1 Evaluation of Active Labour Market Programmes in ...

25 downloads 16584 Views 83KB Size Report
Feb 28, 2003 - Labour market training, subsidized employment and business start-up .... the active labour market programmes the number of participants ...
Evaluation of Active Labour Market Programmes in Estonia

Draft

Reelika Leetmaa Andres Võrk

University of Tartu, Estonia

28 February 2003

Abstract The paper analyses the net impact of active labour market policies on employment and wages in Estonia. Labour market training, subsidized employment and business start-up grants programmes are considered. The analysis is based on micro- level datasets from administrative records and the data gathered through follow-up survey of the nationally representative samples of unemployed conducted in Fall 2002. The administrative records contain information on people’s participation in various active and passive labour market programmes and the characteristics of the programmes. The data from the follow-up survey allows us to learn about the labour market status of the unemployed over the period from January 2000 to September 2002. The employment outcomes of the participants are compared with the control group of unemployed with the similar characteristics. We used statistical matching based on propensity score to construct the control group. The study shows positive and statistically significant impact of the active labour market programmes on employment probability. The magnitude of the effect is around 7 per cent after one year of the completion of the programmes. Analysis also shows that the effect of labour market programmes is relatively homogenous within different socio-demographic groups and geographical regions. Theme: Labour market policy Keywords: Active labour market programmes, propensity score matching, Estonia JEL-code: J68

1

1. Introduction At the beginning of transition open unemployment was practically non-existent in the former centrally planned economies of Central and Eastern European countries. The introduction of political, economic and social reforms that started between 1989 and 1992 has led to the profound structural changes and the emergence of the labour surplus. Unemployment has risen dramatically and remains between 15% and 20% in Poland, Slovakia, Bulgaria and Lithuania, between 12% and 15% in Latvia and Estonia and under 8% in Hungary, Czech Republic, and Romania (Estonian Labour Force Surveys, 2001). Since unemployment was practically non-existent in the former centrally planned economies before 1990s, the regulatory framework for labour market policies and a network of employment offices, where unemployed persons could register, receive labour market services and collect benefits had to be created from the scratch. Currently both unemployment compensation systems and the wide array of active labour market programmes (ALMPs) comparable to those found in developed market economies are found in transition countries and in Estonia. The menu of active labour market programmes includes placement services, retraining, employment subsidies and job creation measures. Active labour market policies in general suffer from a very serious lack of resources. The public expenditures on ALMPs measured as a percentage of GDP, can be five to ten times less than the amounts devoted to similar measures in EU countries with comparable or even lower unemployment rates. While the empirical literature on evaluations of ALMPs is vast and the experiences of developed market economies are well summarized by the studies of OECD (see Fay (1996); Martin (1998)) and World Bank (see Dar and Tzannatos (1999)), the evidence on transition countries is scarce. The most important reason behind this is probably the inadequate data sets, especially the lack of long-term data in transition countries. The major study on transition countries experiences with ALMPs was the Cross Country Evaluation of Active Labour Programmes initiated by the World Bank in 1997. The countries studied were Czech Republic, Poland, Hungary and Turkey. In addition to the cross-country studies mentioned above, the effectiveness of ALMPs has been recently analyzed in Slovakia by Lubyova and van Ours (1998), in Poland by Kluve, Schmidt and Lehmann (2002) and in Bulgaria by Kotzeva (2000). Lubyova and

