This chapter examines the effects of Lithuania’s employment subsidy programmes on a number of labour market outcomes. In addition to outcomes typically examined in impact evaluations, such as employment probability and duration, the analysis examines the effects on wages, occupational mobility and earnings, including earnings net of the direct subsidy costs. It also compares the results obtained by the counterfactual impact evaluation with those of similar studies, both for Lithuania and for other countries. The estimated effects are examined across sub-groups of workers based on their age, gender, skill level and urban or rural location. The chapter concludes by analysing whether employment subsidies lead to subsidised workers replacing unsubsidised ones.
Impact Evaluation of Vocational Training and Employment Subsidies for the Unemployed in Lithuania
5. Evaluation of employment subsidy programmes administered by the Lithuanian Public Employment Service
Abstract
5.1. Introduction
In addition to vocational training, the employment subsidy programme is one of the key active labour market programmes (ALMPs) used to help connect unemployed people with jobs in Lithuania. By providing employers with a financial incentive to hire certain categories of jobseekers, employment subsidies can facilitate the integration of such individuals into the labour market. This chapter examines how effective Lithuania’s subsidised employment programme has been in placing individuals into sustained employment, how it has affected their career prospects, how the effects vary across individuals, and whether the subsidies lead to subsidised workers displacing unsubsidised ones.
The impact evaluation results indicate that employment subsidies generate large and statistically significant effects on individuals’ probability of being in employment. Compared with the results of other studies of similar programmes in other countries, the estimated effects for Lithuania are generally much larger over the first 12 months and in the lower range of estimates over longer time horizons. Furthermore, employment subsidies have a positive effect on occupational mobility for men, but not for women.
The organisation of the chapter is as follows. The next section presents the overall results of subsidised employment on the key outcomes examined: employment probability and duration, wages, occupational mobility and earnings, including earnings net of the direct subsidy costs. It also compares the results obtained by the counterfactual impact evaluation with those of similar studies, both for Lithuania and for other countries. The subsequent section compares the outcomes observed for subsidised employment across sub-groups of workers based on their age, gender, skill level and urban or rural location. This is followed by an examination of the extent to which effects vary across different attributes of the subsidised employment programmes, including in comparison with other studies. The chapter concludes by examining whether the employment subsidy programmes lead to displacement effects, analysing whether the subsidies lead to subsidised workers replacing unsubsidised ones. This section discusses only direct displacement effects occurring within individual firms, with questions relating to deadweight effects – hiring that would have occurred even in the absence of the subsidies – being outside the scope of questions that can be answered with the available data.
5.2. Employment subsidies have positive effects on most outcomes examined
The next sections describe the aggregate results for employment subsidies on selected labour market outcomes and compare them to the results from other studies. The effects of employment subsidies on labour market outcomes are estimated using the dynamic selection-on-observables approach described in Chapter 3 of this report.
5.2.1. Positive effects of employment subsidies decline over time but remain present even after three years
The results show that employment subsidies generate large and statistically significant effects on individuals’ probability of being in employment. As Figure 5.1, Panel A shows, after 12 months, individuals who entered subsidised employment (the intervention group) were almost 26.7 percentage points more likely to be in employment than those who were not participating in a substantive ALMP measure or another way out of unemployment (the comparison group, see Section 3.6 of Chapter 3 for more details of the econometric approach). Given that the subsidy period lasts for a maximum of six months – and that for over a quarter of participants, the actual duration is less than this long – these effects capture unsubsidised employment (although firms still have an incentive to retain workers for further six months, tied to their continued eligibility to use the employment subsidy scheme for new hires). The effects remained positive for several months over the entire observation period: 36 months after the start of subsidised employment, individuals on employment subsidies were still 10.9 percentage points more likely to be employed than individuals who did not enter an ALMP at the beginning of the observation period.
Paralleling the positive effects on employment probability, jobseekers entering subsidised employment were employed for a considerably longer period than jobseekers who did not enter subsidised employment (Figure 5.1, Panel B). Over the three‑year time horizon studied, they were employed for 269 days more than individuals who were not employed via employment subsidies. Note that this period includes days worked which were directly subsidised (during the first six months), as well as employment during the subsequent period for which employment subsidies were not paid. A majority of this effect – roughly 60% – is attributable to additional days worked during the period after the initial six months. During the first six months – a period during which employers were paid employment subsidies for individuals on subsidised employment – they were employed for 114 days more than individuals in the control group.
In addition to providing a boost to their employment probability and duration, jobseekers entering subsidised employment experienced a short-term boost in their occupational mobility but received similar wages to those who did not enter subsidised employment. Jobseekers entering subsidised employment experience a boost to their occupational index, with positive and statistically significant effects observed in months 9 to 18 after starting the employment subsidies (Figure 5.1, Panel C) While the average effect over this period is positive, it is rather small, amounting to an increase in EUR 10.8 in the occupational index – an increase that corresponds to 1.0% of real average wages observed during this period. At the same time, in terms of wages, the point estimates are statistically insignificant over the entire time horizon studied, indicating that the very positive employment effects are not counteracted by a negative effect in terms of match quality as this is reflected in wages (Figure 5.2, Panel A).
