply help the registered unemployed to reach the location of a new job.

Formal trainings tend to occur earlier in individuals’ unemployment spells than non-formal trainings. At the median, 161 days elapsed between registering as unemployed and the start of formal training programmes, compared with 218 days for non-formal trainings. Indeed, non-formal trainings occur later in the unemployment spell than any other type of ALMP. At least part of the difference in the time between registration and the start of training that arises between formal and non-formal trainings is down to queueing. After the caseworker and SEA client have agreed on the need for training, individuals have to wait 110 days on average for non-formal trainings (at least for those focussed on foreign languages and ICTs) to start compared but around 60 days for formal training.20 These waiting times comprise the time between assignment to a training measure and receipt of a training voucher, and the time between receipt of a training voucher and redemption of that voucher. The implications of these waiting times are discussed in more detail in the penultimate section of this chapter.

Figure 3.10. Variation in the start of ALMP measures

Given the variation in the timing of ALMP measures described above, unemployed individuals that participated in training in combination with some additional ALMP measure typically participated in the additional ALMP measure first. Approximately 93% of those individuals that received both a MIC and formal training began the MIC first while 89% of those that received both a career consultation and formal training began the career consultation first.21 Additionally, 58% of those receiving both an employment measure and formal training began the employment measure first.

Nevertheless, for those individuals that received both support for regional mobility – discussed in more detail in Chapters 2 and 4 – and training, the two measures began simultaneously in 80% of cases. This reflects the fact that such mobility support was often explicitly received by individuals seeking to improve their access to training providers. Individuals were able to receive a reimbursement of EUR 100 per month to cover the costs of transport to training sites or accommodation at training sites, providing the suitable training site was more than 15 kilometres from their place of residence. However, receipt of such mobility support is not automatic: training participants must submit an application form along with supporting documentary evidence to the SEA within 10 working days of the start the training. As discussed in Chapter 4, mobility support is also available to those wishing to travel to take up new employment (regardless of whether or not training has been completed), although the minimum distance to required to qualify for such mobility support is 20 kilometres.

Among those unemployed individuals that received both formal and non-formal training, formal training was only slightly more likely than non-formal training to begin first (Figure 3.11). For approximately 56% of such individuals, formal training was sequenced before non-formal training. As such, the differences between formal and non-formal training – in terms of the average time between registration and training start – are not so stark for those individuals that receive both types of training. Individuals are also able to express a preference to their caseworkers over how formal and non-formal trainings are sequenced (if they are to be combined), although there are certain restrictions. For example, Latvian language training is prioritised for those individuals whose inability to speak the state language prevents them from integrating into the labour market.

Figure 3.11. Participation in multiple trainings

Complex patterns of participation in training necessitate a pragmatic econometric approach

In order to rigorously assess the impact of training measures, it is necessary to try and compare the employment outcomes of training recipients with what would have happened, had they not received the training: the latter can never be observed so it is necessary to find some way of constructing this “counterfactual” from the data. Normally, researchers would do this by comparing the outcomes of those individuals that participated in training and those that did not. Such comparisons may be biased because certain types of individuals (e.g. more motivated individuals) are more likely to participate in training and have better employment outcomes for reasons besides their participation in the training. Conversely, certain individuals that face additional barriers to employment – and therefore have worse employment outcomes – may be more likely to be directed towards training by caseworkers. To address such sources of bias, researchers may try to control for observed differences (in gender, education, age, and so on) between training participants and non-participants. Such methods would then produce an estimate of the “treatment effect” by effectively comparing individuals that appear similar in terms of their observable characteristics.

In the context of this evaluation, making such comparisons (even with controls for observable characteristics) may provide biased estimates of the true effects of training, because individuals participate in trainings at very different times throughout their unemployment spells. Since it takes time for unemployed individuals to begin training programmes after registering with the SEA, it may not be valid to compare trained individuals with those who receive no training. Many of the untrained may not be treated simply by virtue of the fact they find a job quickly (and exit unemployment) without support from the SEA. This latter group of individuals may have better future employment outcomes than training participants by construction: if they exit unemployment again quickly they have a good chance of keeping that job, and are much more likely to be employed in several years or months than if they had remained unemployed. Additionally, the SEA will certainly not profile them as someone needing training support, so time in unemployment will not be extended mechanically by becoming locked-in to a training course. At the same time, they may be systematically more motivated or more able than training participants, factors for which it may be difficult to adequately control with the available data.

One way to estimate the effects of training programmes that are assigned at different times throughout the unemployment spell is to focus on those individuals who have endured a set number of months in unemployment and compare the labour market outcomes of those who begin training in that month with who are still “waiting”, either for support from an ALMP measure or some other way out of unemployment. The application of this “dynamic selection-on-observables” methodology – pioneered by Sianesi (2004[39]) and Fredriksson and Johansson (2008[40]) – which is used in this analysis is explained in more detail in Box 3.2. The advantage of this approach is that it ensures trained individuals are compared with those who spent at least as long in unemployment, reducing bias when trying to estimate the effects of training.

Box 3.2. Econometric approach – Dynamic selection-on-observables

When individuals begin ALMP measures at different times throughout their unemployment spells, selecting “dynamically” into such measures, the set of individuals who were never treated does not serve as a suitable comparison group for those who were treated. Individuals only become available for treatment if they stay in unemployment long enough. Conversely, one of the main reasons that some individuals do not get treated is because they are able to find jobs and exit unemployment quickly. This motivates an approach that does not simply compare the ever treated with the never treated, but rather compares those who begin treatment at a given point in their unemployment spell with those who are still waiting for treatment at that time. This is precisely the “dynamic selection-on-observables” method developed by Sianesi (2004[39]) and Fredriksson and Johansson (2008[40]).

Implementing the dynamic selection-on-observables approach requires estimating (then aggregating) separate treatment effects for each pre-treatment duration ( $m$ , the amount of time between registration and the start of treatment) and for each time horizon of interest ( $t$ , the amount of time elapsed since the start of the ALMP measure, when the employment and earnings are measured). The potential labour market outcomes (such as employment or earnings) for an individual ( $i$ ) can be written $Y_{i m t}^{d}$ , where $d = 1$ under treatment and $d = 0$ otherwise. The average treatment effect on the treated ( $D_{i m} = 1$ ) for each $m$ and $t$ can then be written:

$γ_{m t} = E [Y_{i m t}^{1}| D_{i m} = 1] - E [Y_{i m t}^{0} | D_{i m} = 1]$

To estimate this equation:

- The treatment group comprises those individuals who begin treatment in period $m$ . This includes the small minority of individuals that subsequently drop out of training.

- The comparison group comprises those individuals who were still unemployed in period $m$ but were either treated later than period $m$ or never treated.

- Individuals who (1) received treatment or (2) became employed and therefore left unemployment before period $m$ are dropped from the estimation of a particular treatment effect $γ_{m t}$ .

Given this framework, the dynamic selection-on-observables approach can only be used to estimate the treatment effect of the first ALMP measure in which individuals participate. Everything that happens after starting participation in the first ALMP measure is effectively treated as part of individuals’ outcomes, even if that entails not working due to further participations in ALMP measures.

While the $γ_{m t}$ s are revealing by themselves, it is helpful to calculate an overall treatment effect $γ_{t}$ that is specific only to the time horizon $t$ for the outcomes of interest. The analysis in this chapter follows previous applications of the dynamic selection-on-observables approach, by taking a weighted average of all the $γ_{m t}$ s, where the weights correspond to the fraction of the total treated at each pre-treatment duration $m$ (Doerr et al., 2017[41]). For total number of individuals $N$ and maximum pre-treatment duration $M$ :

$γ_{t} = \frac{\sum_{m = 1}^{M} \sum_{i = 1}^{N} D_{i m} \times γ_{m t}}{\sum_{m = 1}^{M} \sum_{i = 1}^{N} D_{i m}}$

The $γ_{t}$ s are the key treatment effects reported for employment chances and earnings in this chapter, looking at 6, 12, 18, 24, 30, and 36 months after the start of the training.

Certain restrictions have to be placed on the pre-treatment durations (the $m$ s) to ensure there is sufficient sample for each $γ_{m t}$ to be estimable. For the main results, the analysis focusses on pre-treatment durations of between 0 and 12 months. This covers 95% of all formal trainings and 93% of all non-formal trainings. Looking between 0 and 12 months ensures that there are enough observations, when the results are broken down into certain sub-groups.

While looking at specific pre-treatment durations partly addresses concerns about selection, which may bias the estimated effects of training, individual characteristics – which are themselves correlated with individuals’ employment prospects – may still influence whether or not individuals begin treatment at a given $m$ . The analysis in this chapter therefore estimates each $γ_{m t}$ using Ordinary Least Squares (OLS), including control variables for gender, age, marital status, number of children, disability and social assistance recipient status at the start of the unemployment spell, education level, ethnicity, citizenship status, Latvian language ability, and information on when previous employment occurred (if known). The OLS regressions also include fixed effects for region, SEA branch, and the calendar month in which the individual registered as unemployed.

One particular source of bias, which needs to be addressed to generate reliable estimates of trainings’ effects, arises from unemployed individuals’ ability to anticipate their future employment prospects. If individuals suspect that they will receive a job offer in the future, even if they are unemployed now, they may have less incentive to participate in training or other ALMP measures. Seasonal workers in particular may have meaningful arrangements with employers, which allow them to predict when new work will come along. This motivates controlling for skill levels (through education level) and month of registration as well as using the information that is available on previous employment.

Training participants are also likely to anticipate the start of their training: they are typically informed that they will receive a training voucher well in advance of receiving it. As a consequence, they may lower their job search effort while they wait for training to start. This should be dealt with by making comparisons between those who do and do not start training at a given month $m$ . Even if the untrained have been exerting more search effort during their unemployment spell, such effort has not been successful up until month $m$ . However, it may be that this extra search effort improves the job chances of the untrained after month $m$ : search effort may, in some sense, be “cumulative”, perhaps if job seekers can build up networks or connections with prospective employers. This may bias the estimates of the effects of training downwards, insofar as training participants-to-be may not have accumulated as much search effort as non-participants. This potential phenomenon should be borne in mind when interpreting the results.

The analysis uses OLS regression rather than matching techniques (such as propensity score matching) to condition on observable individual characteristics when estimating each $γ_{m t}$ for three key reasons. First, matching techniques typically permit a simple comparison between treated and untreated individuals. However, for some of the analysis in this chapter, it is useful to estimate the effects for more than one type of treatment group (for example, when decomposing the treatment effect of training for those that did and did not also receive mobility support). Including more than one dummy variable for different treatment groups is straightforward when using OLS. Second, it is possible to include fixed effects (for example, for SEA branch) when estimating the treatment effects using a linear model like OLS, without encountering the Incidental Parameters Problem. This problem may affect the probit or logit estimates needed to construct a propensity score, for either matching or weighting (Neyman and Scott, 1948[42]; Lancaster, 2000[43]; Söderbom, 2009[44]). Third, estimating the effects using OLS substantially speeds up computation. Similar studies have also noted the remarkable similarity between the results emanating from OLS and other, more complex techniques such as Inverse Probability Weighting (Doerr et al., 2017[41]).

The standard errors are estimated using cluster bootstrapping with 250 repetitions, with the clusters at the level of the SEA branch. This accounts for the fact that the overall treatment effect for a given time horizon $γ_{t}$ is a composite of the treatment effects estimated from multiple pre-treatment durations (the $γ_{t}$ s). Calculating the standard errors in this way follows examples from the existing literature (Biewen et al., 2014[45]).

The analysis that follows will also report more descriptive comparisons of the employment outcomes of individuals that were and were not trained at some point during their first recorded unemployment spell, alongside the results of the dynamic selection-on-observables methodology. Firstly, hazard rates – the chances of transitioning out of unemployment into employment – at different times after registration as unemployed will be reported for trained and untrained individuals. Secondly, employment outcomes at set post-registration times will be reported for trained and untrained individuals, both with and without controls for observable characteristics. These analyses are more in line with the previous evaluation of training programmes in Latvia conducted by Hazans and Dmitrijeva (2013[38]), facilitating comparison with their results.