2

van Ours reported the positive treatment effect of retraining and publicly useful jobs, but negative effect on the transition rate from unemployment to employment for socially purposeful jobs in Slovakia (Lubyovaa and van Ours, 1998). Kluve et al found that Polish training and retraining programmes had a positive effect on the employment probability of the participants. However, the public works and intervention works had negative treatment effects on the employment rate of the men and no effect on women’s employment probability (Kluve et al, 1998). The results of the Bulgarian study show positive net impacts for all ALMP-s, although the positive impact of the temporary employment program is negligible (Kotzeva, 2000). In this paper we complement previous studies with an analysis of the effectiveness of ALMPs in Estonia. We will base our analysis on micro- level datasets using both administrative records available in Estonian Labour Market Board and data from a follow-up survey conducted by the Statistical Office of Estonia in Fall 2002. The administrative records give us information on people’s participation in various active and passive labour market programmes and the characteristics of the programmes. Within the follow-up survey a sample of about 2000 people who received active labour market policy measures or unemployment benefits in year 2000 were interviewed to learn their labour market status over the period from January 2000 to September 2002. The sample size was about 4% of registered unemployed people, whic h allows us to generalize the results on all registered unemployed. To evaluate the impact of ALMP-s we will rely on the non-experimental methodology of program evaluation. The parameter of interest for us is the mean effect of treatment on the treated population. The mean effect is measured as the difference between the outcomes of participants and comparable non-participants, who serve as a control group. In our case participants are those people who received active labour market measures, and non-participants are those who received only unemployment benefits. We use linear regression models and statistical matching of pairs. The survey data allow us to use age, gender, educational level, nationality and place of residence for matching. As an outcome both employment probability and wages are used. The paper proceeds as follows. Section 2 gives an overview of dynamics of unemployment, institutional setup and the measures of active labour market policies in Estonia. Section 3 describes the methodology that we use in our empirical study. After

3

that an overview of the data that we used are given in section 4. Section 5 presents empirical results and main findings are summarized in the conclusion.

2. Unemployment and Active Labour Market Policies in Estonia In this section we will give a brief overview of the dynamics of unemployment, institutional setup and the measures of active labour market policies in Estonia. For more detailed analyses of the development of Estonian labour market since the beginning of transition see Eamets et al (2001). The economic transition has led to a deterioration of the labour market situation. Labour force participation rates have been constantly declining since the beginning of the transition process. The average labour force participation rate in 2001 was 70.1% (see Table 1), which is approximately 8 percentage points lower throughout. However, the indicator is comparable to the EU average (69,2%) in 2001. The second row in Table 1 presents the dynamics of employment rates, which have been sharply declining from an initial 77.4% to 61.1% in 2001. The current level is approximately 16 percentage points lower than in the beginning of 1990s. Compared to the EU average in 2001 (63,9%) the employment rate in Estonia is almost 3 percentage points lower. The third row documents the development of unemployment rates in the Estonian labour market since 1991. Unemployment rose from virtual absence in 1991 to a more or less constant level around 9% on average in 1995-1998 and climbed up to an average of 13.8% in 2000, presumably due to the Russian crises. In 2001 we observe a modest decline of 1 percentage point in unemployment rate compared to previous year, but the level remains remarkably higher than the respective indicator in EU. Note that the data we analyse in our ALMPs evaluation covers the year 2000, when the Russian crises occurred. The evaluation literature suggests that during the cyclical downturn or a shock the cause for unemployment is demand deficiency and hence, the efficiency of active labour market policy is reduced. Programmes are generally most effective when the economy is growing and the programmes are well targeted. (Dar and Tzannatos, 1999) These developments have been accompanied with a substantial increase in the number of inactive persons, which has grown from 304,800 (1992) to 330,100 (2001). Within the inactive population the number of discouraged persons has risen from 10,500 in 4

1993 to 22,300 in 2001. In addition unemployment varies between regions and subgroups. The groups hardest hit are young persons looking for they first job, nonEstonian speakers and older workforce. Table 1. Main Labour Market Indicators in Estonia 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 Labour force participation rate % (15-64)

77.9 77.0 75.1 73.6 73.8 72.6 72.2 72.3 71.7 70.3 70.4 70.1

Employment rate % (15-64)

77.4 75.8 72.3 68.7 68.1 65.5 64.9 65.2 64.5 61.6 60.7 61.1

Unemployment rate 0.6 1.5 3.7 6.7 7.7 9.7 10.0 9.8 %* * 1990-1992 population aged 15-69, 1993-2001 population aged 15-64 Source: Estonian Labour Force Surveys

10.0 12.4 13.8 12.8

Table 2 presents the main labour market indicators in the regions observed in microeconometric analysis. The regions selected represent both the counties characterized by high unemployment and low employment (Ida-Virumaa) and counties close to national average. In addition the regions are spread throughout the country and dominated by different economic activity (industry in Ida-Virumaa (51,5% of the employed), agriculture in Viljandi (19,9% is the highest share in Estonia) and services in Tallinn and Tartu (over 70,3% and 63,5%) in 2001). Table 2. Main Labour Market Indicators in the regions analyzed (population aged 15-74) Labour force participation rate, %