The combined effect of the factors described above – positive effects of employment subsidies on employment and occupational mobility, together with inconclusive effects on wages – is a positive effect on cumulative earnings (Figure 5.2, Panel B). Twelve months after entering subsidised employment, individuals who had become employed through employment subsidies earned EUR 2 801 more than those who had not, with the difference increasing to EUR 4 653 after 36 months. The trajectory of the increase over time, with subsequent increases remaining positive but diminishing in magnitude, parallel the trajectory of the employment effects, which also become progressively smaller in magnitude. These effects are quite sizable also when taken in the context of counterfactual earnings: cumulatively, at 36 months, individuals in the control group earned an average of EUR 6 909. Individuals in the treatment group thus experienced a 64.8% increase in earnings over this period.
The effects of the employment subsidies on cumulative earnings are positive also after subtracting the direct costs of the employment subsidies (Figure 5.2, Panel C). The effects are positive already at three months – at which point the estimated effect on cumulative earnings net of subsidies amounts to EUR 456. The net effects increase until 33 months after becoming employed via an employment subsidy, at which point they amount to EUR 3 675.
Figure 5.1. Employment subsidies have positive effects on employment and, to a lesser extent, occupational mobility
Note: The analysis presents nearest-neighbour propensity score matching results which matches individuals based on a number characteristics: duration of unemployment, age, gender, education, unemployment benefit receipt and level, language and other skills, municipality, employability and barriers to employment, as well as prior employment history and earnings. For every individual in the treatment group, the matching is conducted based on the values of these characteristics in the calendar month when the individual enters the programme. The control group is comprised of individuals with similar characteristics not entering ALMPs in that same calendar month. For paired individuals in the treatment and control groups, this calendar month is then the reference point after which outcomes are measured. The analysis is restricted to the region of common support. The standard errors are calculated based on the adjustment proposed by Abadie and Imbens (2016[1]). The confidence intervals are shown at the 5% level of significance and represented by the whiskers delimiting the dotted lines on the charts.
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
Figure 5.2. Employment subsidies have positive effects on cumulative earnings, but insignificant effects on daily wages
Note: The analysis presents nearest-neighbour propensity score matching results which matches individuals based on a number characteristics: duration of unemployment, age, gender, education, unemployment benefit receipt and level, language and other skills, municipality, employability and barriers to employment, as well as prior employment history and earnings. For every individual in the treatment group, the matching is conducted based on the values of these characteristics in the calendar month when the individual enters the programme. The control group is comprised of individuals with similar characteristics not entering ALMPs in that same calendar month. For paired individuals in the treatment and control groups, this calendar month is then the reference point after which outcomes are measured. The analysis is restricted to the region of common support. The standard errors are calculated based on the adjustment proposed by Abadie and Imbens (2016[1]). The confidence intervals are shown at the 5% level of significance and represented by the whiskers delimiting the dotted lines on the charts.
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
5.2.2. The estimated boost to employment probability by Lithuania’s employment subsidy programme compare favourably with estimates from other studies in the short term
In order to examine how the results obtained by the CIE of the Lithuanian measures compare with those of similar studies in other countries, this section places them in the context of results from a meta‑analysis conducted by Card, Kluve and Weber (2018[2]). The meta‑analysis summarises estimates from over 200 recent impact evaluations of ALMPs. Of these, 15 impact evaluations contain point estimates for the employment effects of private employment subsidy programmes comparable to the one in Lithuania. As noted in the discussion of the results of training in Chapter 4, the meta‑analysis does not provide estimates of the effects of other outcomes analysed for Lithuania, such as earnings or days worked.
Compared with the results of the meta‑analysis by Card, Kluve and Weber (2018[2]), the estimated effects for Lithuania are generally much larger in the short term, and in the lower range of estimates over longer time horizons (Figure 5.3). The estimated short-term effect for Lithuania, 26.3 percentage points, is considerably higher than the median of 0.0 percentage points found in the comparison studies, while the long-term effect, of 11 percentage points, is lower than the 22.7 percentage point median of comparison studies. Nevertheless, it is worth noting that despite the relatively lower point estimate of the long-term effect, the higher coefficients in the other studies are not necessarily statistically significant. In fact, a small majority (58%) of the studies do not find positive and statistically significant results over the long term.
Figure 5.3. Compared to other studies, the estimated effects of employment subsidies on employment probability are particularly positive in the short term in Lithuania
Note: Short, medium and long-term effects respectively refer to effects up to one year, 1‑2 years, and more than two years after programme completion. For Lithuania, results refer to 12, 24 and 36 months after beginning the programme. As such, the observation periods are similar, but potentially not fully aligned. Point estimates are included in the chart even if they are statistically insignificant. The studies presented adopt various research designs and econometric techniques – the results for Lithuania use nearest-neighbour propensity score matching (for details, see Chapter 3).
Source: Card, D., J. Kluve and A. Weber (2018), “What Works? A Meta Analysis of Recent Active Labor Market Program Evaluations”, https://doi.org/10.1093/jeea/jvx028 and OECD calculations based on data from Lithuanian Employment Service.
