Education, training and youth employment outcomes. Evidence from Italy

Education, training and youth employment outcomes. Evidence from Italy Eliana Baici and Giorgia Casalone * 1. Introduction Human1 capital investment ...
1 downloads 0 Views 116KB Size
Education, training and youth employment outcomes. Evidence from Italy Eliana Baici and Giorgia Casalone *

1. Introduction Human1 capital investment is a key public policy for its impact on individual and collective well-being, economic growth and social mobility in presence of remarkable market failures. In particular, education policies should involve young people providing them with skills and competences useful for their entry and staying in the labor market. Training represents, together with formal education, a primary source of human capital and, thereby, an important area of public intervention. As declared by the Council of the European Union (2001) “Education and training are a structural means by which society can help its citizens to have equitable access to prosperity, democratic decision-making and individual socio-cultural development” and human capital improvement and accessibility represent key issues in the European Union policy agenda. Given the relevance of this matter, it is crucial to analyze and evaluate with rigorous methodologies the efficacy of these policies focusing on the characteristics of people involved and on their outcomes in terms of skills acquisition and employment. In Italy each year rather one million of people, that is 4% of the labour force, is involved in training activities (Isfol 2005). The higher quota of training is provided by Training Agencies (36%), followed by associations and non profit organizations (16%). Local authorities, such as Regions and Provinces and Municipalities organize 12% of training courses and the remaining part is provided by schools, Universities or entrepreneurial organizations. The State-sponsored training system in Italy is organized in three main intervention programmes. Firstly, the European Social Fund (FSE) co-finances training programmes organized by public and private training centres and firms. These programmes aim at improving base-skills in order to increase workers employment chances and are addressed to youths in search of first employment, long term unemployed, disadvantaged subjects and women. Secondly, the so-called Integrated Superior Training (FIS) is aimed to youth with a high-school degree to provide them with additional professional competences demanded by firms. Finally, there are programmes specifically devoted to employed workers (Lifelong Learning Programmes) that attend the courses autonomously or in consequence of firms’ proposals. These courses aim in particular at providing workers with new competencies allowing them to the production systems evolution. This paper aims at investigating the relationship between training, education and youth employment outcomes in Italy. In particular it aims at adding to the body of micro-econometric analyses of the impact of training on youth labour market outcomes that is rather small for Italy. The issues analysed in this paper are the following. Firstly, we investigate the decision to attend training programmes paying attention on the role played by previous formal education. The aim is to understanding the type of relationship existing between training and formal education. Secondly, we estimate the impact of training on youth employment condition. In particular we question training effectiveness, namely whether participating in training courses positively affects young people employment chances. Then, we investigate whether training participation reduces the need for informal channels, such as personal or family networks, to find a job. The hypothesis that we aim at testing is whether training strengthens workers’ power in the labor market providing them with some additional “formal” channels to contact potential employers. Finally, restricting our analysis to the youth employed, we aim at establishing if participating in training improves the *

Eliana Baici and Giorgia Casalone, Department of Economic Sciences and Quantitative Methods, Eastern Piedmont University, Italy Corresponding author. Address: Dipartimento di Scienze Economiche e Metodi Quantitativi, via Perrone 18, 28100 Novara, Italy. Email: [email protected] 1 We are grateful to the participants at the XIX Italian Public Economics Society Congress (SIEP 2007) for helpful comments and suggestions. Data availability from ISFOL is gratefully acknowledged. 317

human capital economic returns, measured in terms of wage satisfaction. Training is a source, together with formal education and learning by doing, of human capital and, consequently, of productivity. According to the Mincerian view about education returns, this should be recognized by the labor market through higher wages. Finally, we focus on matching between skills and competencies acquired within the whole educational process and the skills and competences required by the job and we ask if training improves this matching. The paper is organized as follows. Paragraph 2 presents a brief review of the empirical literature with a focus on the results concerning Italy. Paragraph 3 illustrates the data set and provides descriptive statistics. Paragraph 4 describes the empirical strategy used to identify the effect of training. Paragraph 5 presents the results of the analysis of the relationship between education and training and of the effect of training on employment. Finally paragraph 6 concludes.