As well as accounting for trainings beginning at different times throughout individuals’ unemployment spells, the dynamic selection-on-observables approach also offers a practical way to deal with individuals receiving multiple ALMP measures when estimating the effects of training. For individuals that receive multiple ALMP measures, the approach focusses on estimating the impacts of the first ALMP measure that they receive. All subsequent ALMP participations are effectively treated as part of individuals’ employment outcomes: if an individual stays unemployed so that they can complete another training (or is trained again because they have remained unemployed), this is treated as information about outcomes rather than information about subsequent treatments.22

Since the dynamic selection-on-observables approach focusses on the first ALMP measure in which individuals participate, MICs, career consultations, and short other measures are not treated as substantive ALMP measures in the analysis. Treating MICs, career consultations, and short other measures as substantive ALMP measures would drastically reduce the sample of individuals for whom formal or non-formal training was the first ALMP measure received. In turn, this would make it more difficult to estimate the effects of formal and non-formal training reliably. Nevertheless, the effects of training may be estimated for those individuals who did and did not receive MICs: this enables the analysis to explore whether the effects of training are dependent on having previously received a MIC. For example, it may be that training will only affect employment chances if training recipients know how to sell their new skills – through improved CVs or good interview technique – in the labour market.

The dynamic selection-on-observables approach will also be adapted in three ways to investigate whether receiving other ALMP measures alters the estimated effects of formal and non-formal training:23

First, the effects of training can be decomposed into the effect for those who received one formal or non-formal training only and the effect for those who received training and those who subsequently received some other substantive ALMP measures (such as another formal or non-formal training, or an employment measure). However, such results need to be interpreted with some caution because receiving subsequent substantive ALMP measures may also be correlated with future employment outcomes: in order to participate in additional ALMP measures individuals need to remain unemployed for longer by construction.
Second, it is possible to look separately at the effects of training for those who previously received another ALMP measure. In particular, it is possible to check whether receiving an MIC before training starts boosts the effectiveness of that training.
Third, the effects of trainings that begin simultaneously .alongside an additional ALMP measure can be separated from the effects of trainings that begin independently. This approach is used to assess the extent to which mobility support complements formal and non-formal trainings, as mobility support is often explicitly provided to help with travel to training sites.

The outcomes on which the main analysis in this chapter will focus are individuals’ chances of employment and individuals’ earnings.24 These outcomes are considered at several different time horizons: 6, 12, 18, 24, 30, and 36 months after the start of the training. While chances of employment – captured by a variable that takes 1 if an individual is employed and 0 otherwise – can be assessed for each individual in the sample, the effects on earnings can only be observed for those who actually find work. This potential source of bias on the estimated effects of earnings should be borne in mind when interpreting the results. The chapter consequently places more emphasis on the employment effects than the earnings effects. Earnings are specified in logs as this improves the fit of the models, given the long right tail (positive skew) on the earnings distribution.

The results presented in this chapter complement a wide and growing literature evaluating the effects of training programmes on individuals’ labour market outcomes, in many different contexts. This literature suggests that, while training measures’ effects are positive (although small) on average, there is substantial variation in their impacts and many individual training programmes have no positive effects. The main strands of this literature are summarised in Box 3.3.

Box 3.3. Related literature on the effects of ALMP training measures

There is now a large and growing body of evidence on the effectiveness of training programmes and other ALMP measures from around the world. This has allowed economists to conduct “systematic reviews” and even statistical “meta-analyses” to synthesise the findings from multiple studies and start to form coherent messages about what works. While this literature is not restricted exclusively to the unemployed – unlike the ALMP measures considered in this chapter – the results provide a useful starting point for any new study into the effectiveness of training.

The literature suggests that training has a positive impact on individuals’ labour market outcomes – including employment chances and earnings – on average, but there is wide variation in the estimated effect sizes, with some individual studies finding no statistically significant impact from training at all. In one recent meta-analysis, Card, Kluve, and Weber (forthcoming[46]) found that training had positive and growing effects on labour market outcomes when combining the results from more than 200 studies. However, only around 35% of the training-relevant studies found that training had positive effects in the short-term (within a year), 54% found that training had positive effects in the medium term (one to two years), and 67% found that training had positive effects in the long term (two or more years). The authors highlight that the increasing effectiveness of training measures over time is consistent with there being lock-in effects, whereby individuals lower their search intensity while training is taking place, reducing the impact on labour market outcomes in the short-term (Calmfors, 1994[47]). Similarly, the meta-analysis undertaken by Vooren, Haelermans, Groot, and Maassen van den Brink (2018[48]) finds that training has positive impacts on employment likelihood at 6, 12, 24, and 36 months after programme start in terms of the point estimates. Yet the average effect size is not always statistically significant partly because the estimated treatment effects vary so much between the studies included.

Notwithstanding this variation in its effectiveness, training measures appear to have outperformed more direct employment measures such as employment subsidies and public works programmes. This difference holds across different meta-analyses and across different time horizons (Vooren et al., 2018[48]; Card, Kluve and Weber, forthcoming[46]). Nevertheless, this does not imply training is necessarily the best ALMP measure. Firstly, Kluve et al. (2016[49]) show that programmes promoting entrepreneurship generally outperform training, at least for young people, although sample selection should be borne in mind here: such programmes are likely to try and seek out individuals that display some entrepreneurial ability or motivation. Secondly, integrating multiple types of programmes appears to boost ALMPs’ chances of success (Kluve et al., 2016[49]). This way, ALMPs can respond more directly to individuals’ needs, which may change throughout their interactions with policy makers.

Although the treatment effects of training programmes appear to be positive on average, it is less clear that such programmes pass simple cost-benefit analyses (Blattman and Raltson, 2015[50]). In principle, it is useful to set off the overall gains in terms of employment or earnings against the operating costs of the program, the costs of the education or training itself, forgone earnings, and any out-of-pocket expenses the training participant incurs to attend (such as transport or childcare). There may also be further hard-to-measure costs, such as the leisure time that participants forego to attend a training and the possibility that some existing workers are actually displaced by trained individuals, in light of their new skills (Heckman, LaLonde and Smith, 1999[51]). This chapter focuses primarily on the individual-level benefits accruing from Latvia’s training programmes for the unemployed. Yet, decisions about future directions for ALMP in Latvia cannot be taken without consideration of costs.

Training participants differ from non-participants along key observable characteristics

There may still be systematic differences between trained and untrained individuals, even when making comparisons among individuals that have endured a set number of months in unemployment. If these differences are captured by characteristics that are observed in the data, then it is possible to control for them in the analysis, to reduce the chances that the estimates of the effects of training are biased. Fortunately, the data contain detailed information on individual characteristics, on individuals’ labour market engagement, and on individuals’ interactions with and receipts of assistance from different arms of the government in Latvia (including the SEA, the Social Insurance Agency, and any social assistance coming from the municipalities). The full set of sample characteristics for individuals that do and do not receive training is shown in Annex Table 3.B.1.

In terms of individual characteristics, women, those who have completed some form of upper secondary education, married individuals, and those with children are more likely to participate in both formal and non-formal training than the population of unemployed people at large, although the patterns in terms of ethnicity, citizenship, and location are less clear-cut. Ethnic Latvians, Latvian citizens, and those educated in Latvian are more likely to participate in formal training but less likely to participate in non-formal training than other unemployed individuals. Around 31% of those participating in formal training first (and only participating in one training) were not ethnic Latvians, whereas just over half of those participating in non-formal training first (and only participating in one training) were not ethnic Latvians. This may reflect the fact that many non-formal trainings (around one quarter) focus on teaching the Latvian language, which may be less relevant for ethnic Latvians, Latvian citizens, and those educated in Latvian. In any case, non-formal training recipients – at least those for whom non-formal training is their only substantive ALMP measure – come disproportionately from urban areas, including Riga.

In terms of previous labour market engagement, individuals for whom formal or non-formal training was the first substantive ALMP measure in which they participated are more likely to have been employed directly before their registering as unemployed. This implies that trained individuals may be more connected to the labour market than other unemployed individuals. Nevertheless, this information should be treated with some caution, because the data do not allow every individual’s full employment histories to be recovered.

Since the proportion of people who are classified as disabled and who receive social assistance is relatively low, it is difficult to ascertain whether recipients of benefits (besides unemployment benefit) have a higher chance of receiving training. However, the proportion of people who received social assistance and who were classified as disabled as they entered unemployment is significantly higher among those individuals that participated in training after receiving another substantive ALMP measure.

Simply comparing the ever trained with the never trained portrays training programmes in a negative light

Directly comparing training participants with all non-participants shows that it typically takes trained individuals longer to return to employment, on average (Figure 3.12). For those individuals that received just one formal or non-formal training, average unemployment spells lasted 340 and 355 days respectively. By contrast, among those individuals who did not participate in trainings – but including those that potentially participated in other substantive ALMPs such as employment measures – average unemployment spells lasted around half that time (177 days). Similarly, it takes well over a year (465 days) into individuals’ unemployment spells before the hazard rate – the likelihood that an individual will exit unemployment to employment at a given moment – or trained individuals reach the same level as the hazard rate for those individuals not participating in training.25 These results underline the logic discussed in the sub-section outlining the econometric approach: it takes individuals a certain amount of time during their unemployment spell to begin training courses and they then become locked-in (they reduce their job search effort) when participation in training actually starts. This makes it difficult to estimate the true effects of training by simply comparing participants and non-participants.

Figure 3.12. Spell length and hazard rates among trained and untrained individuals

Controlling for observable characteristics does not alter the conclusion that untrained individuals find jobs faster than trained individuals on average, if the fact that individuals select into training at different times throughout their unemployment spell is not explicitly taken into account. To test this, individuals’ employment statuses at 6, 12, 18, 24, 30, and 36 months after the start of their unemployment spell are regressed on their participation in training as well as controls for individual characteristics, previous employment status, disability status and receipt of social assistance, as well as region, SEA branch, and month of registration fixed effects.26 These regressions suggest that, even at the 36-month mark, individuals that received formal training have only just managed to catch up to untrained individuals in terms of their likelihood of being employed, whilst those individuals that received non-formal training remain less likely than untrained individuals to be in employment (Annex Table 3.B.2). The results for earnings at 6, 12, 18, 24, 30, and 36 months after registration paint a similarly negative picture of the effects of training programmes.

Using appropriate econometric techniques, both formal and non-formal trainings have long-lasting positive impacts on employment

The dynamic selection-on-observables approach suggests that formal trainings generate positive and statistically significant effects on individuals’ chances of being in employment relatively quickly. As Figure 3.13 shows, 12 months after the start of formal training, individuals who began training (the intervention group) were almost 7.6 percentage points more likely to be in employment than those who were still “waiting” for a substantive ALMP measure or another way out of unemployment (the comparison group, see Box 3.2 for more details of the econometric approach). The effects remained positive for several years: 36 months after the start of the training, individuals who began training were still 6.7 percentage points more likely to be employed than individuals who were still waiting. Nevertheless, the positive effects of formal training did not appear immediately, as the employment and earnings results at the six-month mark demonstrate. This is consistent with the fact that formal trainings typically take several months to complete. Individuals’ capacity to search for new jobs may be curtailed while they are participating in training and even if job offers do arrive, they may prefer to make sure they complete the training and become accredited.

Figure 3.13. Estimated effects of trainings on employment and earnings

The employment effects of non-formal training were smaller than for formal training, but the effects emerged quicker and still persisted several years after the start of the training. Individuals’ chances of employment increased by 2.6 percentage points just 6 months after the start of non-formal training (compared to those still waiting for a substantive ALMP measure or another way out of unemployment), with these impacts rising to 5.0 percentage points after 12 months and 4.8 percentage points after 36 months (see Figure 3.13). All these effects are statistically significant at the 5% level. The differences between the employment effects of formal and non-formal training were also statistically significant at the 5% level across all the time horizons: such differences were not purely down to random chance.27 The finding that the effects of non-formal trainings on employment appeared more quickly, but were not as sizeable as the analogous effects of formal training, echoes the previous results of Hazans and Dmitrijeva (2013[38]). Part of the difference between the effects of formal and non-formal training may be down to differences in their target groups rather than differences in their effectiveness per se. For example, as Annex Table 3.B.1 shows, formal training recipients are more likely to be female, ethnically Latvian, and living in rural areas than non-formal training recipients.