The estimated effects also compare favourably to previous impact evaluations of employment subsidies in Lithuania. In particular, the estimated effects are more positive than the results reported by PPMI (2015[3]) who examined employment subsidies in Lithuania for an older time period (around 2010). This study found that individuals participating in employment subsidies were employed for roughly two months more during the first year after entering the programme and for an additional one month more during the second year. The effected boost to earnings amounted to EUR 500 and EUR 350 during the first and second years, respectively. Examining the effects on employment subsidy participants among participants entering the intervention in 2016, an impact evaluation by ESTEP (2019[4]) also found positive effects. In that evaluation, the effect of employment subsidies after (roughly) two years amounted to 92 additional days worked and EUR 1 710 in additional earnings. The estimates in these two studies are considerably lower than the ones in this evaluation, which examines entrants into ALMPs during the 2015‑19 periods and finds effects after two years of 224 days and EUR 2 004 on days in employment and cumulative earnings, respectively.
5.3. The effects of employment subsidies vary across sub-groups of unemployed people
This section discusses how the results of the employment subsidies vary across sub-groups of the population. It begins by discussing the detailed results for Lithuania and concludes by contrasting these results with those of other comparable studies.
5.3.1. Employment subsidies are more effective for sub-groups such as older workers
Given that the results above have documented the generally positive effects of employment subsidies in aggregate, an interesting additional set of questions concerns their effects across different characteristics of subgroups of unemployed. Paralleling the analysis of vocation training in Section 4.2 of Chapter 4, the subsequent analysis provides separate estimates for the results along several dimensions: (i) gender, (ii) age, (iii) level of education, (iv) urban vs. non-urban residence, and (v) long-term unemployment status.
Men and women tend to benefit slightly differently from employment subsidies. As discussed in greater detail below, both men and women experience a similar boost in terms of cumulative earnings, but women experience a greater increase in employment probability in the first months after entering a subsidised job. Employment subsidies have a positive effect on occupational mobility for men, but not for women. The effects on wages at given points in time are inconclusive. In terms of employment probabilities, women experience a considerable boost during the first 15 months after becoming employed with employment subsidies, with the average difference in employment probability between men and women amounting to 5.2 percentage points during that time (Annex Figure 5.A.1, Panel A). The effects thereafter point to relatively similar effects for men and women: at 36 months, for example, the point estimates of the treatment effects are 11.5 and 10 percentage points for women and men, respectively. The increased employment probability translates into a greater number of cumulative days in employment: three years after starting subsidised employment, women had been employed for 286 days more than comparable women who did not enter subsidised employment, with the comparable figure for men amounting to 249 days (Annex Figure 5.A.1, Panel B).
Employment subsidies have statistically significant positive effects on occupational mobility for men but not women (Annex Figure 5.A.1, Panel C). Men experience a statistically significant positive effect on occupational mobility throughout the months after starting working with an employment subsidy, with positive but statistically insignificant effects during one‑third of the observed period. The average of the point estimates amounts to EUR 16.8, meaning that the effect represents a boost of 1.6% compared to the average real wage during this period. Women, by contrast, do not experience a statistically significant effect on occupational mobility when compared to other (previously unemployed) women who entered unsubsidised employment at the beginning of the observation period. Underlying these results are different trends in the occupational mobility of the control group (comparable individuals who were unemployed at the beginning of the observation period but did not enter subsidised employment): men in the control group who became employed experienced a slight decrease in their occupational index, but women in the control group did not. For men, subsidised employment can thus be viewed as mitigating the negative effects of unemployment on occupational mobility which are otherwise observed among unemployed men. Women entering employment from unemployment, on the other hand, do not in generally experience these negative effects on occupational mobility, regardless of whether they enter subsidised or unsubsidised employment.
The combined result of these factors is that both women and men experience similarly positive effects on cumulative earnings, including after accounting for the direct costs of the employment subsidies (Annex Figure 5.A.2, Panels B and C). After 36 months, the estimated effects on cumulative earnings for women and men amount to EUR 4 493 and EUR 4 603, respectively. After subtracting the costs of the subsidies, the estimated amounts are very similar, amounting EUR 3 561 for women and EUR 3 608. At the same time, however, there are not any statistically significant effects of employment subsidies on the daily wages of individuals who become employed (Annex Figure 5.A.2, Panel A).
Examining the estimated effects of subsidised employment on employment probability by other jobseeker characteristics show that the positive effects are most pronounced for older women and non-urban jobseekers (Figure 5.4). The effects for women increase considerably among older cohorts: compared to an 11.9 percentage point boost in employment probability experienced by women under 30, women aged 30‑50 and over 50 experience gains of 18.2 and 21 percentage points, respectively. Non-urban jobseekers experienced a boost of 15.8 percentage points, compared to 12.2 percentage points for urban jobseekers. Employment probabilities at 24 months do not have important systematic differences across the other characteristics examined, which include jobseeker age among men, skill level, or unemployment duration.