2. Literature review The empirical literature on the relation between training and formal education is not particularly wide. According to the empirical analyses (see for example Brunello 2001; Ariga and Brunello 2006), formal education and training appear in general complement in human capital formation. Distinguishing between off-the-job and on-the-job training, Ariga and Brunello (2006) in particular find that only off-the-job training is a complement for formal education, whereas more educated workers are less probably involved in firmprovided training activities. The investigation of the effect of training on employment outcomes represents instead the focus of several studies (for instance Lindley 1981, Heckman and Robb 1985 and more recently Booth and Bryan 2005, Muehler et al. 2007) but not a particular attention has been paid to youth as far as we know. Two exceptions are represented by Higgins (1984) and Denny and Harmon (2000) that find a positive impact of training on youth employment outcomes once controlled for selection bias. As regard Italy, Rettore et al. (2002) put in evidence the low quality of training data that, is crucial for the robustness of policies evaluations (Heckman et al. 1999). The main methodological critical feature of the investigations on the Italian training system outcomes stems from the lack of complete information. For instance, Croce and Montanini (1997) in order to overcome the lack of systematic information about trained outcomes, contact them a long time after the end of the programme with a very low response rate implicating an inevitable attrition bias in the results. Centra et al. (2000) and Comi et al.(2002) create a “control group” to be compared with the “treated group” by using a different data set (the Labour Force Survey) with evident problems of information comparability between the two groups. The information availability problem is instead solved by Laudisa (2000) through the availability of information on the non admitted to a state-sponsored training course, but the analysis has to be limited only to individuals that are very close to the acceptance threshold. With a different perspective, Tattara and Valentini (2005) use information from the Italian Social Security System (INPS) archives of two Italian provinces to estimate the average employment gain for firms from training on the job contracts (Contratti di Formazione Lavoro-CFL). Exploiting information about firms that entered a CFL programme (treated) and firms that did not (non treated), they use a difference in difference propensity score matching and find a positive effect of these programmes on youth employment. Our paper aims at giving a new contribution to the analysis of the impact of training programmes in several ways. Firstly, since we have the same information for treated and non treated we can perfectly compare the two groups. Secondly, as we have information not only about the interviewed but also about their households, we can explicitly take account of the effect of family background that is crucial in human capital investment decisions (for Italy see for instance Checchi 2003). Finally, thanks to a wide set of information about jobs, we can analyze the effect of training and education not only on youth employment condition, but also on the quality of the achieved jobs. Since job quality is a crucial issue in the current policy debate, we aim at providing with our investigation some further elements of analysis.

3. Data and descriptive statistics The data used in this paper are drawn from the Isfol “Young People Education and Employment Survey” (YPEES) collecting information on 6,532 individuals aged 21 (3,456) and 31 years (2,896) at the moment of interview and that are representative of these two cohorts Italian population. The questionnaire provides a wide set of information: personal information, family characteristics, education choices and outcomes,

318

employment conditions, job features. A section of the questionnaire is then specifically devoted to training and contains information on the number and type of training programmes attended and on their quality. In order to investigate the impact of training on employment outcomes we consider a YPEES sub-sample defined as follows. Firstly we restrict the analysis to the respondents declaring that their prevailing employment condition is: employed, in redundancy fund (CIG) or in mobility, unemployed or in seek of first job. Since we are interested in the school-to-work transitions we then exclude all full time students, housewives and other “non actives” that generally include well-off persons, but also people that are training for a competitive examination. Secondly we consider only the younger cohort, that is the 21 years olds. The 31 years olds are excluded from the analysis for two main reasons. Firstly because, in order to analyse school-to-work transitions, we need to observe individuals few years after their labour market entry. Secondly, because we are not able, with the available information, at establishing when exactly the individuals participated in training programmes and this could strongly bias the results. As regard training programmes definition in our paper, we adopt the Isfol description. 2 Isfol defines training programmes as “programmes aiming at helping labor market entry, at improving the individual position within the labor market and at replacing in the labor market”. Isfol distinguishes then among three types of trainings: employer-provided, market-provided and state-sponsored courses. As the main purpose of this paper is to investigate the effect of training on employment outcomes, we limit our analysis to training programmes not provided by the employer. The exclusion of these courses, which are obviously attended after having obtained a job, arises for two main reasons. Firstly, when we analyse the probability of obtaining a job or the channels exploited to get it, it is meaningless to include, amongst the characteristics affecting these outcomes, the participation in a training programme that temporally follows the hiring. As regards the investigation of the jobs’ qualitative features (wage satisfaction, skills adequacy) introducing information about training provided by the employer obviously entails a potential endogeneity bias as well.