The effects on earnings appeared to be stronger, emerge more quickly, and last longer for non-formal trainings than for formal trainings. At 18 months after the start of training, non-formal training produced an increase in monthly earnings of 5.8%, while formal training produced an increase in monthly earnings of just 2.2%. After 12 months, there were no statistically significant effects from formal training, while non-formal training’s effects were already starting to emerge. Additionally, 36 months after training began, the effects of non-formal training remained similar to what they were at the 12-month mark, whereas the effects of formal training were no longer statistically significant. The difference in the point estimates between formal and non-formal training were statistically significant at the 5% level at the 6-, 18-, and 36-month marks. This evidence somewhat contradicts Hazans and Dmitrijeva’s (2013[38]) previous results, as their study suggests that the earnings effects are weaker – and even negative – for non-formal training, whilst being positive for formal training. Nevertheless, it is important to recognise the impact that sample selection has on these estimates, as it is only individuals that find jobs for whom earnings data are available. Indeed, the relatively strong effects of formal training on employment chances may actually bring additional individuals with low earnings potential into the sample of employed people who otherwise would not have been observed. This may dampen the estimated effects of formal training on earnings.

The impact of training differs for particular sub-groups of unemployed people

In order to understand the overall effects of training more fully, it is helpful to examine how the results differ for particular sub-groups of the population. In this vein, this sub-section decomposes the overall effects of formal and non-formal training according to: (1) gender; (2) age; (3) rural versus urban; (4) high- versus low-skilled; and (5) receipts of social assistance. This sub-section focusses primarily on differences in the point estimates between different sub-groups. However, we also formally test whether any of the differences observed in the point estimates are statistically significant by running a “fully interacted” model, in which a dummy variable for the relevant sub-group is added into each regression and that dummy variable is interacted with all of the control variables in that regression.28 The relevant charts are reserved for Annex 3.A. The links between the sub-group results in this sub-section and the existing literature are discussed in Box 3.4.

The point estimates alone suggest that women appear to benefit more than men from formal training – at least over a long time horizon – but men benefit more than women from non-formal training, although these differences are only marginally statistically significant (Annex Figure 3.A.3). At 30 months after the start of formal training, the point estimates for the effects on women’s and men’s employment chances were 7.2 percentage points and 4.8 percentage points respectively, although this difference was not statistically significant at the 5% level (p-value=0.1150). At 18 months after the start of non-formal training, women who began training experienced a 4.0 percentage point increase in the likelihood of employment, compared with a 5.7 percentage point increase for men. Again, however, the difference between these point estimates for women and men were not statistically significant at the 5% level (p-value=0.1480). Nevertheless, while these differences are only marginally statistically significant, the story emerging from the point estimates resonates with the evidence in Annex Table 3.B.1 that women’s participation is disproportionately high in formal training, but not non-formal training. Women participate more in the trainings that have a higher relative effect on their employment chances.

Workers aged more than 30 years old experienced stronger positive effects from formal training on employment than younger workers, on average (Annex Figure 3.A.4). At the 18-month mark, the differences between the effects of formal training on workers aged more than 30 years old and younger workers were not statistically significant at the 5% level, but the differences were statistically significant over longer time horizons. At 24 months after the start of formal training, workers aged more than 30 years old who began formal training experienced an 8.7 percentage point increase in their chances of being in employment, while younger workers experienced an increase of just 4.8 percentage points, the difference between these two point estimates was statistically significant at the 5% level (p=0.0004). In contrast, the point estimates for the effects of non-formal training on employment chances were only slightly larger for workers aged more than 30 years old (18 or more months after the start of training), and the differences between these pairs of point estimates were not statistically significant at the 5% level. For earnings, workers aged more than 30 years old appeared to get more of a boost from formal training while younger workers benefited more from non-formal training, although these results – even looking solely at the point estimates – are not clear-cut. The relative success of formal trainings for those aged more than 30 years old – at least in terms of employment chances – may arise because formal trainings explicitly seek to build specific skills among those workers whose skills are being demanded less and less by employers. This may be more important for those who completed formal education a long time ago, prior to any of the reforms to VET and tertiary education discussed above. Additionally, young people are increasingly being channelled towards different types of training outside the auspices of the SEA, including longer programmes lasting up to one and a half years organised by the MoES. In the more recent data, this may change the composition of young people that remain available to actually participate in the formal and non-formal trainings for the unemployed on which this chapter focusses.

Rural dwellers seemed to benefit more than urban dwellers in terms of: (1) the employment effects of both formal and non-formal trainings (although these differences were not statistically significant) and (2) the earnings effects of non-formal training (Annex Figure 3.A.5). Looking at the point estimates alone, formal training increased the chances of employment for rural and urban dwellers by 7.2 and 7.8 percentage points respectively 12 months after training starts, but by the 36-month mark, formal training increased employment chances by 5.0 percentage points in urban areas and 7.4 percentage points in rural areas. Similar patterns emerged for non-formal trainings. The earnings effects of non-formal trainings were also larger in rural areas than in urban areas. At 18 months after the start of training, rural dwellers who found a job experienced a 7.8% increase in earnings, while urban dwellers who found a job experienced a 4.5% increase in earnings. The difference between these two point estimates was significant at the 5% level (p-value=0.0470).

Low-skilled individuals (those with up to lower secondary education) benefited more from formal trainings, especially in terms of earnings, than high-skilled individuals (Annex Figure 3.A.6). At 18 months after the start of formal training, low-skilled individuals that found work experienced a 4.6% increase in their monthly earnings while high-skilled individuals experienced virtually no increase. The difference between these two point estimates was significant at the 5% level (p=0.0462). It should be emphasised that, while there were no clear earnings effects for high-skilled individuals from formal training, they did benefit in terms of their employment chances: the emerging story that virtually all sub-groups have something to gain from training for the unemployed remains intact. The differences in the point estimates between low- and high-skilled individuals in terms of formal and non-formal trainings’ employment effects are largest between 12 and 24 months after the start of the training. After the 24-month mark, however, the gap between low- and high-skilled workers in terms of trainings’ employment effects largely closes.

The employment effects of formal training were (eventually) stronger on recipients of social assistance, but this is not the case for non-formal trainings, and these estimates are somewhat constrained by sample size (Annex Figure 3.A.7). Rather than splitting the sample into those individuals that were and were not receiving social assistance at the start of their unemployment spell, the sample is instead split into those individuals that did and did not receive social assistance at any point during their unemployment spell. Splitting the sample in this way ensures there is a sufficient number of social assistance recipients for the analysis. However, some caution should be exercised when interpreting these results: individuals’ chances of receiving social assistance during their unemployment spell may depend on how long that unemployment spell lasts, which is precisely what the provision of training is trying to affect. While the effects of formal trainings on employment chances do not seem to depend on individuals’ social assistance status at the 18-month mark, 36 months after the start of training social assistance recipients experienced a 10.3 percentage point increase in the likelihood of being employed compared with a 5.4 percentage point increase for non-recipients. The difference between these two point estimates was significant at the 5% level (p=0.0038). This suggests that individuals from poorer households stand to gain the most from learning new, specific skills. However, the positive employment effects arising from non-formal training appear, if anything, to be slightly stronger among those that did not receive social assistance.

Understanding how training specifically affects the long-term unemployed would be a useful complement to this analysis, but assessing trainings’ relative effects on the long-term unemployed is difficult for two key reasons. Firstly, and most fundamentally, unemployed individuals’ transition to employment is the most important outcome variable on which the analysis focusses. It is not possible to simply separate out those individuals who reached more than 12 months in unemployment: the time spent in unemployment is something that training is explicitly seeking to change. One possibility would be to look at individuals that spent a certain amount of time in unemployment before participating in training. That is, the analysis could focus on those individuals who could be classified as long(er)-term unemployed before training began. This, however, leads to the second challenge associated with assessing the long-term unemployed: sample size. Relatively few individuals have to wait more than 12 months into their unemployment spell to begin participation in a substantive ALMP measure, especially for non-formal training. Therefore, this chapter focusses on sub-groups that can be identified at the moment an individual registers with the SEA (as above). The propensity of these particular sub-groups to end up in long-term unemployment is then discussed in more detail in Chapter 5.

Notwithstanding the differences in the treatment effects described above, formal and non-formal trainings appear to have at least some positive and statistically significant impact on the employment chances of the majority of the sub-groups covered. This potentially demonstrates the adaptability of Latvia’s ALMP training measures. Thus, while there may be gains from targeting training programmes according to their estimated impact in order to maximise the benefit to society, such programmes appear to improve the employment and earnings outcomes of many different types of unemployed people. This gives the SEA a certain amount of choice over its targeting approach.

Box 3.4. Heterogeneous effects of training in the existing literature

Linking the sub-group analysis presented in this chapter with sub-group analysis in the existing literature faces two main challenges. First, most existing meta-analyses and systematic reviews consider how programme-level treatment effects differ, rather than considering whether treatment effects differ for certain sub-groups within in a given programme. For example, rather than comparing women and men treated by mixed-gender programmes, such meta-analyses and systematic reviews compare mixed-gender programmes, with all-women and all-men programmes. Second, the sub-groups on which previous meta-analyses – and individual studies – have focussed do not necessarily match the sub-groups on which this chapter focusses. For example, it is rare for studies to break down the results according to social assistance receipts, as in this chapter.

Despite these challenges, some comparisons between the sub-group results in this chapter and the existing literature are possible: participant gender has been a special focus of many previous studies. There is some limited evidence suggesting that training may be more effective for women than men, but much of the existing literature finds any gender differences to be small and not statistically significant. At the programme level, the updated analysis by Card et al. (forthcoming[46]), has found that female-only training (and other ALMP) programmes outperform male-only and mixed programmes, but this finding is not replicated in similar studies by Kluve et al. (2016[49]) and Vooren et al. (2018[48]). Similarly, looking at differential effects by gender within a collection of mixed-gender training programmes in developing countries, McKenzie (2017[52]) finds no clear evidence that women benefit more than men. Nevertheless, previous work on training in Latvia has found similar results to this chapter. In particular, Hazans and Dmitrijeva (2013[38]) show that women experience stronger employment effects from formal training, but men experience stronger employment effects from non-formal training.

Turning to other sub-groups, the evidence on whether younger or older workers benefit more from ALMP measures is somewhat mixed. In a very recent study, Vooren et al. (2018[48]) find that the maximum age of programme participants has no impact on ALMP measures’ effectiveness, just as is observed for non-formal training in this chapter, drawing on a range of studies from a range of OECD and non-OECD countries both within and outside of Europe. However, previous evidence from Kluve (2010[53]) – which focusses solely on studies conducted in Europe – finds that ALMP measures targeting young people are less likely to be effective. Interestingly, Card et al. (forthcoming[46]) demonstrate that, if anything, mixed-age ALMP measures outperform those targeting either young people or older people.

While this chapter does not speak directly to the question of training’s effectiveness for the long-term unemployed, Card et al. (forthcoming[46]) find that the impacts are larger for those programmes that explicitly target those in long-term unemployment. While Card et al.’s estimates are at the programme level rather than the individual level, tentatively, their findings suggest that tailoring the content of the training for long-term unemployed individuals (even after assignment to a training programme) may boost impact.

The effects of training are sensitive to how training is combined with other ALMP measures

The analysis now explores three different ways of assessing the sensitivity of training’s effects to being combined with other ALMP measures, building on the descriptive statistics presented above. First, the analysis decomposes the estimated effects of training into the effects for those who participated in training only and those who went on to participate in other substantive ALMP measures, including employment measures and additional formal or non-formal trainings. Second, the analysis considers how combining training with mobility support alters the impact on individuals’ chances of employment and earnings. Thirdly, the analysis investigates how providing MICs alongside training – either before or after training begins – influences training’s effects. The relevant charts are shown in Annex 3.A.

Trained individuals that go on to participate in other substantive ALMP measures fare worse than both those who participate in just one formal or non-formal training and, for some time after the start of training, those who do not begin training at all (see Annex Figure 3.A.8). These results are, however, unsurprising, insofar as it is primarily those individuals that remain in unemployment after their first training, who are most likely to participate in additional substantive ALMP measures. Subsequently, such individuals may become locked-in to these additional substantive ALMP measures. As such, these individuals are likely to have worse employment outcomes after the start of training by construction. Indeed, this is exactly the same issue that motivated the use of the dynamic selection-on-observables approach to evaluate the first formal or non-formal training that individuals undertook in the main analysis in the previous two sub-sections.

