The Role of Education vis-à-vis Job Experience in Explaining the Transitions to Employment in the Spanish Youth Labour Market by Cristina Fernández* DOCUMENTO DE TRABAJO 2003-06

April 2003

Universidad Pompeu Fabra y FEDEA.

Cristina Fernández¤ Universitat Pompeu Fabra and Fedea

Abstract

It is well known that young cohorts experience higher unemployment rates than their adult counterparts. However, it is less well known that more educated young cohorts may face higher unemployment rates than less educated ones. This seems to be the evidence in some OECD countries such as Spain and Italy. We use data on the Spanish labor market and estimate a duration model for young unemployed people. University graduates’ lack ofjob experience may explain this counterfact.

1 Introduction

When we look at unemployment rates by educational attainment and age across di¤erent OECD countries -Table 1-, two stylized facts come to light.

First, irrespectively of the level of education attained, young males and females always face higher unemployment rates than their adult counterparts. This evidence has led many researchers to study the determinants of the di¢culties in the transition of the young population into the labour-market (OECD 2000, García-Montalvo and Peiró 1999, Gangl 2000, García-Montalvo, 2000 and Martin and Velarde, 2000).

¤This paper formspart ofmy Ph.D. thesisto be submitted at the UPF. I would like to thank especially Juan José Dolado and Gilles Saint-Paul for their helpful guidance always. I would also like to thank Maia Güell for her useful comments and Adriana Kugler for her suggestions at the early stages of the paper. I also acknowledge …nancial support form the Spanish Ministry of Science and Technology (FP2001- 0708). Finally, for their useful comments, I would like to thank the participants at the Microeconometrics Workshop at Universidad Carlos III, at the seminar at Fedea and at the session on “Fertility and Education” at the 2002 Simposio de Análisis Económico (Salamanca, Spain). The usual disclaimer applies.

Secondly, unemployment rates tend to decrease as the level of education attained increases. This is, education seems to provide an insurance against unemployment. However, this fact is not a common feature across countries for the younger cohorts. While in most OECD countries young males and females face lower unemployment rates as their level of education increases, there are also some countries such as Greece, Italy, Luxembourg and Spain where unemployment rates increase with the level of education. This means that better educated young males and females face higher unemployment rates.

In order to understand this counterfact this paper focuses on the analysis of the Spanish case, since it is the one where the increasing trend of unemployment rates with education is clearest¹.

Table 2 provides some relevant information to understand the unemployment rates of the Spanish young population within the age cohorts. As in Table 1, we observe that, unemployment rates,within the same age cohort, increase as the level of education increases, specially among men (see shadowed cells Table 2). Moreover, we observe that this feature has been a characteristic ofthe Sapnish youth labour market for almost three decades. However, in this comparison, individuals are not homogeneous in one relevant variable: experience. When comparing individuals of the same age cohort but di¤erent levels of education we are mixing young males that have …nished school at di¤erent ages and therefore have di¤erent levels of job experience. In fact, we would expect those with higher levels of education to be the ones with lower job-market experience.

Next, if we compare individuals with di¤erent levels of education at the time they are supposed to enter the labour market (striped cells, Table 2), we observe that, as education increases, the unemployment ratedecreases. That is, when we compare individuals with similar labour experience across educational groups, education plays its expected role of insuring the individuals against unemployment.

In view of those facts, this paper focuses on the labour market fortunes of young Spanish workers and analyses whether education helps individuals to …nd a job faster. However, as noted above, to identify properly the e¤ect ofeducation on the probability of leaving unemployment it is crucial to control for labour experience.

To do so, we …rst focus on a sample of young workers who become unemployed after a period of employment and construct two proxies of experience: (i) potential job experience, and (ii) tenure in the previous job. Despite using these proxies, neither primary nor secondary education seem to be penalized with regard to university education. Moreover, those individuals with a university short degree seem to face higher probability of leaving unemployment than those with a university long degree.

¹Greece is a special case since the increasing path of unemployment rates with education is present not only within the younger cohorts but also within the adult cohorts.

Results for experience are more cumbersome. On the one hand, potential experience turns out not to have a signi…cant e¤ect on the exit rate. On the other hand, moderate tenures seem to signi…cantly increase the chances of …nding a job relative to those that had a too short or a too long tenure.

Given the di¢culties in measuring job experience, we try to ascertain the e¤ect of education by using a sample of young workers entering unemployment after …nishing/dropping out of school or after …nishing their military duties², and thus have no job experience. Namely, we intend to capture workers when they are entering the labour market for the …rst time by using a sample of young males and females that di¤er in their education attainment but are homegenous in their potential labour experience. Under this set up, young people with lower levels of education turn out to be signi…cantly penalized in their chances of …nding a job. However, university education does not seem to give any advantage over vocational education, at least during the early steps into the labour market and despite the fact that working experience is set to be null. We understand this last result as evidence of university longdegrees not being as much labour market oriented as university short degrees.

The rest of the paper is structured as follows. Section 2 surveys the studies made so far showing the di¢culties in disentangling the e¤ect of education. Section 3, describes our data sets. Section 4 explains the econometric model we use thorughout the paper while Section 5 presents the results. Finally, section 6 concludes.

2 Review of the literature

Returns to education have usually been studied in terms ofwages. However, in countries where unemployment rates are high, the decision to invest in education is not only regarded as a way to have access to higher wages but as a way of insuring against unemployment.

Determinants of the probability of getting out of unemployment have been widely studied since the seminal studies of Lancaster (1979) and Nickell (1979a). The focus of most of the studies has been centered on the role of unemployement bene…ts and unemployment duration in increasing or decreasing the unemployed’s chances of…nding a job (Nickell, 1979b, Alba and Freemna,1990, García-Pérez, 1997, Bover et al., 1996, Narendranathan and Stewart, 1993b, and garcía-Serrano and Jenkins, 2000). However, less attention has been devoted to the role of education.

²Over the period of study, the military service has been compulsory for young males.

In the context ofthe Spanish labour market there are two papers by Bover et al. (1996) and Calero and Madrigal (2002), that address the role ofeducation attainment on the exit rate from unemployment among the adult population. Both of them obtain similar results despite using di¤erent data.