The final sample is then composed by 2,245 individuals. Table 1 reports some descriptive statistics about individuals’ characteristics and the analysed outcomes. The first column reports the statistics for the control group, that is those individuals that did not participate in any training. This group represents 71% of the whole sample. The second column reports the statistics for the participant in a general training programme both state-sponsored or market-provided. They represent 28.15% of the whole sample. We then split the participants in training programmes in two groups according to the type of training attended. The third and fourth columns report respectively the descriptive statistics for the individuals that attended state sponsored and market provided courses. They respectively represent 16.5 and 11.6% of the whole sample. Treated and controls do not substantially differ regarding personal information (gender, area of residence and dimension of the city of residence). We only observe that the percentage of individuals that attended a market training programmes is a little lower in Southern regions and in small cities, whereas it is a little higher in North-Western regions and in big cities. The groups differ instead as regards the education level. Compared to controls treated are characterized by a lower percentage of low educated people (compulsory school) and by a higher percentage of individual with a 3-years vocational school. The fraction of individuals with an upper secondary school is the same among the first two groups. When we look at the two treated sub-groups however we observe that they strongly differ since state-sponsored trained are in average less educated. Besides, when we consider individuals holding a high school degree we observe that treated have in general more technical or professional backgrounds. But again when we consider the two sub-samples we note that they strongly differ since 67% of the state-sponsored trained has a technical or professional degree, while this percentage is only 54% for the market-provided trained for which general tracks have a higher weight. As regard the reported final marks at lower secondary school the performances of the controls are a little better compared to the trained, but again the two treated sub-groups differ a little. The remarkable no answer rate on this question (especially as regards lower secondary school) is explicitly taken into account in the analysis. Finally, there are no remarkable differences between the two groups as regards the distribution of failures during school. We then report some indicators about the territorial context where the individuals live. The first indicator is the regional public spending for youth training programmes (per youth labour force unit) during 2004. This 2

See the YPEES questionnaire in the section devoted to training. 319

variable, expressed in euros, represents an indicator of the public resources devoted to youth training programmes. The second is the provincial youth unemployment rate again measured in 2004. This variable gives information about the local labour market conditions. As regard these context indicators we do not observe any strong difference among the analysed groups. The bottom part of table 1 provides then descriptive statistics about the analysed outcomes. Concerning employment conditions, for the sake of simplicity, we distinguish two main situations: 1) employed, also including those workers in redundancy funds (CIG) or in mobility3; 2) unemployed, including those seeking their first job. Table 1 DESCRIPTIVE STATISTICS (IN %) Controls

Treated

(I)

General training programme (II)

State-sponsored training program (III)

Market-provided training program (IV)

Observations Personal information

1613

632

371

261

Female

47.1

48.1

48.0

48.3

North West

23.1

25.5

24.3

27.2

North East

23.7

22.3

22.9

21.5

Centre

21.9

22.8

22.1

23.8

South

31.2

29.4

30.7

27.6

Small city (500,000 res.) Highest education degree

33.0

36.4

34.8

38.7

Compulsory school

24.6

18.5

23.5

11.5

Vocational school

10.1

16.0

19.1

11.5

Upper sec. school

65.3

65.5

57.4

77.0

High school tracks (over 100 high school graduated) General tracks (licei)

37

31

25

38

Technical/Commercial tracks

38

43

42

42.5

Professional tracks

13

20

25

11.5

11.3

6.7

8

3.8

Sufficiente (Pass)

19.3

17.9

20.2

14.6

Buono (Good)

24.2

22.6

19.7

26.8

Distinto (Very good)

8.9

5.4

4.3

6.9

Others tracks Lower secondary school final mark

Ottimo (Excellent)

8.3

7.3

5.7

9.6

No answer

39.3

47.8

50.1

42.1

21.8

18.8

17.8

20.3

436.60

425.19

414.29

440.69

24.63

24.25

24.10

24.46

Use of informal channels

60.1 45.0

64.7 35.9

67.9 32.5

60.2 41.4

Wage satisfaction (much or enough)

68.3

66.7

61.9

74.5

Skills adequacy (much or enough)

81.1

84.1

85.7

81.5

Fails One or more Context Regional public expenditures youth training programmes(€) Provincial youth unempl. rate

for

Outcomes Employed

Data: Istituto per lo Sviluppo della Formazione Professionale dei Lavoratori (Isfol), “Young People Education and Employment Survey” (YPEES) 3

