Education and Local Labour Market: The case of Italy
Gaetano Carmeci and Laura Chies
Department of Economics and Statistics, University of Trieste Piazzale Europa 1, 34127 Trieste, Italy
[email protected] [email protected] Working Papers DiSES 85
Abstract The scope of this paper is to analyse the existing link between the local labour market condition and the decision of further education at the end of compulsory school. Spatial disparities affect in different ways the choice of further education. We investigate these relations using individual data from the Italian Quarterly Labour Force Surveys and aggregated data on regional labour market. We estimate a logit model of choice proving that not only the family background and gender are responsible of education choice, but also the local labour market have an important role. We find that local unemployment rates do influence negatively the decision to invest in further education; whereas the annual variation in unemployment has a negligible effect on youth choice. A cyclical component, that is procyclical for females and anti-cyclical for males, seems to be present in the decision of the Italian youths. A better professional structure in the local labour market influences positively the choice of education, particularly for males.
1. Introduction Industrialised societies are well aware that they cannot limit themselves to diffuse education in general, but in order to be more competitive on the global market, they should also try to match education with the needs of the labour market. This can help also to accomplish social expectations induced by education. The aim of this paper is to analyse which link exists between the local labour market condition and the decision of further education. Spatial disparities affect in different ways the choice of further education after compulsory school. In the literature, particularly the English one, there are various empirical works on this issue. The results are contradictory. In some contributions only the individual characteristics and the social background are matter of choice, whereas the local labour market has a low explicative power. If we look at education as an investment factor, the revenue of education depends also from the probability to find the right job on the local labour market. This is the point in the papers of Bradley e Taylor (1996) and Rice (1999) for England, and of Margirier (1998) for France. These authors affirm that a potentially significant factor affecting educational attainment is the availability of employment opportunities to school leavers. A scarcity of employment opportunities to school leavers in the local labour market may demotivate pupils. This will be reinforced if their parents are themselves unemployed. On the other hand, high unemployment may also encourage more young people to stay on at school hoping employment perspectives improve. The empirical results are contradictory. Some contributions conclude that there is a positive relationship between unemployment and further education decision (Pissarides, 1981; Rice, 1987), others find the opposite result (Micklewright et al., 1988; Gray et al., 1994). In this paper we examine the interaction between education and the local economy for Italy by utilising spatially disaggregated data1. To verify the existence of a “labour market effect” on education choice in Italy, we will control for the individual characteristics and the family background. To this scope we will use individual data derived from the Quarterly Labour Force Survey (QLFS) of ISTAT. In the present work we will highlight that not only the family background and gender are responsible of education choice but also regional variables are of importance. The reminder of this paper is organised as following. It begins with a brief description of theoretical modelling. This is followed by the analysis of the data and the model specification. The model is then subjected to empirical testing. In the last section, we examine the
1
The education system interacts with the local economy in such a way that spatial disparities in economic well-being are exacerbated through a cumulative causation mechanism (see Bradly and Taylor, 1996).
2
effects on the predicted probability of a young person undertaking further education of variations in individual characteristics and in local labour market conditions. The conclusion summarizes the main policy implications of our findings.