Employment rate, %

Unemployment rate, %

Tallinn Ida-Virumaa Tartu

68.8 61.9 61.4

60.4 48.8 54.4

12.2 21.1 11.4

Viljandi

63.5

56.3

11.4

54.7

13.6

Estonia average 63.3 Source: Estonian Labour Force Surveys

In order to counteract the adverse effects of unemployment the Government has launched a range of active labour market policies. The responsibility for the development of national labour market policy and active labour market programmes rests with the Ministry of Social Affairs (MOSA). The Estonian Labour Market Board (ELMB) established in May 1990 is responsible for the implementation of the active labour market policies as well as for registration of the unemployed and administration

5

of unemployment benefits. The ELMB has a network of local employment services operating in all 15 counties and in capital city Tallinn. In 2001 there were 36 local employment offices with total staffing of 207 people. Regarding the legal framework, the provision of active labour market policies and payment of unemployment benefits was regulated by Governmental decree until the end of 1994. In 1995 the first Social Protection of the Unemployed Act came into force, which was amended by the Social Protection of the Unemployed Act and the Employment Service Act in October 2000. The main changes brought by new acts were widening the access to labour market services by redefining the unemployed and introducing the new service – vocational counselling. Despite of the widened access to the active labour market programmes the number of participants remained stable, because no additional funding was foreseen for financing the programmes. The programmes examined in the study operate similarly under the old and the new legislation. Therefore the legislative changes do not affect the results of our analyses. The Employment Service Act specifies seven types of active labour market programmes in Estonia: •

Information on the situation in the labour market and the possibilities of employment;



Employment mediation;



Vocational training;



Employment subsidy to start a business (business start-up grant);



Employment subsidy to employers to employ less competitive unemployed persons (wage subsidy);



Community placements;



Vocational guidance (public works).

Out of these we examine the impact of vocational training, wage-subsidies and start-up grants. Community placements were left out, because since 2001 they are no longer financed from the state budget. Instead, community placements are financed by local municipality budgets. As no additional recourses were allo cated to the local level, the interest to organize public works and the number of participants has declined considerably (see Table 3).

6

Table 3. Number of participants in active labour market programmes (1995-2001) Program Participants in employment training Found job after training (%) Employed with subsidies to employer Received business start-up subsidy Participants in community placement Total number of participants in ALMP-s

1995 9,809 52.9 121 459 5,741 16,130

1996 9,434 64 246 456 4,089 14,228

1997 8,241 75.8 216 434 4,661 13,552

1998 7,956 61.9 136 380 3,771 12,243

1999 7,027 56.2 265 433 3,667 11,366

2000 8,150 57.3 189 413 4,177 12,929

2001 10,232 52 356 421 125 11134

Source: Estonian Labour Market Board

The total public expenditures on labour market policies measured as a percentage of GDP have been very low in Estonia ranging from 0,2% to 0,3% of GDP. The comparable indicator in EU countries was 2,6% in 2000. The expenditures on active labour market policies in Estonia account for only 0,08% of GDP, which is more than ten times less than the amount spent in EU countries with comparable or even lower unemployment rates. In 2001 only 26,8% of the total spending was dedicated to active measures in Estonia. Out of the active measures labour market training is the most important both in terms of expenditures and in terms of participants. Employment training may take the form of 1) vocational training or 2) more general training aimed at providing information on the labour market situation and psychological preparation for competing in the labour market. Most of the participants belong to the first group. Training is organized by the local labour offices. The duration of the courses is limited up to 6 months. In 2001 the average duration of the training course was 25 days. Participants in training receive a retraining allowance equal to 1,5 times unemployment benefit. Participation in labour market training decreased steadily between 1995 (9,809 participants) and 1999 (7,000 only). However, sinc e 1999 the number of participants has increased reaching the highest absolute figure since 1995 in year 2001 (10,200). Hence, comparing the relative numbers, we see that whereas in 1995 about 28% of registered unemployed participated in training, in 2001 the respective number was only 19%. This can be explained by the increasing training costs. The business-start-up grant is the second largest measure in terms of expenditures. However, only a small number of unemployed (between 380 and 460 annually since 1995) participated in the program. To apply for start-up subsidy the unemployed must be at least of 18 years of age and have undergone relevant training or showing