Receiving mobility support appears to boost the employment effects and earnings effects of both formal and non-formal trainings (see Annex Figure 3.A.9). This finding arises by separating those individuals who began receiving training and mobility support simultaneously from all other training participants.29 Those who began receiving mobility support after their formal or non-formal training started are not classified as joint training and mobility participants for this analysis: in these instances, it is less likely that the mobility support is being explicitly provided to support training itself, instead supporting subsequent efforts to find work. The differences between those trained individuals that did and did not receive mobility support are clearest several years after the start of training. For both formal and non-formal training, the point estimates at the 36-month mark are higher for those who received mobility support alongside training, although the point estimates for the with-mobility support group are not statistically significant at the 5% level. The differences between those receiving their training with and without mobility support are stronger and clearer in terms of earnings. One particularly striking result is that, 18 months after training start, those individuals who began receiving non-formal training and mobility support (and who were in work) experienced a 23.1% boost in monthly earnings, compared with a 5.5% boost for those who started non-formal training only. The difference between these two point estimates is significant at the 5% level (p-value=0.0213). However, the same caveats regarding sample selection on the earnings results should again be borne in mind.

Separating the results for mobility support recipients in this way suggests that ALMP measures may have complementary effects on unemployed individuals’ labour market outcomes. Some individuals living in remote and rural areas may need mobility support to reach training providers. More fundamentally, providing mobility support may improve the match between the training provider and the trainee: this is one of the key aims associated with providing training through a voucher system. Nevertheless, one key caveat should be borne in mind when interpreting the results. Since individuals have to apply for mobility support and looking for training providers further afield requires a certain level of effort, those who end up receiving mobility support may be more motivated than those who do not. This may inflate the apparent boost to training’s employment and earnings effects offered by mobility support.

Receiving MICs before training starts may increase the employment and earnings effects of both formal and non-formal training (Annex Figure 3.A.10). Since MICs are so widespread, it is possible to separate the effects of training for those who only received MICs before their formal or non-formal training, those who received some MICs after their formal or non-formal training, and those who received no MICs at all. As expected, individuals receiving MICs after training fared worse than those who received no MICs at all. Again, this is because it is only those individuals that remain in unemployment longer who become eligible for additional ALMP measures after training (even if those additional ALMP measures are short, like MICs). However, receiving MICs before training begins appeared to boost slightly the impact on employment and earnings. For example, 18 months after the start of formal training, those individuals who began training but had previously received no MICs experienced a 10.8 percentage point increase in employment chances, while those individuals who began training and had received one or more MICs beforehand (but no MICs after) experienced an 11.9 percentage point increase in employment chances. However, these differences in employment effects from formal training are not statistically significant at the 5% level over any time horizon. Looking again at the 18-month mark, those individuals who began formal training but had previously received no MICs experienced virtually no change in earnings, while those individuals who began training and received one or more MICs beforehand (but no MICs after) experienced a 4.0% increase in earnings. While this pair of point estimates from the 18-month time horizon are not statistically significantly different from one another at the 5% level, there was a statistically significant difference between the point estimates at the 30-month mark. The analogous differences for non-formal trainings were if anything slightly clearer. For employment, there were statistically significant differences between the point estimates (at the 5% level) at the 24- and 30-month time horizons. For earnings, the differences between the point estimates were statistically significant (although only at the 10% level) at the 6- and 18-month time horizons. The potential boost that MICs offer to the effectiveness of training must be borne in mind for the ongoing reforms to MICs. If MICs are reduced in number and bundled into non-formal trainings too much, formal training participants may not receive MICs and hence may miss out on their potential benefits.

Implementation of training programmes in Latvia

This section considers how training programmes for the unemployed in Latvia are implemented, and what that means for their effectiveness. The section focuses in particular on the implications of providing training through vouchers, as has been the case in Latvia since 2011. The section begins by broadly defining what voucher systems look like and then outlining some of their key theoretical advantages. The section then focuses more directly on Latvia, moving to discuss some of the risks associated with providing training through vouchers and, in turn, what the government may do to mitigate them.

Training is provided through a voucher system in Latvia

Voucher systems can be applied to many different components of policy makers’ involvement in training provision. The definition of what constitutes a “voucher” may be fairly broad, although typically they have the following properties: (1) vouchers are rendered in a written, digital, or other format aside from cash, (2) vouchers entitle the recipient to a subsidy or discount, therefore carrying a money-equivalent price, (3) vouchers are redeemable for a good or service that holds a price in the market (Tomini, Groot and Maassen van den Brink, 2016[54]). As such, vouchers can be dispersed for several steps of training provision, including the assessment of individuals’ needs, the training itself, and placement in a job after the training. An extreme application of voucher systems would disperse vouchers for all of these steps. At the other end of the spectrum, for structures like the military, governments typically provide all steps of the training themselves, including recruitment, eligibility determination, assessment, assignment to a specific training programme, provision of the training itself, and subsequent placement (Barnow, 2009[55]).

In Latvia, vouchers are provided only for the provision of training itself. Other steps, such as assessment of the unemployed individuals’ needs, are undertaken directly by SEA caseworkers. The vouchers consist of a physical document, which is received at the branch offices of the SEA. The voucher itself contains various information about the conditions under which it can be redeemed and cancelled, directly informing the recipient of their responsibilities.

Vouchers may give training participants more choice and improve provision

One of the main motivations for providing training through vouchers is that doing so gives participants more choice over the specific types of training that they do and the institutions that provide that training. In addition to “more choice” being a valuable end in itself, giving voucher recipients choice may also improve the match between their needs and the training that is actually provided (Hidalgo, Oosterbeek and Webbink, 2014[56]). Aligning trainings to participants’ preferences is especially important in contexts where training is primarily provided on the job. Firms may be unwilling to provide training on “general” skills to individuals (which may be used outside the firm), focussing instead on “specific” skills that can only be applied within the firm, regardless of what is best for the individual or for society at large (Becker, 1975[57]).30 However, even for unemployed individuals that have no direct pre-existing association with a firm – as is the case for the formal and non-formal trainings in Latvia on which this chapter focusses – vouchers may still improve the alignment between individuals’ needs and the training provided.

Voucher systems may also improve the quality and performance of training providers. Cross-country evidence suggests that giving providers incentives improves ALMP participants’ outcomes. For example, Kluve et al. (2016[49]) show that having some kind of incentive system for providers moderately improves the performance of ALMP measures, although their systematic review focusses primarily on young people and covers all ALMP measures rather than just training. Voucher systems incentivise vendors to provide high quality training, as doing so enables them to attract voucher recipients and increase their profits. If the quality or relevance of the training provided is low, voucher recipients can “vote with their feet” and find preferable alternatives. Nevertheless, there may be practical limits to the extent of competition between providers, which are discussed in more detail in the following sub-sections.

An additional benefit of vouchers is that they potentially simplify the process of providing training for the public employment service and for the government at large. This comes by transferring a certain amount of responsibility for many of the steps described above – including assessment, enrolment, and provision of the training itself – to voucher recipients and/or to private providers. Indeed, even if training is already provided privately, a voucher system allows the government to incentivise providers without directly contracting out training to service providers. Such direct contracting could potentially involve complex tender processes and regulation of prime providers or sub-contractors, as has been the case in the “quasi-market” created to implement ALMPs in the United Kingdom (OECD, 2014[58]). Nevertheless, as the following sub-sections demonstrate, governments still have an important role to play in ensuring that voucher systems, once set up, operate effectively.

In Latvia, transparency was another substantial motivation for allocating training through vouchers. Before the introduction of the voucher system in 2011, Latvia experienced several notable examples of training providers procuring contracts that lasted a very long time, which were difficult to revise or even terminate in response to performance. Consequently, the quality of the training from some providers deteriorated throughout the duration of the contract. By placing responsibility for selecting training providers in the hands of voucher recipients, the voucher system sought to make the mechanism for allocating training more lucid.

Voucher recipients require information

The success of voucher systems hinges on governments cultivating the right conditions for their success, and providing information to voucher recipients is one especially important way in which the government can help. The main information that voucher recipients require is on the relative success of different providers in building skills and placing training participants in good jobs. Without such information, there is no mechanism for voucher recipients to find providers that are a suitable match, nor will the market promote high-quality providers at the expense of low-quality ones. Typically, governments can supply voucher recipients with descriptive monitoring data on training providers. However, supplying rigorous evaluation results at the provider level is not normally possible. Additionally, voucher recipients may also benefit from information about current and forecast labour demand – and hence wages – at the occupation level. This allows them to set career goals that increase their chances of gaining employment and maximising their earnings.

In Latvia, the SEA collects and disburses relatively detailed monitoring information for each training provider. After completing a training programme, participants fill in a special evaluation sheet, which allows them to describe their experience and report their employment status 6 months after the training finishes. This information is then made available online and at local SEA branch offices and new voucher recipients are directed towards these information sources. Short-term labour market forecasts are also made available to prospective training participants to help inform their choices. However, providing information alone may not be sufficient for supporting effective choice. Reading, absorbing, and interpreting such vast quantities of information may be difficult, especially for individuals with low motivation and who may not have a good understanding of their own potential. As such, caseworker guidance may be needed to further support voucher recipients, especially those from disadvantaged groups, as discussed in more detail below.

Such is the importance of information for ensuring that voucher systems operate effectively, governments may even ask voucher recipients to demonstrate a certain level of knowledge regarding their decision before any training actually takes place (Kaplan et al., 2015[59]). However, no such testing of voucher recipients’ knowledge is implemented in Latvia.

In Latvia, the SEA also plays a role in ensuring voucher recipients do not encounter misleading information on training providers. In this vein, aggressive marketing techniques by training providers are banned, to voucher recipients make effective and well-informed choices.

Partly restricting voucher recipients’ options may better align their motives with those of the government

Aligning the incentives of voucher recipients with the motives of the government also presents a key challenge for establishing an effective voucher system.31 For one, voucher recipients may be more inclined to pursue trainings that carry higher “consumption” value rather than investing in their human capital per se: they may select trainings that they perceive to be more enjoyable (Barnow, 2009[55]). It may also be that the time horizons of the government and of voucher recipients are misaligned. Governments may prefer voucher recipients to choose trainings that enable them to return to work (ceasing benefit payments) and reintegrate into the labour market quickly. The voucher recipients themselves, however, may instead prefer to focus on boosting their potential earnings power, even if this takes time. Indeed, evidence from the German system suggests that – while differences are difficult to detect – any benefits in terms of voucher recipients’ employment and earnings outcomes relative to those trained through mandatory assignment, often take several years to emerge (Strittmatter, 2016[60]).

To ensure that the choices made by voucher recipients align with government incentives, the Latvian SEA restricts the way that training vouchers can be used in two main ways. First, the Training Commission – under the auspices of the Ministry of Welfare – meets at least once each year to decide the fields of study for which vouchers are redeemable. This ensures that vouchers are tilted towards occupations in which growth in labour demand is forecast to outstrip growth in labour supply, to ensure that there are sufficient high-paying vacancies. Second, the SEA restricts, or at least guides, the set of educational institutions at which vouchers can be redeemed. Only pre-approved training providers known as SEA “partners” are listed on the SEA website. These SEA partners are either approved by the SEA themselves – through a rigorous registration, accreditation, and licensing process – or by another more relevant organisation. For example, the Road Traffic Safety Directorate or Ceļu satiksmes drošības direkcija (CSDD) accredits and licenses driving schools. The SEA also coordinates with the State Education Quality Service or Izglītības kvalitātes valsts dienest (IKVD), the body that is responsible for assuring the quality of education, when approving SEA partners. As discussed above, the SEA is currently making the criteria for becoming an SEA partner even stricter. In principle, voucher recipients are able to choose an accredited educational institution independently if desired, but SEA partners are likely to be easier to find and enrol in.

Nevertheless, evidence from other countries suggests that creating and maintaining approved lists of training providers may be difficult in practice. Public employment service workers in the United States – where voucher systems have long been used to provide training – report that establishing such lists is burdensome, especially in terms of the intense data collection required to monitor training provider quality (Barnow and King, 2005[61]). Indeed, the SEA devotes substantial resources to collecting relevant data, monitoring training providers, and publishing information about training providers online. The SEA not only explicitly publishes the requirements for becoming an SEA partner on its website, but also disseminates the results of a specific quality performance system that tracks training providers in terms of participants’ subsequent employment outcomes. Moreover, one particular challenge in Latvia has been finding ways to deal with training providers that ostensibly meet all the criteria but where the training outcomes or participant perceptions are not positive.