On the one hand, in the context ofSpanish unemployed males, Bover et al. (1996) obtain³ that “holding a university degree increases the hazard only at the beginning of a spell. After the third month, the presence of negative duration dependence reduces the hazards of college graduates below those of less educated workers, which presumably re‡ects the former’s higher reservation wages”. However, this result could also be interpreted as college skills depreciating faster than primary skills, that is, in terms of the o¤er arriving rate rather than the acceptance probability.

On the other hand, Calero and Madrigal (2002) obtain a similar conclusion with the ECHP⁴. Higher education attainment signi…cantly increases the probability ofgetting out ofunemployment but this e¤ect is overturned by the negative e¤ect that receiving unemployment bene…ts exerts on the more educated unemployed. Again, they interpret this negative overall result on education as evidence ofa higher reservation wage among higher educated workers. However, in this case, the di¤erential e¤ect of unemployment bene…ts across educational groups may not be related with education itself but with the generosity of unemployment bene…ts that depends on previous employment status.

When we focus the attention on studies restricted to the young population, results on the role of education show more divergence.

Alba (1998) uses the Spanish labour Force Survey over the period 1987-1996 to select a sample of young unemployed males and females who report previous working experience.With regard to education, he …nds that young unemployed males with a university degree seem to face lower chances of getting a job than less educated males. However, in the case offemales, education seems to be a factor that neither increases nor decreases their chances of reemployment.

³Bover et al (1996) uses the Sapinsh Labor Force Survey for the period 1987:II-1994:III

⁴European Community Household Panel.

Albert et al. (2000) focus on an earlier stage of young’s working life: their transition from school to work. They use the Spanish labour Force Survey over the period 1992-1999 to select a sample of young males and females that are in school at the …rst interview they are observed.

Their interest is on identifying the role of di¤erent variables on the transition towards employment, unemployment or inactivity among those that get out of school before the end of the interviewing period. They …nd that young educated females face higher probabilities of being employed rather than inactive once they get out of school. However, education does not seem to play any role in determining the employment status among male school-leavers.

Along the same line, Lassibille and Navarro (1999) use the Encuesta Sociodemográ…ca carried out in 1991, to analyze the labour market entrance of recent shool leavers without any previous working experience. They …nd that those with more labour oriented studies -vocational secondary education or universtiy short degree- are the ones that get a job faster. Those with general secondary education are the ones that take longer to …nd a job. And surprisingly, those with university long degree get out of unemployment at the same pace of those with compulsory studies.

Finally, Davia and Smith (2001) report results obtained from the ECHP on the non-parametric probability of getting out of unemployment ofyoung males and females that were unemployed in January 1993. They show that there does not seem to be any di¤erence across educational groups⁵.

In summary, a comparative review of the research carried out so far on this issue does not give a clear answer on whether education helps young unemployed to …nd a job faster. In fact, if anything, we could conclude that education, specially for young males, does not in‡uence at all the exit rate from unemployment.

3 The data

The data we use comes from the Spanish labour Force Survey (Encuesta de Población Activa). This survey is carried out every quarter on a sample of around 60.000 households. Information on employment and educational background is collected for all household members above 15 years ofage. Every quarter one sixth ofthe households is renewed. This enables us to follow individuals for up to six consecutive quarters.

⁵The ECHP allows researchers to distinguish only three educational groups, each of which gathers studies of very di¤erent nature. For example, group 1, the higher level of studies, gathers long and short university degrees and secondary vocational studies. Thisdiverstiy ofstudieswithin the same group isa disadvantage when trying to analyze the e¤ect of education in detail.

Our focus is on the sample ofyoung males and females aged 16 to 30 years-old when …rst interviewed that entered the survey between 1992:I and 2001:II. We restrict the analysis to single males and females that live with their parents⁶. This restriction, together with the time period under study, allows us to assume that males and females have the same attachment to the labour market⁷. Moreover, considering only young males and females that are not married and live with their parents will allow us to de…ne as unemployed not only those registered so by the LFS⁸, but also those classi…ed as inactive (Bover et al., 1996), as long as they do not attend regular studies.

Two di¤erent samples have been designed labelled respectively as: (i) the ”attachment” sample, and (ii) the ”transition” sample. On the one hand, the ”attachment” sample includes young unemployed individuals that have a strong attachment to the job market. Therefore, we exclude individuals that have never been employed and that attend formal education or the military service at any stage of the observed period. However, we do not exclude individuals that are following non-regular courses as long as they do not classify themselves as students.

We also restrict the ”attachment” sample to those individuals we observe entering unemployment⁹. This condition is used because we do not observe key characteristics of the young individuals that report to be unemployed in the …rst interview. Among those missing characteristics there are the level of education at the start of the unemployment spell, whether they have attended regular education or the military service during their unemployment spell, and whether they have received bene…ts at any point of the spell.

On the other hand, the ”transition” sample is designed to isolate the e¤ect of education on the probability of …nding a job from the e¤ect of previous labour experience. Thus, we include only young unemployed males and females that have never been employed before and that have just …nish/drop school¹⁰. Moreover, we require that once we identify their transition out of school we do not observe them going back to the school system nor classifying themselves as students again. In this manner, we intend to capture individuals that are undergoing their initial transition from school to the labour market (Albert et al., 2000).

⁶This assumption does not limit the representativity of our analysis. The data shows that 75% of the Spanish young population -between 16 and 30 years old- is single and lives with his/her parents. More than 70% of young workers who become unemployed and almost a 90% of young males and females that get out of school are also in that situation.

⁷Having the same attatchment to the labour market implies that we do not have to worry about the e¤ects of being married or having children on female labour supply. ⁸Labour Force Survey

⁹We exclude young males and females that report they quit their jobs.

^{1 0}The requirement of going through a period of unemployment intends to exclude young males and females that drop school because they have found a job.

One of the di¤erences from other available papers in this literature, is that we will consider not only transitions from school to work but also from the military service to the labour market. During the sample period under study, themilitary service has been compulsory for themale population implying that the majority of the male population did not start searching for a job until completing their military duties. Hence, …nishing the military service seems to be crucial for the labour search commitment of males as it is …nishing formal education for females and ”objetores de conciencia”¹¹.