7

“sufficient” experience. The upper ceiling of the subsidy amounts to 10,000 EEK (since the beginning of 1998), which is about 640 Euro only. Wage subsidy to the employer for recruiting less competitive persons has been the least important active measure both in terms of expenditure and participants (only between 120 and 350 unemployed annually since 1995). The following persons who are registered as unemployed are considered to be less competitive in the labour market: disabled persons, pregnant women and women who are raising children under six years of age, young people aged 16-24, persons who will be retiring within 5 years and persons who have released from prison. The level of the wage subsidy is 100% of the minimum wage during the first 6 months and 50% of the minimum wage during the next 6 months of her/his employment period. Table 4. Expenditures on labour market policies in Estonia 1993 1994 1995 1996 1997 1998 1999 2000 Expenditures on labour market policies as a % of GDP Total on labour market policies 0.19 0.24 0.17 0.17 0.17 0.16 0.34 0.33 Passive labour market policies 0.17 0.11 0.07 0.07 0.08 0.08 0.26 0.25 Active labour market policies 0.02 0.13 0.10 0.09 0.09 0.08 0.08 0.08 Expenditures on active and passive policies as a % of total expenditures Passive labour market policies 88.6 45.8 40.5 44.8 47.3 49.9 75.7 77.0 Active labour market policies 11.4 54.2 59.5 55.2 52.7 50.1 24.3 23.0 Breakdown of the expenditures on active labour market policies, % Labour market training 62.5 55.5 57.9 59.6 55.4 59.3 58.1 Community placement 5.2 2.7 4.6 4.1 3.3 5.2 4.9 Start-up grants 4.7 9.1 7.7 7.1 6.4 6.6 6.6 Subsidy to employer 0.5 0.9 1.7 1.7 1.9 2.8 3.2 PES administration 27.2 31.9 28.2 27.5 33.0 26.1 27.1 Source: Estonian Labour Market Board, Statistical Office of Estonia

2001 0.30 0.22 0.08 73.2 26.8 63.7 0.0 5.3 4.1 26.9

To conclude, the Estonian labour market is characterized by high unemployment and low employment rates. Furthermore, the number of inactive people has risen considerably compared to the beginning of transition period. The expenditures on labour market policies are very low compared to EU countries as well as the other transition countries in Central and Eastern Europe.

3. Evaluation methodology In this section we briefly discuss the approach that we use to evaluate ALMPs. Because of the small scale of the active labour market programmes, we can hardly find any effect

8

on these measures on employment on macro level. Therefore our approach is to use micro level analysis. The questions we ask are: 1) Does participation in ALMPs increase the chances to find a job? 2) Does participation in ALMPs lead to higher labour earnings? To answer these questions we compare the labour market status and wages of those previously unemployed people who participated in one of three active labour market programmes with those unemployed who did not. Given that the two groups of people are similar with the respect of all other characteristics, we can assign the difference in their labour market outcomes to active labour market programmes. Next we briefly present this formally drawing heavily on Sianesi (2001) and Heckman, Lalonde and Smith (1998). We are interested in the effect of a labour market program on an outcome Y of all unemployed people, who constitute our population, compared to the absence of the program. In our case the population is all registered unemployed people in Estonia in year 2000. We use the following notation:

Y1 – potential result when a person participates in the program, Y0 – potential result when a person does not participate in the program, person participat es in the program 1, D = , 0, person does not participat e in the program X – personal characteristics that affect labour market outcome and potentially participation in labour market programmes, but which are not influenced by the program itself (for example, sex, ethnicity, general education, place of living, etc.). Using the notation above, we can write the observed outcome for an individual i: Yi = (1 − Di )Y0i + DiY1i = Y0 i + Di (Y1i − Y0 i )

(1)

In what follows, we employ the so-called stable unit-treatment value assumption (SUTVA), which means we assume that there are no displacement effects or general equilibrium effects. It means that an individual's decision to participate in ALMPs must not depend on the decisions of other individuals. Also, the potential result of an

9

individual must not depend on the participation of other individuals. Because the scale of the programs is small in Estonia, the assumption should not very strict. The effect of the active labour market programmes or treatment effect for an individual i is defined as the difference of the potential results Y1i − Y0 i . Because we cannot observe a person in two states, we aim only to estimate the average treatment effect on treated: ATET = E (Y1 − Y0 | D = 1) = E (Y1 | D = 1) − E (Y0 | D = 1)