Examining the evolution of the estimated occupational index by age and gender shows stark differences in the profiles by age groups. Figure 5.5 plots changes to the occupational index over time, taking the month when individuals enter the employment subsidies programme as the reference point. In contrast to results presented elsewhere in the chapter, the results here depict gross outcomes and not net outcomes (also known as treatment effects), which can be calculated by subtracting the values for the control group from the values for the treatment group. Several interesting findings emerge from these figures:
For individuals under 30, both men and women generally experience increases in their occupational index over time, and entering employment with employment subsidies generally does not have a statistically significant effect on this trend. This result is in stark contrast to the analogous figures for vocational training discussed in Chapter 4, which show negative effects of training for men and positive effects for women over time horizons under 24 months.
For individuals aged 30‑50, individuals who entered subsidised employment generally do not experience a change in their occupational index.
For individuals aged over 50, both men and women who become re‑employed tend to experience downward occupational mobility. For men, entering employment with employment subsidies helps mitigate this downward mobility, with some of the point estimates indicating that participation confers a statistically significant positive boost. Nevertheless, even men participating in employment subsidy programme experience a slight negative effect on their occupational mobility.
Figure 5.4. The positive employment effects of employment subsidies are particularly strong for certain sub-groups such women over 50 years of age
Note: The analysis presents nearest-neighbour propensity score matching results which matches individuals based on a number characteristics: duration of unemployment, age, gender, education, unemployment benefit receipt and level, language and other skills, municipality, employability and barriers to employment, as well as prior employment history and earnings. For every individual in the treatment group, the matching is conducted based on the values of these characteristics in the calendar month when the individual enters the programme. The control group is comprised of individuals with similar characteristics not entering ALMPs in that same calendar month. For paired individuals in the treatment and control groups, this calendar month is then the reference point after which outcomes are measured. The analysis is restricted to the region of common support. The standard errors are calculated based on the adjustment proposed by Abadie and Imbens (2016[1]).
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
Although a small number of point estimates at specific time horizons showed some statistically significant effects of employment subsidies on occupational mobility, most of the point estimates are not statistically significant. In fact, 24 months after entering subsidised employment, none of the other jobseeker characteristics have a statistically significant effect on occupational mobility (Annex Figure 5.A.3). This means that the positive employment effects experienced do not have a generally measurably negative or positive effect on occupational mobility. The estimated effect at 24 months is statistically insignificant also for men, who otherwise experienced statistically significant positive effects on occupational mobility during two out of three points for which effects were estimated during the observation period.
Figure 5.5. Employment subsidies help mitigate downward occupational mobility for men over 50
Note: The figure plots gross outcomes separately for individuals in the treatment and control groups. The analysis presents nearest-neighbour propensity score matching results which matches individuals based on a number characteristics: duration of unemployment, age, gender, education, unemployment benefit receipt and level, language and other skills, municipality, employability and barriers to employment, as well as prior employment history and earnings. For every individual in the treatment group, the matching is conducted based on the values of these characteristics in the calendar month when the individual enters the programme. The control group is comprised of individuals with similar characteristics not entering ALMPs in that same calendar month. For paired individuals in the treatment and control groups, this calendar month is then the reference point after which outcomes are measured. The analysis is restricted to the region of common support. The standard errors are calculated based on the adjustment proposed by Abadie and Imbens (2016[1]). Estimates are plotted relative to their values at month zero. Shaded circles denote point estimates for which differences between individuals in the treatment and control groups are statistically significant at the 5% level of significance.
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
The effects of employment subsidies on earnings at 24 months are heterogeneous across jobseeker characteristics. Some of the estimated effects parallel those observed for employment probability, especially those for older women. Paralleling their boost to employment, women over 30 years of age also experience a considerably greater boost to earnings compared to women under 30 years of age. Similarly, women in general benefit slightly more than men. In other respects, however, the results differ substantively: although urban jobseekers experience a smaller boost in employment probability than their non-urban counterparts, they experience a considerably larger boost in earnings. A related discrepancy concerns the earnings of individuals according to education level: individuals with higher education level experience a considerably larger boost to earnings than individuals with lower education do (although both groups experience similar boosts to employment probabilities). The two results are related, given that individuals with higher education disproportionally reside in urban areas. Nevertheless, these are two distinct effects, as shown by the divergence in the employment effects, with region, but not education level, playing an important explanatory role in employment probabilities.
5.3.2. Heterogeneous effects from other studies show wide variation in effects across sub-groups of jobseekers
Paralleling the findings for training discussed in the previous chapters, meta‑analyses of the effects of subsidised employment programmes find substantial variation in effects on participant employment across programmes (Card, Kluve and Weber, 2018[2]). This variation may be attributable to a variety of factors, including differences in the target groups, the features of the programme (in particular, the generosity and duration of the subsidy as well as the obligations of employers), as well as differences in the economic environment during which they take place.
Nevertheless, comparing the evaluation results for Lithuania with that of other studies can provide a useful basis for gauging its effectiveness. Figure 5.6 overlays the results in this impact evaluation with the ones included in the Card et al. (2018[2]) meta‑analysis. Similar to the aggregate results, the estimated effects for Lithuania are generally much larger in the short term, and in the lower range of estimates over longer time horizons. Compared to the estimates in the meta‑analysis, the estimates for Lithuania are more consistent across the different demographic groups included in the comparisons. One of the more notable outliers is related to the long-term effects observed for women: this is one of the few sub-group estimates where the comparison studies consistently find more positive results. At the same time, however, it is worth noting that for a specific subgroup not included in the figures separately – women over the age of 50 – the results for Lithuania find very positive employment effects (see discussion in Section 5.3.1 above).