2. A theoretical framework The theory of human capital suggests that the revenue of education depends not only from the years of education and the student ability, but also from some other factors. Among these factors we can list the social background, the disposable financial means, the public or private organisation of the school system and the local labour market performance. In particular, the interaction between the education system and the performance of the economy could be important in the case of Italy. In fact, Italy is trapped in a low skilled equilibrium, with the majority of firms being staffed by poorly trained managers and workers, who produce low quality, low value-added products. Economic growth depends on the education level, but also the education level depends on the economic performance, so they are mutually influenced. In fact, the young individual can choose to continue in full-time further education if the market can offer better possibilities of earning with higher level of education or in alternative to work if earning probabilities are high enough without further education. There is sufficient literature on the different probabilities of employment by academic attainment levels and on the causal effect of education on earnings, but there is minor work on the relationship between employment (unemployment) opportunity in the local market and the school choice. In order to investigate this problem we have to focus on the condition of regional labour market and its effect on the years of schooling (see Heckman et al., 1996, Bradley and Taylor, 1996 and Rice, 1999). This is a particularly important relationship for Italy, where the structure of the labour market is characterised by a deep dualism between the north and the south part of the country. The local structure is an important issue in valuating the return of education. Heckman et al. (1996), among others, researched this issue for the United States and they sustain, in line with Oswald and Blanchflower (1994), that regional labour market variables affect only the economic returns of less skilled and less educated workers. The model of heterogeneous human capital, in which high skilled labour trades in a national labour market and low skilled labour is affected by regional labour market shocks, justifies this finding. As we are looking at the choice of young people 15 years old, we expect that this choice will be affected by an aggregate information on future success of further education given by the local labour market structure and by the individual and
3
social peculiarities. It follows that we consider in our model not only individual factors but also labour market variables influencing the return of education. The framework for the empirical analysis is inferred from a simple model of optimal investment in human capital. In our contribution an individual at the age of 15 has a stock of education from compulsory school which may be employed either on the labour market to generate an amount of income that can be utilised for consumption or in the school, in order to acquire additional human capital through education. In so doing the individual chooses between two alternatives maximising the present discounted value of expected future net benefit. In our hypothesis for the general individual i2: Ii = I(Hi) = I(Si)
I’ > 0,
I” < 0
(1)
Where Si is the stock of years of schooling for each individual i at time t, so that Hi can be seen as the stock of human capital at the age of fifteen, since the youth is at the end of compulsory school and has no on-the-job experience as normally it is accounted for in a human capital framework. We assume the current education is directly related to the future income perspective at a decreasing rate. The economic convenience of further education is given by the net present value of the investment. It derives from the maximisation of two different functions representing the discounted value of expected net benefit, one for the full time further education and one for entering the labour market and seeking employment. The present discounted value of expected net benefit to further education is given by:
S
[
]
l
~ V (t 0 ) = ∫ I t S − ct e −βt dt + ∫ I tS e −βt dt t0
(2)
S
In equation (2) the term (ĨSt) is the revenue obtainable at time t during the school time less direct costs (ct) of further education. The expected value increases with the investment in education and decreases if the costs related to further education increase. ISt denotes the expected income at time t after the education period. The alternative is to leave school and seek employment. If this is the case, the individual receives income from employment and in addition, her stock of human capital increases as a result of work experience. The present discounted value of expected net benefits is in this case:
2
See Checchi (2001), p. 28.
4
l
V
w
(t 0 ) =
∫I
n t
e − β t dt
(3)
t0
Where Itn is the expected value of income received from an individual who decide to leave the school at time t03, n is the period of previous education. If the marginal benefit of more education V(t0) is less than the marginal benefit of the decision going to work Vw(t0), she decides to leave the school. The opposite is true when the benefit from further education is higher [V(t0)Vw(t0)>0]. In order to simplify the rate of return of the education choice (β) the expected value must be reduced. Under the constraint that a) the stream of income (ĨSt ISt e Int) are constant in spite of augmenting age and experience, b) the work life is identical between the two choices and c) ĨSt = ct. The solution is immediate: S
∫I t
n − βt
e
dt =
l+S
∫I
S − βt
e
dt
S
And from the hypothesis a) that In and Is are constant terms, we obtain:
IS In
= e βS , individual
with more human capital gets a higher income during her life4. This model does not take account of the important aspect of the local labour market conditions. In order to consider both the regional structure of employment and the probability of obtaining an employment as factors conditioning the choice of further education we have to modify the model. We have to take account also of the fact that a better regional qualification structure can strengthen the education choice. This is a key variable giving the individual better perspectives of income. In fact, a region with a higher skill/unskilled ratio will imply a higher demand for further education. The general employment perspectives may have a positive or negative effect on the education choice and it depends on income perspectives of not schooled youths in the local market. Analysing the professional structure we have to capture the particular character of the independent professions. Italy is, in fact, characterized by a scarce education among the independents even if they are classified in the statistics as skilled, where the professionals are an
3 4
In our hypothesis is the age of fifteen. See Checchi (1999), p. 44ff in the discrete case.