(2)

ATET shows the effect of the program for those people who actually participated. Because we cannot observe the last term of the equation (2) E (Y0 | D = 1) – the average outcome of those people who participated in the program if they had not participated – we have to construct it. First we may use the non-participants actual outcome as the counterfactual: E (Y0 | D = 1) = E(Y0 | D = 0)

(3)

We know that both because of observed and potentially unobserved differences in the characteristics of the participants and non-participants we cannot rely on the equality (3). We therefore use the conditional independence assumption (CIA), which states that given the observed characteristics X, the outcome of participants if they had not participated is equal to the actual outcome of non-participants: E (Y0 | D = 1, X ) = E (Y0 | D = 0, X )

(4)

To find the ATET we have to find the average over X based on the distribution of X among participants. ATET = E (Y1 | D = 1) − E (Y0 | D = 1) =

= E X [{E(Y1 | D = 1, X ) − E (Y0 | D = 1, X )} D = 1] = E X [{E (Y1 | D = 1, X ) − E (Y0 | D = 0, X )} D = 1]

CIA

(5)

= E X [{E(Y | D = 1, X ) − E (Y | D = 0, X )}D = 1]

Whether the CIA is satisfied in our case is open to discussion. Because both our participants and non-participants are registered unemployed from the same time period, and we control for the other main socio-demographic characteristics that may influence probability of working, we hope that we can rely on the CIA.

10

Because of the dimensionality problem, we estimate the effect of ALMPs using propensity score matching. It means instead of conditioning on X, we condition on the probability to participate in the program, which is the function of X: E (Y0 | D = 1, Pr ( D = 1 | X )) = E (Y0 | D = 0, Pr (D = 1 | X )) .

(6)

The propensity score matching does not rely on the functional relationship between probability of work and explanatory variables. We use one-to-one matching with replacement, where propensity score is found using a logistic regression in a such a way that the main observed socio-demographic characteristics are balanced in the treated and non-treated samples. In addition to propensity score matching we also estimate linear regression models of employment probability and net wages, where one of the explanatory variables is the ind icator of participation in ALMPs.

4. Data description We use data gathered through follow-up survey of the nationally representative sample of unemployed conducted in September 2002. The population for our study contained all registered unemployed who in 2000 participated either in active or passive labour market programmes or both (see Figure 1). The data from the follow- up survey contained basic socio-demographic characteristics, information on participation in active and passive labour market programmes, self-evaluation of the effectiveness of these programmes, and labour market status and wages over the period from January 2000 to September 2002. Figure 1. Survey design

Active and passive measures Unemployed

Measure Passive measures

Dec 2000

Jan 2000

Sept 2002

Time

Survey with retrospective part

11

Because there was no central database in Estonian National Labour Board until year 2002, the availability of the data from the local employment offices determined the four regions – Tallinn, Ida-Viru, Tartu, Viljandi – that we could use in our analysis. Because these include major regions in Estonia and they have different unemployment levels and industrial structure (see also table 2), we believe that they represent the Estonian labour market and the functioning of active labour market programmes well. For these regions 22292 people remained as a study frame after removal of missing or incomplete records. The administrative records contained information on people’s participation in the programmes,

characteristics

of

the

programmes

and

main

socio-economic

characteristics of the unemployed. On the basis of these administrative records a systematic sample of the size 3024 people was drawn and 1839 people were actually interviewed (see Table 5). Table 5. Sample construction

Received ALMPs – treated Labour market training Start-up grants Subsidy to employer Received unemployment benefits only – controls Total

(1) Population of registered unemployed in 2002 2121 1968 101 52 20170

(2) Sample drawn

(3) Interviewed

(4) Response rate

1013 860 101 52 2011

637 540 64 33 1202

63%

22292

3024

1839

61%

60%

The relatively low response rate is mostly due to inadequate addresses that were recorded in the labour board office databases. In 63% of cases there were no contact, 19% refused to answer, and 18% had other reasons. We have not analysed yet if there is any systematic pattern related to dropout. Most of the people who participated in ALMPs had labour market training; the share of start-up grants and subsidies was small (Table 5, column 3). It means when not distinguishing between different active policy programmes, we essentially analyse the effect of training. There is a small discrepancy what administration records say on participation and what those surveyed answered (compare Table 5 column 3 and Table 6). The larger number of people having received labour market training according to survey compared to registry data is due to different time span, but either recording or recalling errors must