None of the trainees is in CIG or in mobility, while 8 controls (1.21%) are in this condition. 320

From the reported data it could be argued that treated have a slight higher (4.6%) probability of being employed. Concluding from this first evidence that training programmes have a positive impact on employment chances would clearly be a mistake, since we can not exclude the presence of selection bias: trained, perhaps, could have been more “attractive” workers independently from their participation in training. The matching estimator allows detecting the net training effect once the determinants of the selection process are taken into account. Again splitting the treated sample in two groups appears meaningful since we observe that participants in state training have a higher probability of being employed than the other group (67.9 vs. 60.2%). The performance of the whole group of treated is the (weighted) average of these two figures and it does appear only a little better than the control group. With regards to the second examined issue, we focus on the channels used to find the jobs, distinguishing between “formal” and “informal”. Formal channels include ads in newspapers or on the internet, CVs sent to employers, competitive examinations, services offered by training centres, public or private jobcentres, stages or other work experiences. Informal channels are instead represented by every kind of “signalling” to the potential employer from relatives, friends or other people. Table 1 shows that controls use these informal channels more frequently than treated (45% vs. 35.9%) and that again there are significant differences between the two treated sub-groups. Finally, as regards job characteristics, the questionnaire contains a large amount of information on the (perceived) job quality. For our purpose, we focus on two features: wage satisfaction and adequacy of the 4 acquired skills with respect to the job actually done . Table 1 reports that whole treated group and the control group do not significantly differ according to these indicators and that they are in average rather satisfied as regard these two job features. However, we observe that the group of market trained is in average more satisfied than the group of state trained about wage (74.5 vs. 61.9%) and that, conversely, this latter is more satisfied than the former about skills adequacy (85.7 vs. 81.5%).

4. Empirical strategy The main problem to cope with in policy evaluation is to distinguish their effect from the effect of observable (and unobservable) individual or context characteristics. Since it is not possible to observe simultaneously the outcome of a person involved in a programme and the outcome of the same person if not involved, a typical selection bias problem arises because people participating in a programme could differ from people who do not participate. As observed by Heckman et al., “the fundamental aspect of the programme evaluation problem is that one cannot simultaneously observe the same person in a programme and out of it” (Heckman, Smith and Clements 1997), hence a different strategy has to be implemented. To deal with cross-sectional data in presence of selection bias we adopt in our analysis a non parametric methodology represented by the propensity score matching estimator (Rosenbaum and Rubin 1983; Dehejia and Wahba 1998a, 1998b; Dehejia 2005). This estimator takes account of the selection bias by assuming that it depends on observable characteristics. Given this assumption, selection bias can be overcame by comparing (matching) individuals who are as similar as possible in these characteristics. This methodology, known as “covariates matching”, has the advantage of being very accurate since it takes account of every single relevant individual characteristic. The disadvantage is that in presence of a high number of characteristics to be compared, it is difficult to implement it from a computational point of view. The propensity score matching estimator allows overcoming this difficulty by reducing the multidimensionality problem through the estimation of the conditional (to observable characteristics) probabilities to take the treatment. Two conditions have to be verified so as to eliminate selection bias with this methodology: 1. Controlling for X treatment must be independent from the outcomes, namely selection on treatment only 5 depends on the observable characteristics (Conditional Independence Assumption – CIA);

4

These indicators are constructed on the basis to the answers to this question: “Are you satisfied about the following features of your current job?” where the examined features are Wage and Adequacy of the acquired skills to the current job. The possible answers are four: Yes, much; Yes, enough; Not much; Not at all. 5 The CIA condition states that selection depends on observable characteristics and that all variables that can influence both the treatment and the outcome are observed. 321

2. The probability of participating in treatment, conditional on X, cannot be zero or one (common support condition6). While the common support condition can be easily verified, the CIA condition is more difficult to check. However Caliendo and Kopeinig (2005) states that the probability that the CIA condition verification strongly depends on the choice of the appropriate variables to be included in the propensity score estimation. The choice of the relevant variables should be made according to the following assumptions. Firstly, the variables have to be considered relevant from an a priori judgement of the researcher founded on his knowledge of the issue to be analysed and on previous empirical and theoretical literature results. Secondly, the chosen regressors must be unaffected by participation decision, that is they must be fixed (such as gender) or measured before participation decision (education levels). Thirdly they have to influence simultaneously the participation decision and the outcomes. Fourthly, regressors for treated and controls should stem from the same source, assuring a perfect comparability of information. Finally the adopted specifications should satisfy the so called “balancing hypothesis”. This hypothesis states that the distribution of X in (1) conditional to the balancing function (the propensity score in our analysis) is independent from the treatment and it assures that if CIA is verified for X, then it is also verified for Pr(X) (Rosenbaum and Rubin 1983). The weakness of the matching estimator rests on the non verifiability of the CIA. Once propensity score is estimated, the Average Treatment effect on Treated (ATT) is calculated as follows: ATT = EPr|T=1 {E[ (O1i | Ti =1, Pr(Xi)]- E[ (O0i | Ti =0, Pr(Xi)]}