5
exception. Only a quarter of the independent workers have a diploma of upper secondary school (17,7%) or a university degree (laurea) (7,4%), whereas one third of the dependent workers has at least a diploma. Considering the continuous increase of independent activities in Italy, the better perspectives of employment in this sector can negatively influence the choice of youth towards further education. The professional structure is therefore expected to be an important factor in the choice of further education. With respect to the standard human capital model, in order to take account of this aspect we can assume that both Itn and Its are function of the regional professional structure (a), i.e.: Itn = Itn (a) A priori we expect that a higher skilled/unskilled ratio will positively affect the education return. Another important factor that could influence the education decision is the probability to be unemployed as youth so that the discounted value of income is reduced because of the failed increase in human capital (learning by doing) and the missing wage. Under this hypothesis we have to consider indirect costs (opportunity costs) of education. We expect they will be reduced by high and increasing youth unemployment. A contradictory result based on youth perception of the indirect costs can also emerge. We may specify the education cost function as following: c t = f (c s , de , u15 - 19 ) ;
Individual costs of education (ct) are not only due to direct costs of education (cs), but also to indirect costs. Among these latter costs the variation of employment probabilities for individuals (de) may be an important factor to analyse, which has an important regional connotation in Italy. It affects negatively the individual decision of going to school. On the contrary, the probability of not finding a job will be inversely related to the education costs. Youth unemployment rate (u15-19) can be taken as a proxy for this latter probability5. We will sketch hereafter the model we use in the empirical analysis based on what we have highlighted in the last part of this section.
5
The use of this variable and not that of the total unemployment rate, appear to be more appropriate because of the real work perspectives of youths without second level education in the age band 15-19.
6
Two sets of variables will be considered for estimation purposes. One set contains personal attributes such as gender, age and the family background of individual (education, profession and age of parents, number of siblings, etc.). The second set of variables reflects the market condition faced by all individuals (i.e. unemployment rate and its variation, occupation variation, professional structure). For the purposes of empirical analysis, the expected net return of further education for the ith individual of a given region is described by a function of the general form:
(V - V w ) i = R( h i , k, ε i ) Where hi is a vector of observable attributes of the individual and of the family, k is a vector of observable variables which reflect the labour market conditions facing all young people of the region and εi is an unobservable random component influencing the expected net benefits for the ith individual. For the ith member we can define an indicator function, CSi that assumes the value 1 if the individual undertakes a period of further education at the age 15 after compulsory school and 0 otherwise. It follows that:
Pr(CSi = 1) ≡ Pr (R(hi, k, εi) > 0) Adopting a linear approximation for the function R(.) and the logistic cumulative distribution function for the random error εi we obtain a logit model of binary choice further analysed in section 4.
3. Data analysis The analysis focuses on the individual decision to invest in further education. As suggested in section 2, we take account of two subsets of variables. The first one is the family background and the gender of the individual. The second one concerns the labour market performance of the region in which the youths live. In Italy there exist wide North–South differences in this respect, as we will present hereafter. The former group of variables is measured individually or at the family level, while the latter group of variables is measured at the regional level.
7
3.1
The individual characteristics
In this section we present a preliminary data analysis based on the individual data from the Italian Quarterly Labour Force Surveys (source ISTAT). We extract from the dataset the Italian family records with at least a fifteen years old youth having finished compulsory school at the time of the survey. Only the records of the second quarter of each year from 1993 to 1999 are regarded. Table 1 reports the descriptive statistics for our sample. Time and regional dummies indicate the structure of our data. Parent’s education is not very high: over 60% of parents have reached only compulsory school. The mean dimension of the family is of four persons and the one parent’s household has little statistical evidence. About 7% of households have only the mother present and about 2% only the father. Fathers are on average one year older then mothers, whose medium age is about 42 years. Fathers are mainly employed in manufacture, public services, commerce and transportation, whereas mothers are active principally in public services. About 49% of fathers are unskilled, whereas 58% (in the case of youth females) –59.6% (in the males case) of mothers are not working. If we look deeper in the structure of the sample of youths, there are some interesting evidences. About the half of the youth live in the south (47,1%) of Italy and only one third (35,3%) in the North. If we look at the gender a deep North-South difference is evident (Table 2). In the regions of north and central Italy the females continue in further education more than males, the opposite is true in the South. Moreover, the youth males of the northern part of the country seem to be not aware of the education importance. In fact, they present the lowest participation rate in further education (68.8% versus 72.3% in the South). The ISTAT dataset doesn’t permit the extraction of other individual characteristics, so we don’t have any information on the individual, such as ability or preferences. On the other side we have enough variables that describe the family background, such as education, professional and occupational status of parents and we know that these characteristics are very important factors influencing the educational decisions of youth. The statistical analysis indicates that the level of parents’ education is linked to the professional status and the economic sectors of occupation, particularly in the South. More than 70% of fathers working in public services are graduates, but this finding is characterised by a high regional variability. 59% of fathers are in the same condition in the Centre and only 48% in the North. The northern part of the country presents a diversified sector structure with higher work opportunities for graduate fathers, in fact, we find a higher percentage of graduates than elsewhere in the bank and financial market (23,7%) and in