Table 1 displays the summary statistics of the explanatory variables in the “attachment” and “transition” samples. Appendix C and D, in turn, present a more detailed description of the construction of both samples as well as of the de…nition of each variable. One of the advatnages of using the LFS is that it allows us to specify our variable of interest, education, in detail distinguishing seven educational dummies. This precision is not allowed by other surveys, e.g., the ECHP (see footnote 5).

Before moving ahead, however, it is important to go further discuss the decision of limiting the analysis to young single males and females that live with their parents throughout the survey period. We have previously noted that within the Spanish context this limitation does not imply a lack of representativeness of the analysis. However, some doubts may still arise on whether our results remain biased due to this restriction. In fact, this will be the case if we can not follow with the survey those unemployed workers that have to move to another region and therefore leave their family homes when they …nd a job. Specially, we will be biasing the e¤ect of our variable ofstudy, ‘education’, if there is a group of individuals with a common characteristic, e.g. those with a university degree, that showa higher propensity to migrateand therefore be dropped from the sample. Fortunately, this is not the case. The survey registersas still living with the family unit all theyoung malesand females that have moved to a di¤erent region, either to study or to work. The only requirement is that they eventually plan to return to their family homes in the near future¹². As a piece of evidence, we …nd that in the second quarter of 1999, almost a 4% of the young single working males and females that are registered as living with their parents, turn out to be working in a di¤erent region from the one they are registered living. Some of these cases could in fact refer to daily commuting, but a considerable percentage -at least a 50%- necessarily refers, because of the distance, to cases where the young worker lives in the region where he works and not in the one where data on the family unit is collected.

^{1 1}“Obejetores de conciencia” is the term that refers to the young males that choose to perform social services as a substitute of the military duties. This option characterizes because of its ‡exibility for combining it with labour market activities. It has also been used as a way to postpone military duties in the hope of avoiding them.

4 The model

Asjob search theory states, the conditional probability ofleaving unemployment in any period can be viewed as the product of the probability of receiving an o¤er and the probability that this o¤er is acceptable, i.e., exceeds the reservation wage (Mortensen, 1985).

Prob of exit = prob receiving an o¤er prob (o¤er > reservation wage)

In this paper, however, we do not impose the restrictions implied by thestructural model ofjob search. Instead a reduced-form model is used. Themain reason to opt for a reduced-form model is that our dataset does not contain relevant information, e.g. reservation wages, that is required to estimate the structural model. Hence, we will specify a hazard rate that relates to di¤erent regressors and to the elapsed unemployemnt duration.

Nickell (1979) points out that we can classify the di¤erent regressors that in‡uence theexit rate from unemployment into four categories: personal characteristics, local labour demand characteristics, family composition and income variables. Income variables can be thought of as in‡uencing reservation wages while variables included in the …rst three categories can be thought of as determining both the probability of getting an o¤er and the probability of accepting it.

The main drawback ofthis reduced-form approach is that wewill only observe the overall e¤ect of the variables we control for, e.g., education , on the exit probability. This is, we will not be able to disentangle the e¤ect of each variable on the probability of receiving an o¤er from the e¤ect on the reservation wage. We need to keep this in mind when interpreting the results.

^{1 2}The LFS considers as subject of interview within the family unit “ [...] all the students and workers that have temporarily move to other region or country and are planning to come back to their homes when the reason for moving gets to an end, even though they have been absent for three or more months. In the case we do not know whether they eventually plan to come back, they will not be taken into consideration if they have been absent for more than a year.” (EPA, 2000)

4.1 Speci…cation of the hazard without unobserved heterogenity.

As discribed above, the LFS allows us to observe individuals quarterly. Then, since we only observe the employment status of the individual at the extremes of the interval, we will be missing employment spells shorter than one quarter. In fact, the data reveals that the proportion of …xed term contracts signed with a duration shorter than 3 months amounts up to 20%.

Given this structure of the data set, we need to follow a continuous time hazard funtion approach where durations are interval censored (Narendranathan and Stewart, 1993 and Jenkins, 1995).

Suppose that the transition out of unemployment is a continuous process with proportional hazard

\[\mu_ {i} (t; X) = \mu_ {0} (t) \exp \left(^ {- 0} x _ {i}\right)\tag{1}\]

where is the vector of time-invariant explanatory variables¹³, µ₀(t) is the baseline hazard function which is common to all persons -does not depend on and is a vector of unknown coe¢cients. Then the continuous survivor function at time t can be written as:

\[S \left(t _ {i}; X\right) = \exp_ {i} \int_ {0} ^ {Z _ {t _ {i}}} \mu_ {i} (u; X) d u = \exp_ {i} \int_ {0} ^ {Z _ {t _ {i}}} \mu_ {0} (u) d u \exp \left(^ {- 0} x _ {i}\right),\tag{2}\]

Since the time axis is partitioned into 15 monthly intervals indexed by

\[[ 0 = a _ {0}; a _ {1}); [ a _ {1}; a _ {2}); [ a _ {2}; a _ {3}); \dots ; [ a _ {1 4}; a _ {1 5})\]

the survivor function for individual i at the start of the t-th interval and at the end of the t-th interval are, respectively

\[S \left(a _ {t _ {i i} 1}\right) = \operatorname * {P r} \left(T _ {i}, a _ {t _ {i} 1}\right) = 1; F \left(a _ {t _ {i i} 1}\right)\]

and

\[S \left(a _ {t _ {i}}\right) = \operatorname * {P r} \left(T _ {i}, a _ {t}\right) = 1; F \left(a _ {t _ {i}}\right)\]

where is a discrete random variable representing the time at which the end of the spell occurs and is the discrete time cumulative distribution function. The probability of exit within the t-th interval is:

\[\operatorname * {P r} \left[ T _ {i} ^ {2} \left[ a _ {t _ {i}} 1; a _ {t}\right) \right] = F \left(a _ {t _ {i}}\right) i F \left(a _ {t _ {i} i} 1\right) = S \left(a _ {t _ {i} i} 1\right) i S \left(a _ {t _ {i}}\right)\]