A caveat in interpreting the results concerns underlying differences in the programmes for which the effects of the comparison countries are calculated. It is worth noting that Card et al. (2018[2]) do not examine the effects for different groups within the same programmes, but compare the effects of programmes that are targeted to specific groups with those that are not. Within this framework, Card et al. (2018[2]) find that programmes targeted to women are more effective than the average programme or programmes targeted towards men. They also find that targeted programmes towards older workers, perform better than average.
While differences in target group can matter (as Card et al. (2018[2]) show), differences in programme design are also important. Programmes vary in subsidy generosity and length, which naturally make them more or less attractive to employers. Also important are the obligations imposed on employers after employment subsidies end. For example, in Lithuania employers must keep workers on for six months following the end of the programme or else the employer is excluded from using the scheme for 12 months. This condition may reconcile the discrepancy in the effects over the short, medium and long-term: employment subsidies in Lithuania appear to have stronger effects on employment in the first several months following programme completion, but not over longer horizons.
Figure 5.6. Employment subsidies have heterogeneous effects on employment probability across sub-groups in both Lithuania and other countries
Note: Short, medium and long-term effects respectively refer to effects up to one year, 1‑2 years, and more than two years after programme completion. For Lithuania, results refer to 12, 24 and 36 months after beginning the programme. As such, the observation periods are similar, but potentially not fully aligned. Point estimates are included in the chart even if they are statistically insignificant. Ages refer to “under 30” and “30 and over” for Lithuania. Categories with less than five point estimates in the meta‑analysis are omitted from the above figures. The studies presented adopt various research designs and econometric techniques – the results for Lithuania use nearest-neighbour propensity score matching (for details, see Chapter 3).
Source: Card, D., J. Kluve and A. Weber (2018), “What Works? A Meta Analysis of Recent Active Labor Market Program Evaluations”, https://doi.org/10.1093/jeea/jvx028 and OECD calculations based on data from Lithuanian Employment Service.
5.4. Any direct displacement effects of the subsidised employment programme appear to be small
The discussion in the previous section has considered direct impacts of employment subsidies on participants. However, when designing such programmes it is also important to consider indirect effects whereby employment subsidies can (i) benefit participants at the expense of other jobseekers (substitution); (ii) result in hires of workers that would have occurred even in the absence of the subsidy (deadweight effects); or (iii) displace other non-subsidised jobs in the economy (displacement). Thus, these possible indirect, distortionary effects of such large‑scale government interventions like employment subsidies can adversely affect individuals not participating in the programmes.
From a theoretical point of view, if the demand for labour is relatively fixed, there may be considerable negative spill-over effects from government interventions: treated individuals may benefit, but mainly at the expense of others. For example, the theoretical model outlined by Landais, Michaillat and Saez (2018[5]) shows that an increase in labour supply arising from training or increased search intensity may merely place the worker at the front of the queue for a fixed supply of jobs. This implies that an accurate assessment of the broader effects of ALMPs may need to take into the effects on individuals who are not participating in the programmes. At the same time, relatively few studies have been conducted to examine these effects, and the findings are far from conclusive (see Box 5.1).
The analysis in this section examines the question of whether negative spill-over effects are present within individual firms in the case of employment subsidies in Lithuania. The analysis is facilitated by access to the data on persons in employment (all employment contracts) over the 2018‑20 period in Lithuania. This allows for a broad comparison of the extent to which newly hired workers are entering job positions previously occupied by unsubsidised workers. However, the absence of detailed firm-level information which could allow for estimating whether a firm would be expected to be expanding or contracting means that questions of whether hiring would have occurred even in the absence of subsidies is outside the scope of the analysis. As a result, the analysis does not fully capture potential deadweight effects: the results refer to a more narrow question of whether specific hires are replacing existing workers, not whether these hires would have occurred in the absence of the subsidy.
In the analysis below, job positions are defined based on detailed occupational codes. Shifts in the composition of employment within firms by detailed occupational code from one‑quarter to the next are analysed to ascertain whether individual job positions are being created or destroyed.1 If the composition of employment in a given firm according to occupational codes remains the same from one period to the next, but with changes in the individuals employed there, these changes are construed as replacement flows. To give a simple example, if an individual employed in a given detailed occupation at a given firm at the beginning of one‑quarter is no longer employed in that occupation (and firm) at the next quarter, that job position is deemed to be either (i) replaced, if another individual is then employed in that occupation within that firm, or (ii) destroyed otherwise. Conversely, an individual becoming employed at a firm in an occupation that is not yet present at the firm is construed as a job position being created.
Box 5.1. Research on the indirect effects of ALMPs
The direct effects of ALMPs are widely studied and their effects well-documented – with the meta‑analysis in Card et al. (2018[2]), covering over 200 studies – but the possible unintended, distortionary effects of these large‑scale government interventions are comparably less well-researched. These comprise substitution effects where participants benefit at the expense of other jobseekers, deadweight losses when workers would have been hired even in the absence of an intervention or displacement effects when subsidised jobs displace other non-subsidised ones. Accurately measuring such effects has proven more elusive, with a variety of possible channels through which they take place.