8
manufacture (15,1%). It is also to notice that the majority of fathers work in manufacture and commerce in the North and Centre and in the public sector and commerce in the South. To summarize, data suggest that better educated parents work mainly in two sectors, i.e. in the bank system and in the public services and that their children are willing to continue education (respectively 80.8% and 78%), whereas the children of farmers and of construction’s employees have a minor bent to further education (see table 4). If we analyse the data on the professional status by level of education of parents, we observe that if they are employed in less skilled occupations they have also little education. 88% of mothers have almost a vocational school, whereas 91% of unskilled fathers has no upper secondary school. A larger divergence can be observed between the skilled males and females. With respect to males, females with higher education present more skilled jobs (72.5% versus 65% of males). With a better professional status of parents (and in particular of father) the choice of further education is likely to be almost certain. Children of skilled continue studying in 82% of cases, otherwise only 70% of children begin the secondary school. 8.8% of youth in the sample has unemployed parents. In this case the attendance of a secondary school decrease substantially. It stands by about 60% for youth with a parent in unemployment but it is even lower when is the father that is unemployed. Overall, it seems to be the high level of education of both parents that induce the children to continue in education6, but if we look at the quota on total sample of educated parents, we note that only 30% of youth have a mother with high education (diploma or laurea – university degree) and 37% have a good educated father. Given the other factors, the probability of youth to continue in education decreases dramatically to about 55%, when no parents are in the family.
3.2
The regional variables
As we have stressed in the previous sections the education choice depends not only from the individual and socio-economical background, but also from the local labour market condition. We consider the following indicators of regional economic performance. The first one is the difference between the regional and Italian value of the ratio skilled /unskilled. This variable is depicted in graph 1 for the period 1993-99. It is evident how the differential ratio for central Italy is positive with respect to the Italian mean value. This can be explained by the fact that there are proportional more state and private employees, entrepreneurs
9
and independent professionals in the Centre with respect to the other regions of the country. The graph 1 makes also clear that there is a gender bias, caused by the higher proportion of both skilled males in the Centre of Italy and skilled females in the South. It is precisely during the Nineties that a continuous improvement in the professional structure of females in the South has taken place. In the same period the females in the Centre are performing worse. It seems that only the persons employed in northern Italy are characterised by a higher proportion of unskilled males and females. This is probably due to the old industrial structure of northern Italy marked by a high amount of blue collars. During the last decade the better perspective on the labour market and the constant increase of employees in the North induces a relative improvement in the professional structure of females. The second indicator we consider is the unemployment rate of young people in the age band 1519. Graph 2 evidences that the Italian youth unemployment rate is generally higher than that of the northern regions except for the Liguria. In the south the contrary is true. Only the region Abruzzo is below the mean rate for Italy. On this issue is also to stress that youth unemployment is higher for the macro-regions with a better professional structure. This can be interpreted in various ways and our empirical analysis in section 4 can give some explanation on this topic7. In order to take account of the occupational dynamics of the Italian regions we have considered two other variables: the annual change in the regional unemployment rate and the annual change in the regional employment rate. The reason to take separately account of these two effects is twofold. First of all, the two variables are not always varying in the opposite direction and with the same magnitude because of the frictional effects of time and of the different level of labour demand. Moreover, changes in employment and unemployment could have divergent effects on the education choice of male with respect to females. Looking at the graph 3 and 4, the sign of the unemployment variation is the same in the second half of the 90ties for the males in all areas of the country. In the female case the economic recovery produces positive effects in the north, but in the south the unemployment increases all the time around. The positive variation of employment turns up particularly for the females in the north-central regions beginning from 1995, where the positive variation is yearly over 3% in various regions: Marche, Umbria, Toscana, Emilia Romagna and Friuli-Venezia Giulia. The male
6
85% of youth with both good educated parents continue to study. The higher unemployment may be the result of an historical process due to an early concentration of manufacture in the north and later introduction of public enterprises and services in the south. Or it can be explained by the mean of the economic theory. The search theory suggests that an individual is willing wait for a long period performing a better job at the end. The theory of contracts justifies the involuntary long-term unemployment by the rigidity of labour market. A more general explanation is that a scarce demand of products implies a latent demand of workers. 7
10
employment too is increasing, unfortunately only beginning from 1998 and by a yearly rate of about 0.5%.