^{1 3}X includes the personal, local labor demand, family composition and income variables mentioned above.

and the dicrete time hazard of exit in the t-th interval, which is the probability ofexit in the t-th interval given that the individual has been unemployed up to t-1 is:

\[h _ {i} \left(a _ {t _ {i}}\right) = \operatorname * {P r} \left[ T _ {i} ^ {2} \left[ a _ {t _ {i} 1}; a _ {t}\right) = T _ {i}, a _ {t _ {i} 1} \right] = \frac {\operatorname* {P r} \left[ T _ {i} ^ {2} \left[ a _ {t _ {i} 1} ; a _ {t}\right) \right]}{\operatorname* {P r} \left(T _ {i , s} a _ {t _ {i} 1}\right)} = 1 i \frac {S \left(a _ {t _ {i}}\right)}{S \left(a _ {t _ {i} 1}\right)}\]

Substituting the expression obtained in (2) leads

\[\begin{array}{r l} h _ {i} \left(a _ {t _ {i}}\right) = 1 _ {i} \frac {S \left(a _ {t _ {i}} ; X\right)}{S \left(a _ {t _ {i i}} 1 ; X\right)} & = 1 _ {i} \exp_ {i} \exp \left(^ {- 0} x _ {i}\right) ^ {Z _ {a _ {t _ {i}}} a _ {t _ {i i}}} \mu_ {0} (u; X) d u \\ & = 1 _ {i} \exp \left[ i \exp \left(^ {- 0} x _ {i} + ^ {\circ} t\right) \right] \end{array}\tag{3}\]

where is the log of the integrated baseline hazard µ over the interval . We will assume can di¤er within each duration interval, and estimate the model semiparametrically.

To simplify notation and since monthly intervals are of unit length, we can denote the recorded duration interval for each individual as 2 where

The likelihood contribution of a spell of length can then be written as:

\[\begin{array}{r l} L _ {i} & = c _ {i} \operatorname * {P r} \left(T _ {i} 2 \left[ t _ {i, 1}; t _ {i}\right)\right) + (1 _ {i} c _ {i}) \operatorname * {P r} \left(T _ {i, s} t _ {i}\right) \\ & = c _ {i} \log \frac {h _ {i} (t _ {i})}{1 _ {i} h _ {i} (t _ {i})} + \underset {s = 1} {\times} \log (1 _ {i} h _ {i} (s)) \end{array}\]

where is an indicator variable that takes value one when the observation is not censored. And the log likelihood for the whole sample is:

\[\log L = \underset {i = 1} {\times} c _ {i} \log^ {"} \frac {h _ {i} (t _ {i})}{1 _ {i} h _ {i} (t _ {i})} ^ {\#} + \underset {i = 1 s = 1} {\times} \log (1 _ {i} h _ {i} (s))\]

At this stage we can follow Allison (1982) and Jenkins (1995) and rewrite the model as a binary choice model. The main advantage ofthis transformation relies on the wider availability of statistical packages to estimate binary choice models.

In a duration model we are dealing with a sequence of time where we observe at each time unit whether the individual is employed or still remains unemployed. Hence ”the discrete duration model can be regarded as a sequence ofbinary choice equations de…ned on the survivor population at each duration” (Bover, 1996). Each individual can then be seen as contributing observations, one for each time unit his/her spell lasts. This amounts to form a sample of size observations.

Let be an indicator variable that takes value 1 if the observed duration of the spelll falls in the interval

\[y _ {i t} = 1 (T _ {i} 2 [ t _ {i} 1; t))\]

The log-likelihood of the sample for given that

\[L _ {t} = \underset {i = 1} {\overset {x ^ {n}} {\sum}} 1 \left(T _ {i} > t _ {i} - 1\right) \left[ c _ {i} y _ {i t} \log h _ {i} (t) + \left(1 _ {i} c _ {i} y _ {i t}\right) \log \left(1 _ {i} h _ {i} (t)\right) \right]\]

where n is the number ofindividuals/unemployment spells in the initial sample and t ranges from 1 to 15.

Summing up the for all observed durations, we obtain the loglikelihood we …nally estimate:

\[L (\mu) = \underset {t = 1} {\overset {5} {\times}} L _ {t} = \underset {i = 1} {\overset {8} {\times}}: c _ {i} y _ {i t} \log \frac {h _ {i T _ {i}}}{1 i h _ {i T _ {i}}} + \underset {t = 1} {\overset {T _ {i}} {\times}} \log (1 i h _ {i t})\]

where is the vector of parameters to estimate. Substituting the hazard rate by the experssion obtained in (3), yield that

4.2 Speci…cation of the hazard with unobserved heterogeneity.

We introduce a random variable " to capture unobserved heterogenity. Following Jenkins(1997), the discrete time hazard rate for each person in each duration interval can be written as:

\[h _ {i} \left(t _ {i}\right) = 1, \exp f _ {i}" _ {i} \exp \left(\overline {{{\mathbf {k}}}} _ {i} + ^ {\circ} _ {t}\right) g\]

where is a positive valued random variable with unit mean and nite variance. Moreover, we assume the unobserved heterogenity term is distributed independently of t and X and that it follows a gamma distribution function.

The log-likelihood (Jenkins, 1997) then becomes

\[\log L = \underset {i = 1} {\mathbf {X}} \log \left[ \left(1 _ {i} c _ {i}\right) A _ {i} + c _ {i} B _ {i} \right]\]

with

\[A _ {i} = 1 + \frac {3}{4} ^ {2} \underset {s = 1} {\overset {\times} {\operatorname{X}}} \exp \left(x _ {i} \mathbb {T} + \circ_ {s}\right) ^ {\# _ {i} \frac {1}{\frac {3}{4} 2}}\]

Yit

t

^{1 4}The sample Y_it is formed by all the individuals whose spell is longer than t_i.

\[B _ {i} = \begin{array}{c} \stackrel {{8}} {{< }} h \\ : \end{array} 1 + \stackrel {{3 / 4}} {{2}} P _ {s = 1} ^ {t _ {i i}} ^ {1} \exp (x \mathbb {I} + \circ_ {s}) ^ {i _ {i}} \stackrel {{1}} {{\frac {1}{3 / 4}}} ^ {2} i A _ {i} \quad \text {if t_{i} >1}\]

where and ° are the usual censoring indicator and the function describing duration dependance respectively, and is the variance of the gamma-distributed random variable.