Disadvantaged groups may need help exercising effective choice

Not all voucher recipients will be able to exercise choice effectively, meaning that voucher systems can amplify existing inequalities among individuals receiving ALMP measures. Redeeming vouchers relies on recipients being sufficiently motivated to find themselves a suitable provider, enrol, and then stay in the training, potentially without much guidance or supervision from the public employment service. Some individuals may be more able to deal with these “hassle factors” than others (Babcock et al., 2012[62]). Indeed, this is often cited as a potential reason for the widespread observation – coming from many contexts in both Europe and in the United States – that low-skilled workers are less likely to redeem their vouchers than high-skilled workers (Barnow, 2009[55]; Kruppe, 2009[63]). Additionally, disadvantaged workers, including those from low-income households, may be more susceptible to form unrealistic expectations about their prospects in the labour market, using their vouchers to pursue trainings that are inappropriate and overly ambitious (Dickinson and West, 1983[64]). As Bruttel (2005[65]) notes, based on the experience in Germany, higher-skilled individuals are better able to “articulate” their training needs to caseworkers and are thus matched to more appropriate training, even under a system where the public employment service assigns people mandatorily to training. However, these inequalities may be exacerbated when yet more responsibility is given to potential training participants, as is the case in a voucher system.

Since not all vouchers are redeemed in Latvia, one potential way to detect inequalities among voucher recipients is to verify whether certain individual characteristics influence the likelihood of voucher redemption. Approximately 77% of the vouchers received were used to start training, leaving the redemption rate in the Latvian system comparable to Germany but somewhat lower than the United States (Huber, Lechner and Strittmatter, 2015[66]; Strittmatter, 2016[60]). Some instances of non-redemption may arise from hard constraints on finding suitable training in remote and rural municipalities, as discussed in the following sub-section: training programmes will not start if registration for such programmes is too low and people in such areas may not be mobile enough to reach suitable training sites. Other instances of non-redemption may arise from individual expectations and levels of motivation, which may be correlated with observed characteristics such as age, gender, and educational level. In principle, the SEA can sanction those who fail to redeem their vouchers: if a client twice fails to realise their voucher with no good justification, their assignment to training may be cancelled. In practice, however, it appears that sanctions for non-redemption are rarely applied, although there are no specific administrative data on this issue. In principle, non-redemption may also arise if certain individuals are able to find a job while waiting for training to start, although given that vouchers are only valid for two weeks on average, this phenomenon is not likely to be prevalent.

Regressing voucher redemption on a range of individual characteristics, language abilities and age have the largest effects on the likelihood that voucher recipients actually redeem their vouchers, but many individual characteristics have only very small effects (Table 3.2 and Annex Table 3.B.3). Individuals with at least basic Latvian language skills are 7 percentage points more likely to redeem their vouchers, even when controlling for gender, age, education and a host of other individual characteristics. Additionally, being one year older increases the likelihood of redemption by between 0.1 and 0.2 percentage points on average, which is a small effect in itself but adds up over individuals’ life cycles. Indeed, those aged more than 55 years old are 7 percentage points more likely to redeem their vouchers than those aged 15-24 years old. Nevertheless, there may be other factors, besides their being less able to make effective choices, which influence rates of voucher redemption among young people. In particular, young people may be channelled towards different types of training, either within the framework of the voucher system or outside the voucher system (including longer programmes lasting up to one and a half years organised by the MoES).

Table 3.2. Regressions of voucher redemption on individual characteristics
	(1)	(2)	(3)	(4)	(5)
Male? (1=Y; 0=N)	0.0087**	0.0085**	0.0083**	0.0080**	0.0049
	(0.0037)	(0.0037)	(0.0038)	(0.0038)	(0.0037)
Married? (1=Y; 0=N)	0.0078**	0.0076**	0.0071**	0.0071**	0.0064*
	(0.0031)	(0.0031)	(0.0034)	(0.0034)	(0.0035)
Has children? (1=Y; 0=N)	0.0060**	0.0062**	0.0066**	0.0067**	0.0031
	(0.0028)	(0.0029)	(0.0029)	(0.0029)	(0.0042)
Age (years)	0.0016***	0.0017***	0.0017***	0.0017***	0.0012***
	(0.0004)	(0.0004)	(0.0004)	(0.0004)	(0.0002)
High-skilled? (1=Y; 0=N)	0.0108*	0.0108*	0.0103*	0.0104*	0.0087**
	(0.0058)	(0.0057)	(0.0052)	(0.0053)	(0.0041)
Latvian language at least basic? (1=Y; 0=N)	0.0653***	0.0646***	0.0643***	0.0644***	0.0693***
	(0.0190)	(0.0190)	(0.0188)	(0.0188)	(0.0205)
Received social assistance in January 2012? (1=Y; 0=N)			-0.0087	-0.0087	-0.0078
			(0.0113)	(0.0113)	(0.0106)
Disabled in January 2012? (1='Y; 0=N)			-0.0026	-0.0026	-0.0029
			(0.0050)	(0.0050)	(0.0049)
Time for which voucher is valid (days)				0.0238***	0.0232***
				(0.0033)	(0.0029)
Time for which voucher is valid squared				-0.0006***	-0.0006***
				(0.0001)	(0.0001)
Voucher for non-formal training? (1=Y; 0=N)					0.0362***
					(0.0131)
Youth Guarantee vouche?r (1=Y; 0=N)					-0.0193
					(0.0193)
Region fixed effects and urban dummy	No	Yes	Yes	Yes	Yes
N	51 925	51 925	51 925	51 925	51 925
R-squared	0.0605	0.0607	0.0607	0.0609	0.0626
Note: Data from 19 November 2015 to 31 October 2017 only. High-skilled refers to those individuals with more than lower secondary education.
Standard errors in parentheses.
Standard errors clustered at the level of the SEA branch office.
Dependent variable: Was voucher redeemed? (1=Y; 0=N).
SEA branch fixed effects in all specifications.
* p<0.10; p<0.05; * p<0.01.
Source: Latvian State Employment Agency (SEA), Latvian Social Insurance Agency and OECD estimates.
StatLink https://doi.org/10.1787/888933961296

On the other hand, many other individual characteristics appear to have little effect on the chances of voucher redemption. Even though the difference is statistically significant at the 5% level, low-skilled individuals (those with up to lower secondary education) are just 1 percentage point less likely to redeem their vouchers than high-skilled individuals. There are similarly small effects for gender, marital status, and having children.

In general, caseworkers in voucher systems face a difficult balance between supporting clients in using their vouchers and not overly interfering such that they effectively make choices on clients’ behalf. Getting this balance right is particularly important for potentially disadvantaged groups. On the one hand, caseworkers are likely to have a good understanding of clients’ needs given their regular interactions with clients and their experience of assigning clients to different ALMP measures. For example, in the Workforce Investment Act effected in the United States, policy makers explicitly experimented with the extent to which caseworkers could guide the choices of voucher recipients, by limiting voucher redemption to high-demand occupations, by screening vendors for quality in advance, and by ensuring decisions about how vouchers were used were taken jointly by caseworkers and voucher recipients. Extra guidance increased voucher recipients’ future earnings, although voucher recipients themselves appeared to prefer having fewer constraints on their choices (McConnell et al., 2006[67]). On the other hand, it may be difficult to reap the full benefits of having a voucher system – in terms of choice and competition – if caseworkers are too heavy-handed in their support, taking the assumption that voucher recipients know their needs best. In Germany, for example, caseworkers are not allowed to guide voucher recipients’ choices over training programmes, and can only restrict the set of options available by specifying educational goals, which are recorded on a (often ambiguous) handwritten note (Strittmatter, 2016[60]).

Providing training through vouchers rather than mandatory assignment also alters the decisions that caseworkers need to take at the assignment stage, so governments may need to adjust the guidelines given to caseworkers themselves. Caseworkers may assign vouchers in several ways, including: (1) trying to give vouchers to those who will experience the largest effects on their employment outcomes; (2) trying to give vouchers to those most in need or those who would fare least well if they did not receive a voucher; or (3) trying to give vouchers to those with the best post-training outcomes (sometimes known as “cream-skimming”) (Poeschel, 2014[68]). Switching from a system of mandatory assignment to a voucher system may change caseworkers’ approach, because they know that dispersal of certain types of vouchers to certain groups may lead to non-redemption or to potentially prolonged periods where individuals try to match with a suitable training course.

In Latvia, there are regular meetings between the SEA and voucher recipients, which may support effective choice especially among the potentially disadvantaged groups – those who may struggle to redeem their vouchers – identified above. The caseworker and the registered unemployed person agree an Individual Action Plan (IAP), which sets out potential pathways back to work. The agreement of the IAP, in itself, may give caseworkers some influence over the types of training that voucher recipients choose. Registered unemployed individuals and caseworkers are also obliged to meet every two months. This level of interaction between voucher recipients and the SEA helps ensure the former can make choices effectively.

As in other European countries, there are some limits over the extent to which caseworkers can influence vouchers recipients choices in Latvia, but caseworkers still have several key avenues for guiding those choices. While caseworkers cannot recommend specific training programmes at specific training providers, they are able to recommend particular occupations or types of training programmes from the full list of programmes for which a voucher is eligible. Caseworkers may also recommend additional services, including career consultations, to voucher recipients. This provides another latent channel through which caseworkers can guide voucher recipients’ choices. Nevertheless, throughout these interactions, caseworkers are expected to focus on supplying clients with objective information – including information about salaries and vacancies in different professions as well as information about training programmes – rather than simply their own subjective assessments of what trainings to choose.

Some remote and rural areas may require support to promote competition

The geographical spread of training providers in Latvia is uneven, meaning that voucher recipients in some municipalities may struggle to find training options locally. All other things equal, this may hamper competition and, in turn, any potential improvements in training quality that could be brought about by having a voucher system. Voucher recipients would find it harder to “vote with their feet” if training quality were low or the curricula were not well aligned with their needs. As discussed in detail below, the SEA is currently aiming to consolidate and reduce the number of training providers to ensure training is of high quality and to limit potential lock-in effects. This makes Latvia’s ongoing efforts to support regional mobility all the more important for promoting competition among training providers, especially in certain remote and rural areas.

Official data on training providers, which contain the full addresses of so-called SEA partners, may illustrate where there are geographical gaps in training provision. In principle, both formal and non-formal training voucher recipients can choose to go to accredited training providers, which are not SEA partners, but given that the SEA partners are advertised explicitly on the SEA website and in SEA branch offices, they are likely to be easier to find and access than other training providers. Thus, the data on SEA partners presented here serves as a useful proxy for the extent of training provision in each municipality. In Figure 3.14, Panel A shows the absolute number of sites at which SEA partners (providing either formal or non-formal training or both) were located in each municipality. Panel B then adjusts these figures by dividing the number of SEA partner sites by the number of unemployed people registered with the SEA (in October 2017) in that municipality. These figures are recreated, separating out those SEA partners providing only formal and only non-formal training in Annex Figure 3.A.11 and Annex Figure 3.A.12.

Three key messages emerge from the SEA’s data on its training partners. Firstly, there are large clusters of municipalities where there are no SEA partners at all, especially in the Kurzeme and Zemgale regions. Secondly, while cities tend to contain the largest absolute number of SEA partners, this is not the case when the figures are adjusted according to the number of unemployed people who are resident there. The municipality (or “republican city”) of Riga, for example, contains 38 SEA partners, the highest of any municipality in Latvia. However, Riga contains just 9.6 SEA partners per 1 000 registered unemployed people, ranking it 49^th out of the 119 municipalities in Latvia. Finally, there are relatively high numbers of SEA partners per municipality – both with and without the adjustment for the number of unemployed people – in eastern Latvia. Thus, while some municipalities in Kurzeme and Zemgale appear to have fewer proximate training providers, it is not universally the case that remote and rural areas totally lack training providers. However, the sheer size of some of the municipalities in the Latgale region should be taken into account: even if voucher recipients do not have to cross municipality boundaries to reach a SEA partner, they may still have to travel some way (and certainly more than the 15 kilometres required to be eligible for mobility support).

Figure 3.14. Number of all accredited training sites (SEA partner sites) by municipality in Latvia

The SEA is in the process of consolidating and reducing the number of providers of training for the unemployed, focussing training provision in locations with high levels of economic activity. This in line with the broader reforms to the VET system described above. As such, simply expanding the number of training providers does not appear to be a tenable strategy for promoting choice and competition.

The consolidation of training provision for the unemployed is driven not only by a desire to improve quality, but also to limit potential lock-in effects that may arise. Part of this consolidation effort involves making the selection criteria for becoming and remaining an SEA partner stricter, which has direct positive effects on training provider quality. For example, as of 2017/2018, providers must hold formal accreditation from the SEA (or some other relevant institution) for 6 years in order to become a SEA partner. SEA partners must also meet more stringent criteria in terms of (for example) teacher qualifications and access for disabled people: if training providers fail to meet these criteria, they risk having their status as a SEA partner suspended for up to a year. At the same time, training classes have often proved difficult to fill in certain remote and rural areas, which lengthens the time that voucher recipients have to wait before training can begin and in turn increases potential lock-in effects. Reducing the number of SEA partners means that classes should fill up quicker, alleviating this problem.