12

be present in case of smaller number of start-up grants and subsidies in Table 6 than in Table 5. In what follows we use people's own answers both on participation and the beginning and end dates of active and passive labour market programmes in our analysis; only when analysing the effectiveness of different programmes we use indicators both from registry and survey. Table 6. Participation in ALMPs according to the survey answers

ALMP Labour market training Start-up grants Subsidy to employer Control group (non-participants in ALMPs) Received unemployment benefits Total

Number of people 730 647 69 14 1109 1646 1839

There were 96 people who according to their answers had not participated neither in active nor passive labour market programmes. In our analysis they were dropped when using participation indicators from survey data. In addition, we used only those people whose participation in the active labour market programmes had ended at least by the end of 2000. Also, the reception of unemployment benefits during the spell that they had in 2000, had to end by that year. We do allow that people received unemployment benefits again in 2002. All these reduced our sample to less than 1600 people. The end month of the participation in active labour market programmes was taken the latest of the following three: 1) the last month of labour market training, 2) the last month of work with subsidy to employer, 3) the month of receiving a start-up grant. Most of the registered unemployed received also unemployment benefits, i.e. passive measures. The end months of policy measures of those who received both the active and passive measures was taken the latest.

13

5. Empirical results In this section we present the results of our analysis starting with simple two-way tables of past participation in ALMPs and future labour market participation, in the whole sample and within different socio-demographic groups. Then we estimate linear regression models to find the effect of ALMPs on the probability of being employed and net wages after controlling for other covariates, and finally we use propensity score matching. Because of the survey design, all the distributions and estimated models presented below are calculated using the probability weights, unless emphasized otherwise. All the empirical estimation is done in Stata™ version 7. 5.1. Two -way tabulations Those people who participated in the ALMPs in 2000 had 12% higher probability of working compared to non-participants in September 2002 (Table 7). Because we do not control here for other confounding variables, which usually reduce the effect due to selfselection and cream skimming, this may be considered as an upper bound of the estimate of the impact. Table 7. Participation in ALMPs in 2000 and the distribution of labour market status in September 2002 Status Employed Unemployed Inactive Total Design-based Fstatistic and p-value

Partic ipated in ALMP 69% 22% 9% 100% 7.9 (p=0.000)

Did not participate in ALMPs 57% 28% 15% 100%

Difference 12% -6% -6%

According to point estimates most effective active labour market programme is start-up grants (Table 8). The probability of being employed is about one third higher for those who received business start- up grants compared to those not participating in any of the ALMPs. Next come subsidies to employer, and then labour market training. The former is, however, insignificant because of the so small sample size. 1

The 14 observations of subsidies to employers did not allow estimating the impact using the indicators from survey data. 1

14

Table 8. Impact of different measures on the probability of being employed in September 2002+ Impact Participation and time periods according to survey data Labour market training 12% *** Start-up grants 35% *** Participation and time periods according to registry data Labour market training 11% *** Start-up grants 31% *** Subsidy to employer 14% + Here and afterwards three stars *** means that the design-based F test-statistic, which tests the differences between the distribution of the labour market states between two groups is significant at the significance level 0,01; ** significant at 0,05; * significant 0,10.

The self-evaluation of the participants in labour market training confirms that training has a positive effect on employment probability – two thirds of the participants considered their course useful (Table 9).