5 Results

In this section we present the results obtained from the estimation of the econometric model described in Section 4 under di¤erent alternative speci…cations of the controls for job-market experience.

5.1 Attachment sample

First, we focus on the re-employment probability of young unemployed males and females with a strong attachment to the labour market. Table 4 displays the results.

When we do not control for labour experience (…rst column of Table 4), the results are those expected from Table 2: Higher educational attainments, with the exception of graduates with a university short degree, do not help at the time of exiting unemployment.

Next, in order to control for labour experience, two di¤erent proxies have been constructed: potential experience and job tenure. Potential experience is the current age minus the age associated to the level of education attained, while Job tenure measures just the length of the last employment spell. Results, which are displayed in the second and third columns of Table 4, show again that being a university graduate with a long degree does not enhance the chances of leaving out of unemployment. Moreover, they are penalized with regard to those college graduates with a short degree (Graph 1).

The e¤ects of the experience proxies turn out to be more di¢cult to interpret. On the one hand, we obtain that potential experience does not signi…cantly a¤ect the exit rate¹⁵. On the other hand, as regard job tenure, we obtained that having a moderate tenure increases the chances of exiting unemployment with respect to those that had a too-short or a too-long tenure (Graph 2).

We have also tried interactions of di¤erent variables with the dummy for gender to check whether there is a di¤erential behaviour between young males and females within a speci…c category. Only the interactions with the educational dummies have turned out to be signi…cant.Results, displayed on Table 5, show that educated males face di¤erent probabilities of re-employment when compared with their females counterpart.

^{1 5}We have also follow Soro-Bonmatí and Ianelli (2001) and study the probability of re-employment of young males and females that have no more than 5 years of portential working experience. This is, we restrict the sample to individuals that are homegeneous in their potential working experience. Results reported on Table 1 of Appendix E show that coe¢cients on education remain insigni…cant.

In fact, while it remains true for males that university long degrees do not seem to increase the exit rates out of unemployment, this result no longer holds for females with low levels of education: women with primary or compulsory education arepenalized with respect to university graduates in their chances of getting out ofunemployment. However, as occurs with males, there does not seem to be any di¤erence between levels of education above compulsory education, despite the di¤erences in the characteristics of the studies.

The reason for females’ pattern on education to be di¤erent from the one of males may be due to their di¤erent labour attachment intensities across educational levels. While males labour market attachment is given for granted, …rms may think that better educated females will have a higher compromise with the labour market than those with compulsory studies. Then, we should expect that relatively to males, better educated females face higher o¤er arriving rates than those with compulsory studies, and therefore, higher exit rates out of unemployment.

5.2 Transition sample

So far, our main …nding is that higher levels of education do not seem to gurantee a faster exit out of unemployment, specially among men.

In view of this puzzling result, it seems natural to conjecture that a failureto account properly for youth labour market experiencecould bias our results on education. However, the Spanish LFS does not provide a precise measure of young males and females working experience and it is likely that the two proxies we have used so far are plagued by measurement error.

In e¤ect, the …rst proxy of experience, potential experience, makes two strong assumptions. On the one hand, it assumes that all individuals …nish their studies on time, and on the other hand, it assumes they have been working all the time since they …nish school up to the current interview. The second variable, job tenure, as we noted, measures only the tenure on the previous contract. Given the wide spread use of temporary contracts among the young population (60% in the second quarter of 1999) we should expect again that it won’t be a good proxy of labour market experience.

Faced with this di¢culty, the transition sample provides us an alternative way to control for working experience: construct a sample where working experience is null. In this manner, the e¤ects of educational attainments on the probability of leaving unemployment will not be contaminated by the measurement errors associated to our proxies of labour experience.

Table 6 displays the estimates of the probability of …nding a job right after …nishing school for young males and females that have never been employed. In the case of males we also consider those that are observed …nishing their military duties The …rst column displays results without control for unobserved heterogeneity while the second column displays estimates where we do control for it. Results indicate that there is no evidence of biases due to unobserved characteristics.

Under this new set-up, it turns out that low educated males and females face lower chances ofgetting out ofunemployment than university graduates¹⁶ (graphs 3, 4 and 5). However, we observe again that there is a threshold level ofeducation that makes the di¤erence. We can distinguish two groups of studies: group A, composed of university graduates (short and long degree) and those with vocational studies, and group B, composed of males and females with compulsory and secondary studies. The evidence shows that young males and females that belong to group A face higher chances of …nding a job than those that belong to group B. However, within groups, e.g., when we compare vocational and university studies, we do not …nd di¤erences in the exit porbabilities.

Therefore, wecan say that, once we account for experience, education matters. This is, higher levels of education increase young males and females chances of getting out of unemployment. However, we do not …nd a signi…cant di¤erence on the exit rates among individuals with high levels of education but di¤erent patterns of specialization: vocational versus university education.

The …rst explanation that comes to the mind of a researcher to explain this lack of signi…cant di¤erence between the e¤ect of university and vocational studies may rely on reservation wages. As noted above, in Section 4, one ofthe withdrawals ofthe reduced-form approach for modelling the exit rate from unemployment is theimposibility to disentangle the e¤ect of the variables of interest on the o¤er arrival rate and on the aceptance rate. And when dealing with variables such as education, the e¤ect on each of the rates may compensate since education in‡uences each rate in opposite directions: higher education attainments tend to increase the o¤er arrival rate but also tend to decrease the aceptance rate becuase it makes individuals more demonading (higher reservation wages).

However, when dealing with young unemployed workers that livewith their parents and are looking for their …rst job in an environment ofhigh unemployment rates, there may be reasons to think that reservation wages are determined by parents’ socioeconomic status and that the negative e¤ect of education on the aceptance rate is not so strong to outweight the positive e¤ect on the o¤er-arrival rate.