With the number of SEA partners declining, other strategies – such as the existing programme of mobility support – are likely to be especially important in municipalities where, even now, there are relatively few training providers. As the analysis demonstrates, voucher recipients in remote and rural areas in the Kurzeme and Zemgale regions are likely to be particularly dependent on mobility support if they are to exercise effective choice.

Regressing voucher redemption on location characteristics, it emerges that redemption is less likely in urban areas and in Riga in particular, and is most likely in the Latgale region (Table 3.3).32 This paints a nuanced picture of the link between local training options and voucher redemption. Amongst the regions outside Riga and Pieriga, Latgale has both the highest redemption rate and the highest concentration of SEA partner training sites. This is consistent with the intuitive notion that having better access to training options – the effects of mobility support notwithstanding – supports voucher redemption. However, the fact that redemption is lowest in urban areas (and especially Riga) runs counter to this notion. One possible explanation is that voucher recipients are more likely to receive job offers while their vouchers are valid in urban areas than in rural areas. This would allow them to exit unemployment without needing training and without needing to redeem their voucher. Nevertheless, vouchers are typically only valid for 14 days: it is not clear that this is long enough for the higher possibility of job offers in urban areas to have a substantive effect on redemption rates.

Table 3.3. Regressions of voucher redemption on location characteristics
	(1)	(2)	(3)
Urban? (1=Y; 0=N)	-0.0802**	-0.0320**
	(0.0307)	(0.0156)
Pieriga? (1=Y; 0=N)		0.0355	0.0633***
		(0.0274)	(0.0203)
Vidzeme? (1=Y; 0=N)		0.1141***	0.1425***
		(0.0223)	(0.0187)
Zemgale (1=Y; 0=N)		0.1071***	0.1280***
		(0.0338)	(0.0325)
Kurzeme (1=Y; 0=N)		0.1481***	0.1636***
		(0.0150)	(0.0165)
Latgale (1= Y; 0=N)		0.1857***	0.2042***
		(0.0191)	(0.0182)
N	51 925	51 925	51 925
R-squared	0.0152	0.0427	0.0418
Note: Data from 19 November 2015 to 31 October 2017 only.
Standard errors in parentheses.
Dependent variable: Was voucher redeemed? (1=Y; 0=N).
Standard errors clustered at the level of the SEA branch office.
All regressions contain full set of individual and voucher characteristics.
Base category for regions is Riga.
* p<0.10; p<0.05; * p<0.01.
Source: Latvian State Employment Agency (SEA), Latvian Social Insurance Agency data and OECD estimates.
StatLink https://doi.org/10.1787/888933961334

Voucher systems may compound lock-in effects

In some situations, providing training through vouchers may prolong the time that individuals spend locked-in to unemployment, even if they do not go on to redeem their voucher. Voucher systems may lead to additional lock-in time over and above systems involving mandatory assignment to training because voucher recipients require (and be afforded) time to find a training course that is suitable for them. In some systems, receipt of a voucher may also insulate recipients from loss of benefits or other sanctions and also from assignment to other onerous programmes to which caseworkers could designate them under a system of mandatory assignment (Strittmatter, 2016[60]). Relatedly, suggestive evidence from the reform of the German training system indicates that mandatory course assignment outperforms voucher systems during the first two years after the start of training (in terms of trainees’ employment and earnings outcomes), but after that voucher systems produce larger positive effects on labour market outcomes (Rinne, Uhlendorff and Zhao, 2013[69]). This emphasises a crucial point regarding lock-in effects: prolonging an individuals’ spell in unemployment may be justified if doing so ultimately improves their productivity (which the positive earnings results in the main analysis above suggest is the case in Latvia). Indeed, there may be a trade-off between getting unemployed people back into work quickly and building their skills for the long run. If providing training through vouchers improves the match and the quality of the training – even if it takes longer for the unemployed individual to receive this training – the additional lock-in effects may be worth it. Nevertheless, evidence from the same German reforms also suggests that individuals who receive a training voucher but do not go on to redeem it experience statistically significant negative effects on their employment chances – reaching a drop of around 5 percentage points – for up to three years after voucher receipt (Huber, Lechner and Strittmatter, 2015[66]). As such, the possibility of non-redemption spreads lock-in effects to an even larger pool of individuals than simply those who are trained.

In Latvia, additional lock-in effects from having a voucher system may potentially be sizable, but are difficult to pin down. This makes it tricky to assess whether the clear gains in employment chances and earnings that accrue to training recipients outweigh any lock-in effects that arise: the effects estimated in the main analysis use the start of training as the reference point from which future outcomes at 6, 12, 18, 24, 30 and 36 months are considered. The time for which vouchers are technically valid – as per the administrative data – is in fact remarkably short. Since, for both formal and non-formal training vouchers, there are just 14 days between the issue date and the expiry date on average, redemption and the start of training itself normally takes place within two weeks of issue and must take place within one month.33 However, voucher recipients are assigned to the training voucher programme – at which point the broad field of study on which they will focus is decided – in advance of actually receiving the voucher. During this time they are expected to search for suitable training providers and training programmes, potentially at the expense of searching for a job. However, whether they fully believe the voucher will arrive after this queuing period, such that they should reduce job search effort, cannot be easily observed. It is also unclear whether voucher recipients know how long they will be queuing for a voucher in advance of the waiting period. For some types of vouchers, the queuing time between assignment to the training voucher programme and actual receipt of the voucher can be lengthy. For non-formal training vouchers (focussing on those for foreign languages and ICTs), the average time between assignment to the training and voucher receipt is 96 days on average. For formal training vouchers, the average time between assignment to the training and voucher receipt is 46 days on average. This system potentially creates an unusual set of lock-in effects in Latvia, which arise not only while the individual holds a voucher, but also in the period before receiving the training voucher, when their status is slightly unclear.

Three strategies are already underway or under consideration in Latvia, which may help to limit the extent of any lock-in effects that arise from training for the unemployed (notwithstanding the current set up of Latvia’s voucher system). First, the consolidation of training providers discussed above may help to solve the issue of long waiting times. With fewer training providers, filling – and hence starting – classes should take less time. Second, previous reforms have allowed training participants to begin employment before the end of their training (if the employer-to-be agrees). This increases training participants’ incentives to continue job search, even when training is still ongoing. In principle, this may also mean that voucher recipients could continue their job search before training even begins (for example, when looking for a suitable training provider), although it may be harder to convince potential employers that new skills will be acquired before the training has even started. Third, the possibility of varying the specific times of day at which training occurs is currently being considered in order to make job search easier. If training is concentrated in the early morning or in the evening, then meeting employers and attending interviews on the same day may be plausible. However, this requires not only that training participants receive a manageable level of homework, but also that they are sufficiently motivated to juggle concurrent training and job search.

Conclusion

This chapter has evaluated how effective providing training to unemployed people has been in helping them find good jobs. The chapter shows that both formal and non-formal training have had positive effects on unemployed individuals’ labour market outcomes in Latvia. These effects differ according to individuals’ characteristics and depending on how trainings are combined with other ALMP measures, although virtually all sub-groups experience at least some boost to their employment chances from training. The chapter has also discussed the potential benefits of providing training to unemployed individuals through vouchers, but also some of the risks. Latvia faces three particular challenges in providing training through vouchers: (1) certain groups redeem vouchers less than others, (2) training providers are distributed unevenly across Latvia’s municipalities underlining the need to support regional mobility, and (3) the current voucher system may prolong the time for which individuals are locked-in to unemployment. These three challenges present possible areas for future policy work.

References

[51] Ashenfelter, O. and D. Card (eds.) (1999), The economics and econometrics of active labour market programs, Elsevier.

[62] Babcock, L. et al. (2012), “Notes on behavioral economics and labor market policy”, IZA Journal of Labor Policy, Vol. 1/2, https://doi.org/10.1186/2193-9004-1-2.

[55] Barnow, B. (2009), “Vouchers in U.S. vocational training programs: an overview of what we have learned”, Journal for Labour Market Research, Vol. 42/1, pp. 71-84, https://doi.org/10.1007/s12651-009-0007-9.

[61] Barnow, B. and C. King (2005), The Workforce Investment Act in Eight States.

[57] Becker, G. (1975), Human Capital, Columbia University Press.

[45] Biewen, M. et al. (2014), “The Effectiveness of Public Sponsored Training Revisited: The Importance of Data and Methodological Choices”, Journal of Labor Economics, Vol. 32/4, pp. 837-897, https://doi.org/10.1086/677233.

[50] Blattman, C. and L. Raltson (2015), Generating employment in poor and fragile states: Evidence from labor market and entrepreneurship programs, https://www.povertyactionlab.org/sites/default/files/publications/Blattman_Employment%20Lit%20Review.pdf.

[36] Bratti, M. et al. (2018), Vocational training and labour market outcomes: Evidence from Youth Guarantee in Latvia.

[65] Bruttel, O. (2005), “Delivering active labour market policy through vouchers: experiences with training vouchers in Germany”, International Review of Administrative Sciences, Vol. 71/3, pp. 391-404, https://doi.org/10.1177/0020852305056809.

[24] Cabinet of Ministers (2010), Guidelines for the Optimisation of the Network of Vocational Education Institutions 2010-2015 (Informative Part), http://www.vvc.gov.lv/export/sites/default/docs/LRTA/Citi/Cab._Order_No._5_-_Guidelines_-_Optimisation_of_the_Network_of_Vocational_Education_Institutions.doc.

[22] Cabinet of Ministers (2009), Concept “Promotion of Interest in Vocational Education and Participation of Social Partners in Assuring the Quality of Vocational Education”, http://www.vvc.gov.lv/export/sites/default/docs/LRTA/Citi/Concept_-_Promotion_of_Interest_in_Vocational_Education_and...doc.

[47] Calmfors, L. (1994), Active labour market policy and unemployment - a framework for the analysis of crucial design features.

[46] Card, D., J. Kluve and A. Weber (forthcoming), “What works? A meta analysis of recent active labor market program evaluations”, Journal of the European Economic Association.

[27] CEDEFOP (2018), Developments in vocational education and training policy in 2015-17: Latvia, http://www.cedefop.europa.eu/en/publications-and-resources/country-reports/vetpolicy-.

[72] CEDEFOP (2015), Vocational education and training in Latvia: Short description, http://www.cedefop.europa.eu/files/4134_en.pdf.

[73] CEDEFOP (2014), Apprenticeship-type schemes and structured work-based learning programmes: Latvia, https://cumulus.cedefop.europa.eu/files/vetelib/2015/ReferNet_LV_2014_WBL.pdf.

[11] Central Statistical Bureau of Latvia (2015), IZG29. Graduates with degree or qualification in higher education institutions and colleges by education thematic groups, http://data.csb.gov.lv/ErrorGeneral.aspx?aspxerrorpath=/pxweb/en/Sociala/Sociala__ikgad__izgl/IZ0290.px/.

[29] CSB (2019), Education, https://www.csb.gov.lv/en/statistics/statistics-by-theme/social-conditions/education.

[17] CSB (2019), Immigration, emmigration and net migration, https://www.csb.gov.lv/en/statistics/statistics-by-theme/population/migration/key-indicator/immigration-emmigration-and-net-migration.

[31] Daija, Z., G. Kinta and B. Ramiņa (2014), Latvia: VET in Europe: Country Report 2014, http://www.refernet.lv/uploads/Country_Report_LV_2014.pdf.

[28] Daija, Z., L. Krastina and S. Rutkovska (2018), CEDEFOP opinion survey on vocational education and training in Europe: Latvia, https://cumulus.cedefop.europa.eu/files/vetelib/2018/opinion_survey_VET_Latvia_Cedefop_ReferNet.pdf.

[41] Doerr, A. et al. (2017), “Employment and Earnings Effects of Awarding Training Vouchers in Germany”, Industrial and Labor Relations Review, Vol. 70/3, pp. 767-812, https://doi.org/10.1177/0019793916660091.

[37] EACEA (2018), EACEA National Policies Platform: 3.3 Skills Forecasting - Latvia, https://eacea.ec.europa.eu/national-policies/en/content/youthwiki/33-skills-forecasting-latvia.