One strand of the literature attempting to measure indirect effects has focused on the effects from a micro-level perspective: exploiting exogenous variation in treatment probability to identify externalities. Crépon et al. (2013[6]) present the results of a randomised experiment designed to evaluate the direct and indirect (displacement) impacts of job placement assistance on the labour market outcomes of young, educated job seekers in France. Exploiting experimental variation in the probability of assignment into the treatment programme, they find that treated individuals had a significantly higher probability of entering stable employment eight months after assignment than untreated individuals. However, the effects were short lived, and appear to have come partly at the expense of untreated individuals, particularly in labour markets where they compete mainly with other educated workers, and in labour markets with above‑average unemployment. Building on the assumption that indirect effects can be expected to be higher wherever a higher share of unemployed individuals are in the treatment group, Attanasio et al. (2017[7]) exploit variation in treatment probability across regions and occupational groups to explore the displacement effects of vocational training in Colombia. In contrast to Crépon et al. (2013[6]), they find no evidence of negative externalities. Blundell et al. (2004[8]) examine the effects of a programme that provided intensive job-search assistance and wage subsidies for jobseekers aged 18‑24 in the United Kingdom. The programme effects are identified by exploiting the fact that area- and age‑based eligibility criteria varied across individuals of identical unemployment durations. Although they find sizable treatment effects of the programme – an increase of 5 percentage points in employment probability in the treatment group compared to a baseline of 26% – they do not find that these led to substitution or displacement effects.
Another strand of the literature has focused on the effects using aggregated data, typically utilising variation across geographic regions for identification. Forslund and Krueger (1997[9]) examine employment in non-subsidised jobs in a geographical area on the number of subsidised jobs lagged one period and other control variables for Swedish construction workers. They estimate a spill over coefficient of ‑0.69: for each worker hired via public subsidy, there are 0.69 fewer private construction workers hired. Dauth et al. (2014[10]) exploit the variation in the ALMP participation rate across regions in Austria over time to examine their effects on the probability of job matches occurring amongst all unemployed. They find evidence of positive spill over effect of ALMPs, including for wage subsidies: in regions with large shares of former participants in this programme a higher number of matches are expected. Examining country-level data from OECD countries for the 1991‑2011 period, Goalas and Zervoyianni (2018[11]) find evidence of a small net positive output-growth differential associated with implementing ALMPs. Furthermore, they find that this differential becomes larger during economic upturns when market conditions are improving relative to trend.
Source: Attanasio, O. et al. (2017), “Vocational Training for Disadvantaged Youth in Colombia: A Long-Term Follow-Up”, https://doi.org/10.1257/app.20150554; Blundell, R. et al. (2004), “Evaluating the Employment Impact of a Mandatory Job Search Program”, https://doi.org/10.1162/1542476041423368; Card, D., J. Kluve and A. Weber (2018), “What Works? A Meta Analysis of Recent Active Labor Market Program Evaluations”, https://doi.org/10.1093/jeea/jvx028; Crépon, B. et al. (2013), “Do Labor Market Policies have Displacement Effects? Evidence from a Clustered Randomized Experiment”, https://doi.org/10.1093/qje/qjt001; Dauth, W., R. Hujer and K. Wolf (2014), “Do Regions Benefit from Active Labour Market Policies? A Macroeconometric Evaluation Using Spatial Panel Methods”, https://doi.org/10.1080/00343404.2014.931571; Forslund, A. and A. Krueger (1997), An Evaluation of the Swedish Active Labor Market Policy: New and Received Wisdom; Goulas, E. and A. Zervoyianni (2018), “Active labour-market policies and output growth: Is there a causal relationship?”, https://doi.org/10.1016/j.econmod.2017.11.019.
Before discussing the relationship between replacement flows and employment subsidies, it is instructive to examine the calculated replacement flows in Lithuania. Given that the relatively short time period spanned by the data (2018‑20) does not provide a long-enough time horizon for a meaningful analysis of their trends over time, the discussion here will focus on the relationship between these flows and firm size. As shown in Figure 5.7, replacement flows increase monotonically with firm size. In firms with over 500 employees, an average of 6% of job positions which are vacated by existing workers are maintained through hires of new workers from one‑quarter to the next.2 Smaller firms tend to experience greater job creation than larger ones, with virtually all such creation in the smallest firms being accounted for by new hires – worker accessions.
Figure 5.7. Larger firms experience more replacement flows
Note: Quarterly calculations made based on changes in employment stocks in the first calendar month of the quarter (e.g. changes for Q2‑2018 reflect shifts from January to April 2018). Calculations are made based on data for 2018‑20. Firm sizes are defined based on the observed number of employees in the SODRA data. Worker accessions are defined as the total number of workers not employed in a given firm in a given quarter who are employed by this firm in the next quarter. Job position creation is defined as the increase in a firm’s total number of workers employed in a given job-position type from one‑quarter to the next, summed across all the job-position types in the firm in the next quarter. Internal reassignments are defined as the total number of workers employed at a given firm in two consecutive quarters who were in a different job-position type in the first quarter than they were in the second.