^{1 6}The interactions ofthe educational dummies with the dummy for sex do not turn out to be signi…cant now. Results are displayed on Table 2 of Appendix E.

Then, we think that the reason for the lack of signi…cant di¤erence between vocational and university studies in explainning exit rates may rely on the fact that we are not properly isolating one crucial characteristic of the Spanish educational system: vocational degrees and university short-degrees are more labour market oriented studies than university long-degrees. Then, even though university graduates stay longer at school, they lack the working experience that is necessary for a fast integration/transition into the labour market.

In fact, a piece of evidence supporting this explanation could be the increasing e¤ort of public universities to organize services for their students that provide summer internships at local …rms. Also, a key characteristicin the advertisement ofnewly private graduateschools and post-grad courses has been precisely their provision of such internships combined with scholarly activities.

5.3 Other results

In this section we discuss the way other characteristics, beside education attainment and working experience, a¤ect the exit rate from unemployment.

With regard to age, we obtain that younger cohorts face lower reemployment probabilities (Table 4), capturing the fact that they do not have as much working experience as their older counterparts. However, when we consider the sample of the young males and females that are getting out of school or the military service (Table 6), the younger they are, the higher their exit rate towards employment. In this case, lower ages, once we control for educational attainment, provide a signal of higher intelectual skills since it took them shorter time to …nish a given degree.

Economy-wide variables behave according to a priori beliefs (Tables 4 and 6). The better the evolution of the business cycle, the higher the exit probability. Both variables, gdp growth and regional unemployment rate, verify this presumption.

Finally, when we focuson the role parents play on the exit probability towards employment of their young “kids”, we …nd an interesting result. Estimations in Tables 4 and 6, this is, both attachment and transition sample, show that living in a family where either the father or the mother works, increases the chances of the young male or female of getting out of unemployment. We could expect that young unemployed that live in families where parents work are more choosy in accepting job o¤ers. However, these young males and females can also bene…t from their parents labour network (graphs 6 and 7). Results show that parents do not seem to increase the reservation wage of their children, but rather they provide their kids with contacts helping them to …nd a job¹⁷, or at least, the e¤ect of the latter role is bigger than the former.

Congregado and García (2002) obtain a result in the same direction although they focus their attention in the role of parents’ education (Bote,2002). Bote (2002) and Congregado and García (2002) argue that parents with higher education attainment are able to provide their kids with higher contacts in the labour market. However, the intuition that we obtain from our results is that it is not just a matter of education but a matter of parents participation in the labour market.

6 Conclusions

A cross country comparisson of unemployment rates across ages and educational attainments shows that the Spanish youth labour market presents a peculiar feature: higher levels of education do not seem to ensure young males and females from being unemployed. In fact, we observe that, within a given age group, the higher the level ofeducation, the higher the unemployment rate males and females face.

A closer look at the data shows that the explanation for this counterfact could come from the role of job-market experience. Given the duality between the student-life and the working-life that characterizes the Spanish labour market, we would expect that, within an age cohort, young males and females that attain higher levels of education will have acquired lower working experience, and vice-versa.

A duration model ofthe probability of getting out ofunemployment has been developed to isolate the e¤ect that education has from the e¤ect that experience has on the exit rate.

The results we have obtained suggest that once we control for experience, education helps young males and females to get out of unemployment. However, there seems to be a threshold level of education above which higher levels of education do not give higher returns in terms of employment. In fact, we could distinguish two groups: group A (university graduates with short or long degrees and males and females with vocational studies), and group B(young males and females with compulsory or secondary studies). The evidence shows that those that belong to group A face higher chances of getting out of unemployment than those who belong to group B. However, within group A, there do not seem to be di¤erences in the exit probabilities despite the di¤erences in the characteristics of the studies. The reason may be due to the fact that university long-degrees are not as much labour market oriented as the labour market demands.

^{1 7}Montalvo (2000) reports that 26.4% of Spanish university graduates found a job because of personal contacts (parents, relatives and friends) compare to an average of 13.46% in the rest of countries that participated in the CHEERS programme.