[32] ENQA (2018), ENQA Agency Review: Academic Information Centre, https://enqa.eu/wp-content/uploads/2018/07/External-Review-Report-AIC-FINAL.pdf.

[1] EURES (2018), Short overview of the labour market: Latvia, https://ec.europa.eu/eures/printLMIText.jsp?lmiLang=en&regionId=LV0&catId=2776.

[34] European Commission (2016), Towards Managing Human Capital in Latvia: A Quick Look Underneath the Surface.

[30] European Commission (2015), Country report Latvia 2015, http://ec.europa.eu/.

[5] European Commission (2011), Special Eurobarometer 369: ’Attitudes Towards Vocational Education and Training’.

[15] Eurostat (2018), Tertiary educational attainment, age group 30-34.

[21] Eurostat (2017), Adult participation in learning by sex.

[20] Eurostat (2016), Adult Education Survey.

[18] Eurostat (2015), Continuing vocational training in enterprises.

[3] Eurostat (2015), Early leavers from education and training by sex, http://ec.europa.eu/eurostat/tgm/table.do?tab=table&plugin=1&language=en&pcode=t2020_40.

[40] Fredriksson, P. and P. Johansson (2008), “Dynamic Treatment Assignment”, Journal of Business and Economic Studies, Vol. 26/4, pp. 435-445.

[14] Hazans, M. (2015), Emigration from Latvia: Return intention of post-2000 emigrants from Latvia, OECD.

[13] Hazans, M. (2013), “Emigration from Latvia: Recent trends and economic impact”, in Coping with Emigration in Baltic and East European Countries, OECD Publishing, Paris, https://dx.doi.org/10.1787/9789264204928-7-en.

[38] Hazans, M. and J. Dmitrijeva (2013), An Evaluation of Active Labor Market Programs (ALMPs) and Related Social Benfit Programs.

[56] Hidalgo, D., H. Oosterbeek and D. Webbink (2014), “The impact of training vouchers on low-skilled workers”, Labour Economics, Vol. 31, pp. 117-128, https://doi.org/10.1016/j.labeco.2014.09.002.

[66] Huber, M., M. Lechner and A. Strittmatter (2015), Direct and Indirect Effects of Training Vouchers for the Unemployed.

[2] ILO (2019), National Labour Law Profile: Latvia, https://www.ilo.org/ifpdial/information-resources/national-labour-law-profiles/WCMS_158912/lang--en/index.htm.

[59] Kaplan, D. et al. (2015), Training Vouchers and Labor Market Outcomes in Chile.

[53] Kluve, J. (2010), “The effectiveness of European active labor market programs”, Labour Economics, Vol. 17/6, pp. 904-918, https://doi.org/10.1016/j.labeco.2010.02.004.

[49] Kluve, J. et al. (2016), Do Youth Employment Programs Improve Labor Market Outcomes? A Sytematic Review.

[63] Kruppe, T. (2009), “Bildungsgutcheine in der aktiven Arbeitsmarktpolitik”, Sozialer Fortschritt, Vol. 58, pp. 9-19.

[43] Lancaster, A. (2000), “The incidental parameter problem since 1948”, Journal of Econometrics, Vol. 95/1, pp. 391-413.

[71] Lechner, M. (2009), “Sequential Causal Models for the Evaluation of Labor Market Programs”, Journal of Business & Economic Statistics, Vol. 27, pp. 71-83.

[67] McConnell, S. et al. (2006), Managing Customers’ Training Choices: Findings from the Individual Training Account Experiment.

[52] McKenzie, D. (2017), How Effective Are Active Labor Market Policies in Developing Countries? A Critical Review of Recent Evidence.

[25] MoES (2019), Number of VET schools: up-to-date information.

[23] MoES (2015), Country background report Latvia.

[4] MoES (2014), Education Development Guidelines 2014-202.

[9] MoES (2014), Pārskats par Latvijas Augstāko Izglītību 2013.gadā: Galvenie Statistikas Dati [Overview of Latvian Higher Education 2013: Key Statistics], http://www.izm.gov.lv/images/statistika/augst_izgl/01.pdf.

[42] Neyman, J. and E. Scott (1948), “Consistent estimates based on partially consistent observations”, Econometrica, Vol. 16/1, pp. 1-32, https://doi.org/10.2307/1914288.

[7] OECD (2018), Education at a Glance 2018: OECD Indicators, OECD Publishing, Paris, https://dx.doi.org/10.1787/eag-2018-en.

[8] OECD (2016), OECD Reviews of Labour Market and Social Policies: Latvia 2016, OECD Publishing, https://doi.org/10.1787/9789264250505-en.

[12] OECD (2016), Reviews of National Policies for Education: Education in Latvia, OECD Publishing, https://doi.org/10.1787/9789264250628-en.

[58] OECD (2014), Connecting People with Jobs: Activation Policies in the United Kingdom, Connecting People with Jobs, OECD Publishing, Paris, https://dx.doi.org/10.1787/9789264217188-en.

[26] Pilz, M. (2012), “Modularisation of vocational training in Germany, Austria and Switzerland: Parallels and disparities in the modernisation process”, Journal of Vocational Education and Training, Vol. 64/2, pp. 169-183.

[68] Poeschel, F. (2014), Assignment vs. choice: lessons from training vouchers.

[69] Rinne, U., A. Uhlendorff and Z. Zhao (2013), “Vouchers and caseworkers in training programs”, Empirical Economics, Vol. 45, pp. 1089-1127, https://doi.org/10.1007/s00181-012-0662-5.

[39] Sianesi, B. (2004), “An Evaluation of the Swedish System of Active Labor Market Programs in the 1990s”, Review of Economics and Statistics, Vol. 86/1, pp. 133-155.

[44] Söderbom, M. (2009), Estimation of Nonlinear Models with Panel Data, http://www.soderbom.net/lecture15final.pdf.

[64] SRI International (ed.) (1983), Impacts of Counseling and Education Subsidy Programs, SRI International.

[70] Stevens, M. (1994), “A Theoretical Model of On-the-Job Training with Imperfect Competition”, Oxford Economic Papers, Vol. 46/4, pp. 537-562.

[60] Strittmatter, A. (2016), “What effect do vocational training vouchers have on the unemployed?”, IZA World of Labor, Vol. 316, https://doi.org/10.15185/izawol.316.

[54] Tomini, F., W. Groot and H. Maassen van den Brink (2016), The effectiveness of the voucher training programs: A systematic review of evidence from evaluations.

[6] UN DESA (2017), World Population Prospects: The 2017 Revision, https://population.un.org/wpp/Download/Standard/Population/.

[48] Vooren, M. et al. (2018), “The Effectiveness of Active Labor Market Policies: A Meta-Analysis”, Journal of Economic Surveys, Vol. 00/0, pp. 1-25, https://doi.org/10.1111/joes.12269.

[16] World Bank (2019), World Development Indicators: school enrollment, tertiary (% gross), Latvia, https://data.worldbank.org/indicator/SE.TER.ENRR?locations=LV.

[33] World Bank (2017), Internal Funding and Governance in Latvian Higher Education Institutions: Status Quo Report, http://www.izm.gov.lv/images/izglitiba_augst/Pasaules_Banka/LV_2nd_HEd_RAS_Ph1_Status_Quo_Report_EXT_FINAL_13Feb17.pdf.

[19] World Bank (2015), The Active Aging Challenge: For Longer Working Lives in Latvia.

[10] World Bank (2014), Assessment of Current Funding Model’s “Strategic Fit” with Higher Education Policy Objectives, http://viaa.gov.lv/files/news/24067/lv_r2_strategic_fit_18april_vfinal.pdf.

[35] Zvīdriņa, I. (2015), ICT-related active labour market policies in Latvia, http://eskillsforjobs.lv/wp-content/uploads/2015/03/ilze-zvidrina-eskillsforjobs-2015.pdf.

Database references

World Indicators of Skills for Employment (WISE) Database, http://stats.oecd.org/Index.aspx?DataSetCode=WSDB.

European Labour Force Survey (Eurostat), http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=lfso_16workexp&lang=en.

Adult Education Survey (Eurostat), http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=trng_aes_103&lang=en.

Annex 3.A. Additional figures

Annex Figure 3.A.1. Breakdown of average lengths for second, third, fourth and fifth unemployment spells

Annex Figure 3.A.2. Hazard rates of trained and untrained women and men

Annex Figure 3.A.3. Estimated effects of training on employment and earnings by gender

Annex Figure 3.A.4. Estimated effects of training on employment and earnings by age

Annex Figure 3.A.5. Estimated effects of training on employment and earnings by urbanicity

Annex Figure 3.A.6. Estimated effects of training on employment and earnings by skill level

Annex Figure 3.A.7. Estimated effects of training on employment and earnings by social assistance receipts

Annex Figure 3.A.8. Estimated effects of training on employment and earnings depending on participation in other substantive ALMPs

Annex Figure 3.A.9. Estimated effects of training on employment and earnings depending on participation in mobility support

Annex Figure 3.A.10. Estimated effects of training on employment and earnings depending on participation in Measures to Improve Competitiveness

Annex Figure 3.A.11. Number of accredited formal training sites (SEA partner sites) by municipality in Latvia

Annex Figure 3.A.12. Number of accredited non-formal training sites (SEA partner sites) by municipality in Latvia

Annex 3.B. Additional tables

Annex Table 3.B.1. Sample characteristics of treated and untreated individuals
	Formal training received first: One training only	Non-formal training received first: One training only	Formal training received first: Multiple ALMPs	Non-formal training received first: Multiple ALMPs	Other ALMPs received before training	No training received
Female	65.43	56.39	69.48	54.71	64.88	51.75
Age
15-24	13.99	7.10	12.19	8.22	39.84	24.02
25-34	31.54	24.95	25.20	22.07	14.04	24.99
35-44	22.60	24.76	22.87	25.56	13.39	18.65
45-54	21.80	26.79	27.70	29.01	20.28	18.71
55+	10.07	16.40	12.04	15.13	12.44	13.63
Education
Not known	0.16	0.09	0.23	0.10	0.30	10.99
Basic	15.26	12.28	12.56	12.51	21.42	17.54
Secondary	28.60	23.11	27.57	21.99	31.09	22.37
Vocational	5.18	5.67	4.80	6.13	6.20	5.50
Professional secondary	29.58	32.56	30.89	34.28	27.21	25.90
Professional higher	9.89	13.62	11.36	13.03	7.08	7.78
Higher	11.33	12.67	12.58	11.97	6.70	9.92
Ethnicity
Latvian	68.59	46.46	68.04	50.68	58.11	61.53
Slavic	27.12	46.86	28.54	43.81	37.33	32.97
Other	4.29	6.68	3.42	5.51	4.57	5.50
Non-Latvian citizenship	7.02	22.28	7.05	18.01	11.87	13.87
Language
Not known	0.11	0.04	0.03	0.08	0.00	0.97
None	1.92	7.67	1.88	4.44	5.63	7.18
Basic	5.74	15.47	6.08	12.82	11.38	8.34
Intermediate	13.69	21.27	13.58	21.76	16.55	13.44
Higher	6.02	8.13	8.04	9.25	5.14	4.62
Educated in Latvian	72.52	47.41	70.39	51.64	61.30	65.45
Married	39.56	46.91	41.25	46.93	28.50	34.75
Has children (aged less than 18 years)	45.28	40.04	40.05	39.08	29.83	34.93
Month of registration
January	7.94	8.20	8.30	8.03	8.14	8.37
February	9.21	9.30	9.37	9.43	9.67	10.17
March	7.50	8.60	8.02	7.57	7.95	8.28
April	7.69	7.83	7.55	7.78	6.70	9.97
May	7.38	8.12	7.99	7.27	7.46	10.34
June	7.82	7.96	8.33	7.95	8.83	8.22
July	7.89	7.65	8.75	8.16	9.40	8.11
August	8.28	7.93	7.81	8.27	10.35	7.56
September	9.28	9.00	8.83	8.92	9.74	8.02
October	9.26	8.88	8.64	8.08	7.80	7.48
November	8.79	7.96	8.07	8.30	7.34	6.87
December	8.96	8.56	8.36	10.24	6.62	6.60
Urban	43.03	59.96	34.13	44.47	43.92	46.79
Regions
Riga City	19.32	34.18	12.49	17.11	16.61	27.38
Pieriga	13.32	12.69	14.63	13.85	8.08	17.28
Vidzeme	12.05	7.84	11.68	9.89	11.89	10.94
Zemgale	14.94	9.91	13.04	11.55	12.27	13.55
Kurzeme	16.90	13.89	15.44	17.00	13.90	15.04
Latgale	23.47	21.48	32.71	30.59	37.26	15.81
Time since previous employment
3 months or less	72.79	80.26	71.83	73.53	48.67	66.41
4-12 months	4.38	3.46	4.23	4.05	6.58	5.69
13-24 months	1.59	1.26	1.67	1.96	3.01	2.06
More than 24 months	2.20	2.08	3.26	3.33	2.05	2.78
Never worked/Unknown	19.03	12.94	19.01	17.14	39.69	23.06
Receiving social assistance at unemployment spell start	3.48	2.95	3.52	3.37	6.62	3.65
Disabled at unemployment spell start	7.42	8.40	12.04	12.66	16.10	6.28
Source: Latvian State Employment Agency, Latvian Social Insurance Agency data and OECD estimates.
StatLink https://doi.org/10.1787/888933961581