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
One estimate of the potential size of direct displacement effects – to be interpreted as an upper bound – is the share of hires via employment subsidies which replaced the job positions of existing workers. During the 2018‑20 period, there were 24 000 such accessions of individuals via employment subsidies measured on a quarter-to-quarter basis.3 Of these accessions (hires), 1 200 replaced workers previously employed via an employment subsidy, and 3 900 replaced workers who were not employed via an employment subsidy. Taken together, accessions of workers on employment subsidies replaced existing workers in 22% of cases. This 22% could be interpreted as an estimate of the direct displacement effects of employment subsidies within specific firms. However, as the following discussion will make clear, it should be interpreted as a maximum upper bound.
Another important related finding pertains to the share of worker accessions that constitute replacement flows among workers who do not receive the employment subsidy. Such accessions can constitute a reasonable estimate of the comparison, baseline rate of replacement flows against which to interpret the analogous statistic for subsidised employment. This comparable figure for non-subsidised workers amounted to 38% – meaning that almost two in five non-subsidised hires in Lithuania during the 2018‑20 period replaced the job position of an existing worker. The fact that this share is considerably higher than the respective share for those receiving an employment subsidy casts doubt on the interpretation of the above estimate of replacement flows as reflecting displacement effects.
Given the available data, an alternative estimate of displacement effects can be constructed under a different set of assumptions. The key assumption in this case is that only a subset of occupations are suitable for hiring individuals via an employment subsidy in practice. In this case, assuming a relatively constant share of individuals being employed via employment subsidies as a share of total employment, one would expect that the share of subsidised individuals replacing unsubsidised individuals would equal the share of unsubsidised individuals replacing subsidised individuals. There exists a period where this proportion is indeed constant, from Q3‑2019 through Q1‑2020: during this period, the share of individuals in subsidised employment was almost perfectly stable and amounted to 0.62% of total employment. The replacement flows during this period do not support the presence of any displacement effects, as in fact the number of subsidised individuals replacing unsubsidised individuals was lower than the number of unsubsidised individuals replacing subsidised individuals – 691 to 751, respectively. While far from conclusive, these figures indicate that any displacement effects occurring within firms in Lithuania are likely to be small.
The finding that the pattern of replacement flows in Lithuania is not consistent with the presence of large displacement effects has several possible explanations. First, the conditions for the receipt of the subsidy may provide a disincentive for employers to engage in such strategic behaviour. From July 2017 onwards, employers who dismiss a worker in the six months from the last subsidy payment for that worker are not eligible for further employment subsidies for 12 months. Second, the relatively strong performance of the Lithuanian labour market during the period studied may have made employers wary of dismissing workers just to gain access to employment subsidies. Even with the subsidies, employers still pay a sizable share of the wages of workers hired: the programme subsidises 50% of participant’s wage costs, with a ceiling that amounted to twice the statutory minimum wage during the 2017‑19 period and one and a half statutory minimum wages thereafter. This also limits the financial incentive of employers to replace existing workers. It is worth emphasising, however, that the findings discussed in this sub-section pertain only to displacement effects relating to existing workers. To the extent that employment subsidies result in deadweight effects – job position creation that would have occurred even in the absence of the subsidies – these are not captured in the present analysis.
References
[1] Abadie, A. and G. Imbens (2016), “Matching on the Estimated Propensity Score”, Econometrica, Vol. 84/2, pp. 781-807, https://doi.org/10.3982/ecta11293.
[7] Attanasio, O. et al. (2017), “Vocational Training for Disadvantaged Youth in Colombia: A Long-Term Follow-Up”, American Economic Journal: Applied Economics, Vol. 9/2, pp. 131-143, https://doi.org/10.1257/app.20150554.
[8] Blundell, R. et al. (2004), “Evaluating the Employment Impact of a Mandatory Job Search Program”, Journal of the European Economic Association, Vol. 2/4, pp. 569-606, https://doi.org/10.1162/1542476041423368.
[2] Card, D., J. Kluve and A. Weber (2018), “What Works? A Meta Analysis of Recent Active Labor Market Program Evaluations”, Journal of the European Economic Association, Vol. 16/3, pp. 894–931, https://doi.org/10.1093/jeea/jvx028.
[6] Crépon, B. et al. (2013), “Do Labor Market Policies have Displacement Effects? Evidence from a Clustered Randomized Experiment *”, The Quarterly Journal of Economics, Vol. 128/2, pp. 531-580, https://doi.org/10.1093/qje/qjt001.
[10] Dauth, W., R. Hujer and K. Wolf (2014), “Do Regions Benefit from Active Labour Market Policies? A Macroeconometric Evaluation Using Spatial Panel Methods”, Regional Studies, Vol. 50/4, pp. 692-708, https://doi.org/10.1080/00343404.2014.931571.
[12] Davis, S., J. Haltiwanger and S. Schuh (1996), Job Creation and Destruction, MIT Press.
[4] ESTEP (2019), Aktyvios darbo rinkos politikos priemonių bei tarpininkavimo įdarbinant ir įdarbinimo tvarumo stebėsenos ir vertinimo sistemos sukūrimo paslaugos.