References

1. [1] Alba-Ramirez, A. and Freeman, R. (1990) “Job…nding and Wages when Longrun Unemployment is really Long: the case of Spain”, Working Paper no. 3409, NBER.
2. [2] Alba-Ramirez, A. (1998). “Re-employment Transitions of Young Workers in Spain” Investigaciones Económicas vol XXII(2), pp. 201- 224.
3. [3] Alba-Ramirez, A. (1999). “Explainning the Transitions out of Unemployment in Spain: the E¤ect of Unemployment Insurance”. Applied Economics, 31, pp 183-93.
4. [4] Albert, C. et al. (2000), “La transición de la escuela al mercado de trabajo en España: años noventa”. Papeles de Economía Española 86, pp 42-58.
5. [5] Allison, P.D. (1982). “Discrete-Time Methods for the Analysis of Event Histories”. Sociological Methodology. Samuel Leinhardt Ed. pp61-98.
6. [6] Becker, G.S (1975) Human Capital, a Theoretical and Empirical Analysis with Special Reference to Education. Columbia Unviersity Press.
7. [7] Bote, V. (2002) “Sorting, Job Contacts and Inequality”, Tesina CEMFI n^o0206.
8. [8] Bover, O. et al. (1996), “Unemployment Duration, Bene…t Duration, and the Business Cycle” Estudios Económicos n^o57. Banco de España.
9. [9] Congregado, E. and García, J.I. (2002) “El problema del desempleo en la economía andaluza (1990-2001): análisis de la transición desde la educación al mercado laboral”, centrA working paper 2002/17
10. [10] EPA. (2000), “Descripción de la encuesta, de…niciones e instrucciones para la cumplimentación del cuestionario”. INE
11. [11] Davia, M.A. and M. Smith (2001). “ Youth Labour Markets in Spaon and UK Di¤erent Systems: Di¤erent Outcomes?. IV Jornadas de Economía Laboral.
12. [12] Dolado, J.J. et al. (2000), “Explainning Youth Labor Market Problems in Spain: Crowding-out, Institutions, or Technology Shifts?”
13. [13] Gangl, M. (2000), “Education and Labor Market Entry across Eu-
14. rope: The Impact of Institutional Arrangements in Training Systems and Labour Markets” MZES Working Paper n^o25.
15. [14] García-Montalvo, J. and J.M Peiró (1999). El mercado laboral de los jóvenes: formación, transción y empleo. Bancaja.
16. [15] García-Montalvo, J. (2000) “Formación y empleo de los graduados de enseñanza superior en España y en Europa” Fundació Bancaixa.
17. [16] García-Pérez, J.I (1997) “La tasa de salida del empleo y el desempleo en España (1978-1993)”, Investigaciones Económicas, 21, 29- 53.
18. [17] García-Serrano, C. and S.P. Jenkins (2000) ”Re-employment probabilities for Sapnish men: what role does the unemployment bene…t system play?” ISER working papers 2000-17.
19. [18] Jenkins, S.P. (1995). “Easy Estimation Methods for Discrete-Time Duration Models” Oxford Bulletin of Economics and Statistics, 57- 1, pp129-39.
20. [19] Jenkins, S.P. (1997), “sbe17: Discrete Time Proportional Hazard Regression” in Stata Technical Bulletin, n.39, pp.22-32.
21. [20] Lassibille, G. and Navarro, L. (1999), “Youth Transition from School to Work in Spain” WP-08-99.
22. [21] Lancaster, T (1979) “Econometric Methods for the Duration of Unemployment” Econometrica 47.
23. [22] Martin Serrano, M. and O. Velarde Hermida (2000) Juventud en España. Injuve.
24. [23] Mortensen, D.T. (1985) “Job Search and Labour Market Analysis” in Handbook of Labour Economics. Volume 2. North Holland.
25. [24] Narendranathan, W. and M. Stewart (1993a), “Modelling the Probability of Leaving Unemployment: Competing Risk Models with Flexible Base-line Hazards”. applied Statistics 42, No 1, pp63-83.
26. [25] Narendranathan, W. and M. Stewart (1993b). “How does the bene…t e¤ect vary as unemployment spells lengthen?”, Journal of Applied Econometrics, 8, 361-381.
27. [26] Nickell, S. (1979b). “The E¤ect ofUnemployment and Related Bene…ts on the Duration of Unemployment”. The Economic Journal, 89, pp 34-49. March.
28. [27] Nickell, S. (1979a). “Estimating the Probability of Leaving Unemployment”. Econometrica, vol47, No5 , pp1249-66. September.
29. [28] OECD (2000) From Initial Education to Working Life.
30. [29] OECD (1997) Education at a Glance.
31. [30] Prentice, R. L. and L.A. Gloeckler (1978). “Regression Analysis of Groupes Survival Data with Application to Breast Cancer Data”. Biometrics 34, pp 57-67. March.
32. [31] Soro-Bonmatí, A. and C. Ianelli (2001). “School-to-Work Transi-
33. tions in Italy and Spain: similar outcomes, di¤erent pathways?”. IV Jornadas de Economía Laboral.
34. [32] Toharia, L. et al.(1998). El mercado de trabajo en España. McGraw Hill.
35. [33] Toharia, L. et al. (2002). Flexibilidad, juventud y trayectorias laborales en el mercado de trabajo español. Opiniones y Actitudes 40. CIS.

A Tables

Source: OECD (1997)

A=Low secondary education or less; B=Upper secondary education; C=University leve Table 1 Table 2

Source: Labor Force Survey and Dolado et al. (2000)

Table 3

Table 4

Table 5

Table 6

B Graphs

Figure 1: Graph 1: Attachment sample. Predicted hazard rates by educational attainment. Figure 2: Graph 2: Attachment sample. Predicted hazard rates by previous job tenure.

Graph 3: Transition Sample. Predicted hazard rates by educational attainment.

Graph 4: Transition Sample. Predicted hazard rates by educational attainment.

Graph 5: Transition Sample. Predicted hazard rates by educational attainment.

Graph 6: Transition sample. Predicted hazard rates by parent’s employment status

Graph 7: Attachment. Predicted hazard rates by parent’s employment status

C Appendix C: the samples

C.1 Attachment sample.

From the sample of single males and females aged between 16 and 31 years old that live with their parents we keep those that:

Answer the …rst two interviews,

Have no missing interview between two relevant interviews,

Do not attend regular school nor classify as students during the interviews,

Are not in the military service in any of the interviews,

Are employed in the …rst interview,

Are unemployed at least in one interview,

Once they become unemployed, they declare loosing their job because the contract ended or because they were …red.

We drop those that when rehired declare to have been on the new job for more than a year. We end up with a sample of 6278 young males and females.

C.2 Sample for transition.

From a sample ofsingle males and females aged between 16 and 31 years old that live with teir parents we keep those that:

Answer the …rst three interviews,

Have no missing value between two relevant interviews,

Follow regular school or the military service during the …rst interview,

Never go back to school or the military service once they …nish school/military duties,

Once they …nsh scool/military service they become unemployed/inactive,

Never worked before.

We drop those young male and female that when employed declare to be working for more than a year. We end up with a sample of 2299 young males and females.

D Appendix D: the variables

Duration.- It measures in months thelenght ofthe unemployment spell¹⁸. It is the sum of three components: durinici, that measures in months the time the individual reports to have been searching for a job in the …rst interviewhe is unemployed; plus three months for each interview we observe him unemployed; plus three aditional months minus end, that measures the months the individual declares to have been on the job when we observe him going out of unemployment¹⁹.

Censored.- It is a dummy variable takes value zero when we do not observe the individual getting a job.

Potential Experience.- It measures in years potential working experience. It is obtained substracting the years needed to complete the education he reports from the age he is when becoming unemployed. We construct three dummy variables: exper5, that takes value 1 if potential experience is lower than 5 years, exper10, that takes value 1 if potential experience is between 5 and 10 years, and exper16, that takes value 1 if potential experience is higher than 10 years.