Annex Table 3.B.2. Regressions of employment status on training participation
	Formal						Non-formal
	6 months	12 months	18 months	24 months	30 months	36 months	6 months	12 months	18 months	24 months	30 months	36 months
Formal training	-0.2160***	-0.1306***	-0.0593***	-0.0187**	-0.0121**	0.0069
	(0.0064)	(0.0057)	(0.0072)	(0.0070)	(0.0060)	(0.0065)
Non-formal training							-0.2449***	-0.1692***	-0.1031***	-0.0540***	-0.0476***	-0.0283***
							(0.0116)	(0.0105)	(0.0062)	(0.0072)	(0.0042)	(0.0051)
Individual controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	383 299	357 500	326 883	293 770	263 275	227 696	383 299	357 500	326 883	293 770	263 275	227 696
R-squared	0.1038	0.1230	0.1415	0.1287	0.1333	0.1214	0.1161	0.1289	0.1441	0.1295	0.1340	0.1216
Note: Standard errors in parentheses.
Dependent variable: employment at set number of months after registration.
Standard errors clustered at the level of the SEA branch office.
Individual controls include age, age squared, education level, marital status, number of children under 6 and 18 years, disability status, ethnicity, citizenship, time since last employment, level of Latvian language skill and receipt of social assistance.
Fixed effects for month of unemployment registration, region and SEA branch office are included.
* p<0.10; p<0.05; * p<0.01.
Source: Latvian State Employment Agency (SEA), Latvian Social Insurance Agency and OECD estimates.
StatLink https://doi.org/10.1787/888933961600

Annex Table 3.B.3. Sample characteristics of voucher recipients by redemption status
	Voucher not used	Voucher used
All	22.68	77.32
Gender
Women	23.10	76.90
Men	22.05	77.95
Age
15-24	26.35	73.65
25-34	24.06	75.94
35-44	21.61	78.39
45-54	19.84	80.16
55+	19.79	80.21
Education
Not known	26.90	73.10
Basic	22.06	77.94
Secondary	22.92	77.08
Vocational	19.21	80.79
Professional secondary	20.39	79.61
Professional higher	23.86	76.14
Higher	27.12	72.88
Ethnicity
Latvian	21.92	78.08
Slavic	23.52	76.48
Other	24.18	75.82
Citizenship
Latvian	22.03	77.97
Non-Latvian	26.42	73.58
Language
Not known	14.89	85.11
None	29.30	70.70
Basic	21.26	78.74
Intermediate	22.89	77.11
Higher	22.03	77.97
Educated in Latvian	22.36	77.64
Marital status
Unmarried	23.61	76.39
Married	21.21	78.79
Children
No children	23.06	76.94
Has children	22.08	77.92
Urbanicity
Urban	26.58	73.42
Rural	18.86	81.14
Regions
Riga City	34.19	65.81
Pieriga	28.16	71.84
Vidzeme	20.34	79.66
Zemgale	22.03	77.97
Kurzeme	18.10	81.90
Latgale	13.89	86.11
Social assistance (January 2012)
Non-recipient	22.72	77.28
Recipient	22.11	77.89
Disability status (January 2012)
Not disabled	22.85	77.15
Disabled	19.80	80.20
Training voucher type
Formal	23.93	76.07
Non-formal	22.13	77.87
Youth Guarantee
Not Youth Guarantee	21.72	78.28
Youth Guarantee	26.37	73.63
Source: Latvian State Employment Agency, Latvian Social Insurance Agency data and OECD estimates.
StatLink https://doi.org/10.1787/888933961619

Notes

← 1. The distinction between formal and non-formal training is discussed in detail at the start of the main analysis section. In short, formal trainings build a specific new skill such as social care, project management, or welding, among participants leading to a professional qualification, whereas non-formal trainings do not necessarily result in a professional qualification, and tend to build cross-cutting skills such as languages and ICTs. Formal trainings typically last longer and require more hours of contact time than non-formal training.

← 2. The nature of “Measures to Improve Competitiveness” is discussed in detail in at the start of the main analysis section. These measures typically comprise very short courses and workshops, that help participants engage with the labour market, including support for writing CVs, succeeding at interviews, and networking effectively.

← 3. By some metrics, job quality is slightly lower in Latvia than in the OECD at large: rates of “labour market insecurity” and “job strain” were at 30.3% and 9.6% in Latvia respectively in 2015, compared to OECD averages of 27.6% and 5.7%. “Labour market insecurity” is defined in terms of the expected earnings loss associated with unemployment. This is calculated based on the OECD Unemployment Duration database, the OECD Benefit Recipients database, the OECD Labour Market Programmes database, and the OECD Taxes and Benefits database. “Job strain” is defined in terms of jobs where workers face more job demands than the number of resources they have at their disposal. This is calculated based on the European Working Conditions Survey and the International Social Survey Programme, and includes factors including long working hours, physical health risk factors, work autonomy and learning opportunities, and social support at work.

← 4. The fact that the proportion of the population who have attained tertiary education is now relatively high in Latvia should be borne in mind when interpreting these earnings premia. Earnings premia for tertiary education in Latvia may be lower than the OECD average because – while there are certain fields of study that are in shortage – overall, tertiary educated individuals are in high supply.

← 5. More than 30 percent of individuals aged 20-24 in Denmark, Luxembourg, and Spain have less than upper secondary education, according to the 2017 European Labour Force Survey (Eurostat).

← 6. The overall statistics mask significant gender differences in educational attainment in Latvia. Amongst 20-24 year-olds, Latvian men are more than twice as likely as Latvian women to have left school without upper secondary education. For the same age group across the EU, men are just 1.4 times more likely than women to lack upper secondary education.

← 7. The main field of study at VET secondary schools in Latvia appears to be engineering, manufacturing, and construction. In 2013, 39.2% of students focussed on engineering, manufacturing, and construction (CEDEFOP, 2015[72]). The next most common fields of study were services (25.0%), social sciences, business, and law (13.7%), and humanities and arts (including design programmes, 10.2%).

← 8. These estimates exclude Erasmus exchange students from Latvia, of whom there were 2 100 in the 2011/2012 academic year.

← 9. Government-subsidised student loans are available for all Latvian residents pursuing tertiary education, assuming they are able to meet co-signatory loan requirements. Other loans are also available from Latvia’s commercial banks (OECD, 2016[12]).

← 10. The gross enrolment rate is the ratio of total enrolment (regardless of age) to the population of the age group that officially corresponds to the given level of education. For tertiary education, this corresponds to the 5-year age group starting from the official secondary school graduation age.

← 11. These results exclude Ireland.

← 12. There is some discussion in the media suggesting that employers’ perceptions of individuals attaining VET may be improving (see, for example, https://nra.lv/latvija/izglitiba-karjera/243139-profesionala-izglitiba-darba-tirgu-kotejas-augstu.htm), but nationally-representative data proving this phenomenon do not yet exist.

← 13. In 2012, just 72 apprentices received diplomas through the Chamber of Crafts (CEDEFOP, 2014[73]).

← 14. A separate commission is responsible for determining the fields of study for training for the employed.

← 15. The term “participations” is used instead of “participants” because, individuals were able to participate in more than one type of ALMP measure each year.

← 16. MICs thus fall under the “Labour market services” category presented in Chapter 2.

← 17. In particular, employment histories are missing for those individuals that had an employment spell after their first recorded unemployment spell, but who were not employed (perhaps due to being out of the labour force entirely) before their first recorded unemployment spell.

← 18. In the Latvia Social Insurance Agency data, those individuals who are classified as registered unemployed can technically also be employed. This may arise, for example, when registered unemployed individuals participate in certain ALMP measures, such as public works schemes and employment subsidies. However, the definition of “unemployed” used in this analysis only includes those individuals that are registered unemployed and are not working.

← 19. It was not possible to separate out formal and non-formal trainings provided as part of the Youth Guarantee easily in the administrative data from the SEA.

← 20. The analogous waiting times for other types of non-formal training are shorter. For example, for car driver training, the waiting time is 91 days on average, while for regular Latvian language training, the waiting time is 83 days on average.

← 21. Not all training participants receive MICs or career consultations. Around 20% of formal training participants did not receive a MIC while 31% of formal training participants did not receive a career consultation.

← 22. The approach taken by this chapter therefore differs from Hazans and Dmitrijeva (2013[38]). In their study, individuals receiving more than one formal or non-formal training, or receiving any other substantive ALMP measures (such as employment measures) are excluded from the sample for the main estimations.

← 23. For a more formal discussion of ways to evaluate sequences of ALMP measures, see Lechner (2009[71]) and Huber, Lechner, and Strittmater (2015[66]).

← 24. Employment could, in principle, be in a subsidised job, although the proportion of such jobs in the overall sample is small.

← 25. Annex Figure 3.A.2 aims to estimate the hazard rates in the same way as in Hazans and Dmitrijeva (2013[38]) on the updated State Employment Agency and Social Insurance Agency data. To replicate the sample used by Hazans and Dmitrijeva, it is necessary to remove all individuals that took part in more than one formal or non-formal training as well as dropping any individuals that participated in other substantive ALMP measures, including employment measures and mobility support. In contrast to Hazans and Dmitrijeva, the hazard rate remains higher for those individuals that did not participate in training for at least two years after registration.

← 26. These results are robust to reapplying the sample restrictions suggested by Hazans and Dmitrijeva (2013[38]) and dropping those individuals that received more than one formal or non-formal training as well as those that received other substantive ALMP measures (including employment measures and mobility support).

← 27. The formal tests were carried out by taking the difference between the point estimates for formal training and for non-formal training, and then bootstrapping this statistic with 250 repetitions (clustering at the SEA branch level) to calculate the standard error.

← 28. This approach follows Biewen et al. (2014[45]), adapting the methodology presented in Box 3.2. Each regression from each month endured in unemployment ( $m$ ) is augmented with a full set of interaction terms as well as a dummy variable for the relevant sub-group. The coefficient on the interaction between the sub-group dummy and the treatment dummy is taken from each regression (from each $m$ ) and then a weighted average over all $m$ is calculated. The weights – which capture the proportion of all treated individuals entering the treatment at $m$ – are taken from the full sample, where both sub-groups are combined. For example, rather than having separate sets of weights for women and men, the weights are taken from a sample that contains both women and men. This means that the difference between the treatment effects for the two sub-groups estimated when splitting the sample is not the same as the difference estimated using the fully interacted model.

← 29. Those individuals that began participating in training and receiving mobility support simultaneously were included as training participants in the main analysis.

← 30. Stevens (1994[70]) suggests a more nuanced view, where there may be “transferable” skills, rather than the simple dichotomy between “general” and “specific” skills. Transferable skills are valuable to more than one firm, but they are traded in a labour market that is not perfectively competitive, so that workers with transferable skills are paid less than their marginal product. Since training in transferable skills therefore raises productivity more than it raises wages, both worker and firm may benefit. However, training in transferable skills still results in a positive externality, given that other firms may benefit. This leads to underinvestment in transferable skills, in the absence of government intervention.

← 31. For training provided to employed individuals, there is potentially an additional misalignment between the motivations of worker and firm. For example, in the Netherlands, Hidalgo, Oosterbeek, and Webbink (2014[56]) show employers react less positively to courses provided through vouchers (rather than by the firm itself) while the families and partners of voucher recipients react more positively.

← 32. The null hypothesis that the coefficients on the Kurzeme dummy variable and the Latgale dummy variable are the same can be rejected at the 10% level in Columns (2) and (3).

← 33. Vouchers are typically valid for much longer in other countries. In Germany, similar training vouchers are valid for up to three months.

End of the section – Back to iLibrary publication page