Table 9. "Did the participation in the course increase your chances to find a job?" Number of people Yes No Total

468 226 694

Unweighted share (%) 67 33 100

The analysis so far has used non-participants' actual outcome as the counterfactual for the treated, but this yields incorrect results when the treated and non-treated are different with respect to other characteristics that influence the potential outcome. It is not clear, whether those participating in ALMPs have better labour market prospects on average in our sample. Those having participated in the ALMPs were younger, among them were more women, and more Estonians, and they had more children and higher education levels (Table 10). Table 10. Participation in ALMPs and socio-demographic characteristics

Average age Share of men Share of Estonians Average number of children Distribution of education Up to basic education Basic education + vocational training Secondary education Secondary education + vocational training Higher education

15

Did not participate in ALMPs 40.3 46.8% 30.6% 0.57 9.8% 7.2% 25.3% 47.2% 10.6%

Participated in ALMPs 36.6 33.4% 39.9% 0.66 5.4% 1.6% 24.9% 50.9% 17.2%

Next we find the effect of ALMPs within different socio-demographic groups (Table 11). Because of the relatively small sample, not all the effects are statistically significant, but we see that point estimates are relatively similar within different sociodemographic groups. There is one negative value -26 in the education group 'basic education +vocational training', but because the group is really small one, we should not take tha t seriously. Table 11. Impact of ALMP within different groups on the probability of being employed in 2002 September conditional that the labour market measure ended by December 2000. Impact (%) Gender and ethnicity Estonian males Estonian females Non-Estonian males Non-Estonian females Age groups Aged 18-30 Aged 31-40 Aged 40-66 Regions Tallinn (capital city) Ida-Viru county Tartu county Viljandi Education groups Up to basic education Basic education + vocational training Secondary education Secondary education + vocational training Higher education

13 16*** 15* 9 11 12 14 11** 8 29*** 11 15 -26 17** 8* 11

5.2. Regression models Above we conditioned on Xs one at a time, next we estimate the effect of participation in ALMPs on the probability of being employed and on wages controlling for all sociodemographic variables by linear regression models. 2 We used different time periods to fix the employment status. The basic regression uses September 2002, but we also run regressions for time periods from 1 to 12 months after the end of labour market programmes. The basic forms of our regression models are given below (equations (7) and (8)). Because labour market programmes ended at different calendar time periods, we 2

Using logit or probit model to estimate the probability of working did not change the marginal effect of ALMP indicator at average values of the other exogenous variables.

16

introduce dummies for calendar months (from January 2000 to September 2002) to control for seasonality and plausible trends effects. Employed = β 0 + β1 ALMP + β 2 Male + β3 Estonian + β 4 Age + β5 Age2 + + Education Dummies + County Dummies + ( Dummies for calendar months) Wage = β 0 + β1 ALMP + β 2Male + β3 Estonian + β 4 Age + β 5 Age 2 + + EducationDummies + CountyDummies

(7)

(8)

The results of the linear models show that the effect of ALMP on the probability of being employed is 11% and it is significant (Table 12). The effect on monthly net wages is about 300 kroons, but this result is due to the differences on employment. If comparing only people who work, then ALMP does not have any effect. Tabel 12. Linear regression models of probability of being employed and net monthly wage on September 2002. + (1)

(2)

Probability of Net wage being employed ALMP 0.110 296.1 (3.53)*** (2.19)** Male 0.092 914.8 (2.78)*** (6.34)*** Age 0.036 148.3 (3.32)*** (3.76)*** Age squared 0.045 204.6 (3.25)*** (4.27)*** Estonian 0.090 682.2 (2.35)** (3.85)*** Ida-Viru county 0.014 539.4 (0.38) (4.21)*** Tartu county 0.052 583.3 (0.97) (2.52)** Viljandi county 0.081 107.1 (1.31) (0.25) Basic education + 0.135 402.6 vocational training (1.41) (1.25) Secondary education 0.125 528.3 (1.84)* (2.04)* Secondary education + 0.183 650.6 vocational training (2.80)*** (2.55)** Higher education 0.220 1074.5 (2.79)*** (3.20)*** Intercept 0.310 1709.2 (1.56) (2.07)** Observations 1567 1502 F-statistic (p-value) 5.17 (0.00) 8.2 (0.00) 2 R 0.056 0.125 + Absolute values of t-statistics in parentheses, *** – significant at 0.01, ** – significant at 0.05, * – significant at 0.10

17

We also tested for the presence of all interactive effects between ALMP indicator and the other explanatory variables to allow for heterogeneous effects, but we could not reject the hypotheses that these interactive effects were insignificant individually and together. Therefore the effect of labour market schooling (which dominates in the programmes) is homogeneous in different socioeconomic groups. Next we estimated the linear probability models for twelve different time periods after the end of labour market programmes (end of active labour market measures or end of unemployment benefits). The covariates used in the model were similar to these in Table 12. In addition we included time calendar dummies. The point estimates of ALMP variable, together with 2 times standard errors are graphed in Figure 2. Figure 2. The impact of ALMPs on the probability of being employed over time after controlling for other covariates

Impact of ALMP on employment probability

20% 15% 10% 5% 0% 1

2

3

4

5

6

7

8

9

10

11

12

-5% -10% Months after the end of labour market policy programmes

As we see, the effect of ALMPs turns to be significant after four months of the end of labour market programme and stays at around 10% level thereafter.