Job tenure.- It measures the length ofthe contract previous to becoming unemployed. We construct three dummy variables that take value one depending on whether the lenght of the contract was lower than a year, between one and three years o longer than three years respectively.

Male.- It is a dummy variable that takes value 1 if male.

Age.- It is measured through three dummies: age20, for unemployed younger than 20; age25, for unemployed between 21 and 25; and age31, for unemployed older than 25.

Region.- We control the region through the unemployment rate of the State²⁰ among the population aged 16-24 at the beginning of the unemployment spell.

Studies.- We have constructed seven educational dummies: primary studies,egb, fpI, bup, fpII, short university diploma and long university diploma.

Father’s activity.- We have constructed three dummies to control whether the father is employed, unemployed or inactive.

Mother’s activity.- We have constructed three dummies to control whether the mother is employed, unemployed or inactive.

Father’s education.- We measure the maximun level ofeducation attained by the father through a dicrete variable. It takes value 1 for those with primary studies, value 2 for those with EGB/FPI, value 3 for those with BUP/FPII, value 4 for those with short university diploma and value 5 for those with a long university diploma.

^{1 8}We consider as unemployed those that are unemployed or inactive.

^{1 9}If the individual reports to be working for more than a year, duration will appear with a missing value. If the individual reports to be working for more than 3 months but less than a year, end will take value 3.

^{2 0}Comunidad Autónoma

Mother’s education.- We measure the maximun level of education attained by the mother through a dicrete variable. It takes value 1 for those with primary studies, value 2 for those with EGB/FPI, value 3 for those with BUP/FPII, value 4 for those with short university diploma and value 5 for those with a long university diploma

Unemployment bene…ts.- It measures whether the individual receives unemployment bene…ts at any moment in the nine months following the beginning of his/her unemploymnet spell.

Noregl.- It is a dummy variable that takes value 1 if the indiviual is attending non formal education.

Activ.- It is a dummy variable that takes value 1 if the individual is actively seeking for a job throughout the unemployment spell.

Servmil.- It is a dummy variable that takes value 1 if the young male enters the sample because he is doing the transition from the military service to the labour market. It will take value 0 for all females and for those young males that enter the sample because they are doing the transition from school to the labour market.

E Appendix E

Table 1

Table 2

DOCUMENTOS DE TRABAJO

References

1. 2003-06: “The Role of Education vis-à-vis Job Experience in Explaining the Transitions to Employment in the Spanish Youth Labour Market”, Cristina Fernández.

References

1. 2003-05: “The Macroeconomics of Early Retirement”, J. Ignacio Conde-Ruiz y Vincenzo Galasso.

References

1. 2003-04: “Positive Arithmetic of the Welfare State”, J. Ignacio Conde-Ruiz y Vincenzo Galasso.

References

1. 2003-03: “Early retirement”, J. Ignacio Conde-Ruiz y Vincenzo Galasso.

References

1. 2003-02: “Balance del sistema de pensiones y boom migratorio en España. Nuevas proyecciones del modelo MODPENS a 2050”, Javier Alonso Meseguer y José A. Herce.

References

1. 2003-01: “Convergence in social protection across EU countries, 1970-1999”, Simón Sosvilla-Rivero, José A. Herce y Juan-José. de Lucio.

References

1. 2002-26: “Temporary Employment and Segmentation in the Spanish Labour Market: an Empirical Analysis through the Study of Wage Differentials”, María A. Davia y Virginia Hernanz.

References

1. 2002-25: “Efectos económicos de las inversiones ferroviarias, 1991-2007”, José A. Herce y Simón Sosvilla-Rivero.

References

1. 2002-24: “Assessing self-assessed health data”, Namkee Ahn.

References

1. 2002-23: “Especialización productiva y asimetrías en las fluctuaciones económicas en las regiones europeas”, Jordi Pons Novell y Daniel A. Tirado Fabregat.

References

1. 2002-22: “An Eclectic Approach to Currency Crises: Drawing Lessons from the EMS Experience”, Reyes Maroto, Francisco Pérez y Simón Sosvilla-Rivero.

References

1. 2002-21: “Migration Willingness in Spain: Analysis of Temporal and Regional Differences”, Namkee Ahn, Juan F. Jimeno y Emma García.

References

1. 2002-20: “¿Es relevante el trato fiscal diferencial en el volumen de ahorro de los individuos?”, José A. Herce.

References

1. 2002-19: “Industry Mobility and Concentration in the European Union”, Salvador Barrios y Eric Strobl.

References

1. 2002-18: “The Closed-Form Solution for a Family of Four-Dimension Non-Linear MHDS”, José Ramón Ruiz-Tamarit.

References

1. 2002-17: “Multiplicity, Overtaking and Convergence in the Lucas Two-Sector Growth Model”, José Ramón Ruiz-Tamarit

References

1. 2002-16: “A Matching Model of Crowding-Out and On-the-Job Search (with an application to Spain)”, Juan J. Dolado, Marcel Jansen y Juan F. Jimeno.

References

1. 2002-15: “Youth unemployment in the OECD: Demographic shifts, labour market institutions, and macroeconomic shocks”, Juan F. Jimeno y Diego Rodríguez-Palenzuela

References

1. 2002-14: “Modelling the linkages between US and Latin American stock markets”, José L. Fernández-Serrano y Simón Sosvilla-Rivero

References

1. 2002-13: “Incentivos y desigualdad en el sistema español de pensiones contributivas de jubilación”, Juan F. Jimeno.

References

1. 2002-12: “Price Convergence in the European Union”, Simón Sosvilla-Rivero y Salvador Gil-Pareja.

References

1. 2002-11: “Recent Trends in Occupational Segregation by Gender: A Look Across The Atlantic”, Juan J. Dolado, Florentino Felgueroso y Juan F. Jimeno.

References

1. 2002-10: “Demand- and Supply-Driven Externalities in OECD Countries: A Dynamic Panel Approach”, Salvador Barrios y Federico Trionfetti.

References

1. 2002-09: “Learning by Doing and Spillovers: Evidence from Firm-Level Panel Data”, Salvador Barrios y Eric Strobl.