(1)

where Oji is the outcome of individual i (for instance, employed or unemployed), j is 1 if individual i is treated and 0 otherwise, Ti is the treatment dummy that assume the value 1 if individual i takes the treatment, 0 otherwise and Pr(X) is the propensity score of the treatment, that is the probability of getting the treatment conditioning on a vector X of observable variables. The average treatment effect on treated, i.e. the average difference between the outcomes of the treated and the outcomes of the same individuals if not treated, is then computed according to three estimators7. The first one is the best matching estimator where each treated is compared with the most similar control in terms of propensity score. The advantage of this estimator is that it allows matching each trained with the control that is as similar as possible. The main disadvantage is that a lot of information is not exploited (all the controls that are not the “best”), but it could also occur that the “best” control is distant from the treated because no restriction is imposed on the distance between the two propensity scores. With the aim of exploiting the whole available information, we also present the results of a kernel matching estimator where each treated is compared with an observation that is a weighted average of all controls. Finally, to check the robustness of our estimates we also report the results of the stratification estimator which compares the outcome of each treated with the outcomes of the controls within the same propensity score block, that is with the controls that have the same (in average) observed characteristics.

5. Results 5.1. Training and formal education Table 2 reports in the first column the results of a probit estimate of the participation both in a general training program. In the second and third column we report the result of a multinomial probit of the participation in a state-sponsored or in a market-provided training. In both cases the dependent variable assumes the value zero when an individual did not participate in training. The reported results are the marginal effects. Holding whatever degree higher than compulsory education significantly increases training participation. When we consider specific high schools tracks we observe that individuals that got a general degree (liceo) 6

The common support condition ensures that any combination of characteristics observed in the treatment group can also be observed in the control group that it is sufficient to ensure the existence of potential matches in the control group (Caliendo 2005). 7 On this see Becker and Ichino (2002). The reported results are obtained by using the “pscore” ado file developed by Becker and Ichino with Stata. 322

have a higher probability to be involved in a market-provided course and a lower probability of entering a state-sponsored training. The same occurs for individuals holding technical degrees (Istituti tecnici commerciali/Istituti per geometri), but in this case this does not affect state-sponsored training attendance. Also the final mark obtained at the end of lower secondary school affect the probability of training participation: individuals with higher final marks have a lower probability of entering in state-sponsored training programs. Table 2 THE PARTICIPATION DECISION ESTIMATES (MARGINAL EFFECTS)

Female Big city (> 500.000 residents) Vocational school (3 - years) General tracks (licei) Techical tracks Professional tracks Other tracks Low sec. final mark 2 (Buono) Low sec. inal mark 3 (Distinto-Ottimo) Failure Regional spending for youth training (ln) Youth provincial unempl. rate (ln) Observations

General training program (I) Univariate probit .0197 ( .019) .033 ( .020) .201*** ( .040) .0463 ( .031) .115*** ( .03) .186*** ( .042) -.033 ( .043) -.024*** ( .022) -.080*** ( .026) -.091 ( .022) -.0049 (.0122) -.0027 (.014) 2245

State-sponsored training Market-provided training programme program (II) (III) Multinomial probit .020 -.002 ( .016) (.013) .0179 .0153 ( .016) ( .014) .106*** .098*** ( .035) ( .037) -.059*** .122*** ( .021) ( .0299) .0034 .126*** ( .023) ( .028) .074** .122*** (.035) ( .040) -.0418 .0214 ( .031) ( .038) -.045*** .024 ( .017) ( .017) -.070*** -.008 ( .019) ( .018) -.062*** -.028* ( .017) ( .015) -.019** .015* ( .009) ( .009) -.001 -.0009 (.0120) (.0105) 2245 2245

Standard errors in parentheses. *** p