5.3. Propensity score matching Finally we use propensity score matching to evaluate the active labour market programmes. The estimation is done with Barbara Sianesi's psmatch ado-file written for Stata. We use one-to-one matching with replacement, that is controls were allowed to be used as matches several times. The matching was based on the propensity score that was estimated using logit model including the same variables that regression models above. We did not use sample weights when estimating the propensity score, because the aim of the estimation was only to get similar control group to treated group with respect to

18

observable characteristics. We dropped the treated observation for which there was not any controls within the range of 0,1% to participate. The results are presented in the table 12. The matching on the propensity score which best balanced the observed characteristics yields the impact of 6.7% on the probability of working and 150 EEK on the average net wage.

Table 13. The effect of ALMPs on the probability of working in September 2000 using propensity score matching.

Average result of treated Probability of being employed

0.694

Average result of controls 0.627

Net wage

1890.8

1746.5

Difference (appr. t-statistic) 0.067 (1.81) 144.3 (0.90)

The effect of ALMP over time is given in Figure 3. We see similar pattern as before – the effect turns to be positive 4 months after the end of programmes and it reaches 7-8% during the year after the program. Figure 3. The impact of ALMPs on the probability of being employed using propensity score matching

Impact of ALMP on employment probability

10% 8% 6% 4% 2% 0% -2%

1

2

3

4

5

6

7

8

9

10

11

12

-4% Months after the end of labour market policy measures

It is difficult to explain the initial insignificant or even negative impact of ALMPs. One reason might be that participation in labour market training increases people reservation wage and therefore they do not accept the job immediately. Another explanation might be simply that the search process takes time, and only half a year later people start finding suitable jobs.

19

6. Conclusion In this paper we analysed the effectiveness of active labour market programmes in Estonia. Labour market training, subsidized employment and business start- up grants programmes were considered. The study shows positive and statistically significant impact of the active labour market programmes on employment probability. The effect turns to be positive 4 months after the end of programmes and it reaches 7-8% during the year after the program. Analysis also suggests that the effect of labour market programs is relatively homogenous within different socio-demographic groups and geographical regions. We conclude that the ALMPs used in a small scale such as in Estonia are effective. 7. References 1. Dar, A., Tzannatos, Z. (1999) "Active Labour market programs: A Review of the Evidence from Evaluations", World bank. 2. Eamets, R., Philips, K., Annus, T. (2001), "Estonia: Candidate Country Monograph", ETF Working Paper. Turin. 3. Estonian Labour Force Surveys, various issues. 4. Fay, R.G. (1996), "Enhancing the Effectiveness of Active labor market Policies: Evidence from programme evaluations in OECD countries", Labor Market and Social Policy Occasional Papers No 18. Paris. 5. Heckman, J.J., LaLonde, R.J., Smith, J.A. (1998), "The Economics and Econometrics of Active Labour Market Programmes", in Ashenfelter, O. and Card, D. (eds.), The Handbook of Labour Economics, Volume III. 6. Kluve, J., Lehmann, H., Schmidt, C.M. (1998). "Active Labour Market Policies in Poland: Human Capital Enhancement, Stigmatization or Benefit Churning?" CEPR Discussion Papers with number 2059. 7. Lubyova, M., van Ours, J.C. (1998), "Effects of Active Labor Market Programs on the Transition Rate from Unemployment into Regular Jobs in the Slovak Republic", Tilburg Center for Economic Research Papers No. 98127. 8. Martin, J.P. (1998), "What works among active labor market policies: evidence from OECD countries’ experiences", Labor Market and Social Policy Occasional Papers No 35. Paris. 9. Sianesi, B. (2001) "An evaluation of the active labour market programmes in Sweden", IFAU Working Paper 2001:5

20