[9] Forslund, A. and A. Krueger (1997), An Evaluation of the Swedish Active Labor Market Policy: New and Received Wisdom, University of Chicago Press.
[11] Goulas, E. and A. Zervoyianni (2018), “Active labour-market policies and output growth: Is there a causal relationship?”, Economic Modelling, Vol. 73, pp. 1-14, https://doi.org/10.1016/j.econmod.2017.11.019.
[5] Landais, C., P. Michaillat and E. Saez (2018), “A Macroeconomic Approach to Optimal Unemployment Insurance: Applications”, American Economic Journal: Economic Policy, Vol. 10/2, pp. 182-216, https://doi.org/10.1257/pol.20160462.
[3] PPMI (2015), Final Report on Counterfactual Impact Evaluation of ESF Funded Active Labour Market Measures in Lithuania, https://doi.org/10.13140/RG.2.2.12469.83689.
Annex 5.A. Additional figures
Annex Figure 5.A.1. Estimated effects of employment subsidies on employment probability, employment duration and wages by gender
Note: The analysis presents nearest-neighbour propensity score matching results which matches individuals based on a number characteristics: duration of unemployment, age, gender, education, unemployment benefit receipt and level, language and other skills, municipality, employability and barriers to employment, as well as prior employment history and earnings. For every individual in the treatment group, the matching is conducted based on the values of these characteristics in the calendar month when the individual enters the programme. The control group is comprised of individuals with similar characteristics not entering ALMPs in that same calendar month. For paired individuals in the treatment and control groups, this calendar month is then the reference point after which outcomes are measured. The analysis is restricted to the region of common support. The standard errors are calculated based on the adjustment proposed by Abadie and Imbens (2016[1]). The confidence intervals are shown at the 5% level of significance and represented by the whiskers delimiting the dotted lines on the charts.
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
Annex Figure 5.A.2. Estimated effects of employment subsidies on occupational mobility, cumulative earnings and cumulative earnings net of subsidies by gender
Note: The analysis presents nearest-neighbour propensity score matching results which matches individuals based on a number characteristics: duration of unemployment, age, gender, education, unemployment benefit receipt and level, language and other skills, municipality, employability and barriers to employment, as well as prior employment history and earnings. For every individual in the treatment group, the matching is conducted based on the values of these characteristics in the calendar month when the individual enters the programme. The control group is comprised of individuals with similar characteristics not entering ALMPs in that same calendar month. For paired individuals in the treatment and control groups, this calendar month is then the reference point after which outcomes are measured. The analysis is restricted to the region of common support. The standard errors are calculated based on the adjustment proposed by Abadie and Imbens (2016[1]). The confidence intervals are shown at the 5% level of significance and represented by the whiskers delimiting the dotted lines on the charts.
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
Annex Figure 5.A.3. Estimated effects of employment subsidies on occupational mobility, by jobseeker characteristics
Note: The analysis presents nearest-neighbour propensity score matching results which matches individuals based on a number characteristics: duration of unemployment, age, gender, education, unemployment benefit receipt and level, language and other skills, municipality, employability and barriers to employment, as well as prior employment history and earnings. For every individual in the treatment group, the matching is conducted based on the values of these characteristics in the calendar month when the individual enters the programme. The control group is comprised of individuals with similar characteristics not entering ALMPs in that same calendar month. For paired individuals in the treatment and control groups, this calendar month is then the reference point after which outcomes are measured. The analysis is restricted to the region of common support. The standard errors are calculated based on the adjustment proposed by Abadie and Imbens (2016[1]).
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
Annex Figure 5.A.4. Estimated effects of employment subsidies on earnings by jobseeker characteristics
Note: The analysis presents nearest-neighbour propensity score matching results which matches individuals based on a number characteristics: duration of unemployment, age, gender, education, unemployment benefit receipt and level, language and other skills, municipality, employability and barriers to employment, as well as prior employment history and earnings. For every individual in the treatment group, the matching is conducted based on the values of these characteristics in the calendar month when the individual enters the programme. The control group is comprised of individuals with similar characteristics not entering ALMPs in that same calendar month. For paired individuals in the treatment and control groups, this calendar month is then the reference point after which outcomes are measured. The analysis is restricted to the region of common support. The standard errors are calculated based on the adjustment proposed by Abadie and Imbens (2016[1]).
Source: OECD calculations based on the Lithuanian Employment Service and Lithuanian State Social Insurance Fund Board.
Notes
← 1. The terminology parallels that in the job creation and job destruction literature, which refers to employment expansion and contraction at the firm level (Davis, Haltiwanger and Schuh, 1996[12]).
← 2. Note that internal reassignments – individuals being reassigned to a different job position but remaining within the same firm – appear to play a negligible role in practice in Lithuania (although this empirical result may be due to measurement error: under-reporting of changes within an employer).
← 3. These statistics are somewhat smaller than the number of individuals included in employment subsidies because (i) changes are in fact measured only for 11 quarters during the 2018‑20 period, and (ii) individuals who had employment spells shorter than three months may not be captured in the statistics.