4. Estimation and results In this section we present the Maximum Likelihood (ML) estimate results of the parameters of the logit model based on the individual data taken from the Italian Quarterly Labour Force Surveys 1993-1999. We have included in the model as explanatory variables characteristics of the individual and their family and measures of regional labour market conditions, as described in section 3. They could affect the probability of further education. In addition, the econometric specification takes account of the existing heterogeneity at the regional level by including a set of regional dummy variables for the 19 Italian regions8 to allow for any regional effects not captured by the local labour market conditions. Eventually, as micro-economic data are plagued by heteroscedasticity problems, we consider a multiplicative model of heteroscedasticity (see Green, 1993) that it is not rejected by the data. It ensues that the logit model with multiplicative heteroscedasticity used in the analyses has the following general form: β ' xi , i = 1,..., N , Pr(CS i = 1 / xi , z i ) = Λ ' exp(δ z i )
where Λ (w) =
exp(w) 1 + exp(w)
denotes the logistic cumulative distribution function and exp(δ ' z i ) represents the multiplicative 2
model assumed for the heteroscedastic variance. The column vectors, xi and z i , contain the two set of variables affecting the probability of studying after compulsory school for the youth i, in the mean and in the variance respectively, while β and δ are the respective vectors of parameters to be estimated. Notice that in order to be able to estimate all the parameters z i cannot have a constant term. In the application we pool together the 1993-1999 samples and the time dummies for the three northern, central and southern macro-regions are included in order to control for temporal heterogeneity.
11
The analysis performed with the aggregate estimates highlights that gender differences cannot be caught by a single fixed effect, so the analyses are performed also separately for males and females. In section 4.1 we present the results for the aggregate analysis, while in section 4.2 and 4.3 we present the results for the males and females analysis respectively.
4.1
Aggregate analysis
The parameter ML estimates9 are reported in Table 5 for the total sample of 16,129 observations. It is well known that if some relevant variables are omitted from the logit model, coefficients are not consistently estimated, even if these omitted variables are orthogonal to the inserted ones (see Gourieroux, 2000). Therefore a general to specific approach is advisable in this context. We notice that a high improvement in the fit of the models has been obtained by including the regional labour market variables. In the columns under the label General model we report the initial aggregate model estimated without correcting for heteroscedasticity, while in the right side of the table we report both the coefficients estimates and estimated marginal effects (computed at the means of the covariates) for the final model. In the final model we omit variables jointly insignificant at the 5% nominal level. Whereas we take account of the fact that in the general model estimates the Italian central regions and the professional level of the father result to be significant in explaining the variance term according to the LM test we performed10 including in the final model two dummies. At the bottom of the table we report some tests results and a 2x2 table of hits and misses using the standard prediction rule that cˆsi = 1 if estimated probability is greater than or equal to 50%. As we can see the null hypothesis that all the slopes β are zero is rejected both in the general and in the final model. There is a tendency for both models to over-predict participation in further education, but the final model that corrects for heteroscedasticity works a little better in this respect.11 To evaluate the marginal effects we take as base for the general model an individual with following characteristics: 15 years old, male, interviewed in 1993, who have both parents in the family and live in the Italian northern region Trentino-Alto Adige, with both parents in
8
Italian regions are twenty but we have no data for the region Valle d’Aosta. All estimations are carried out using LIMDEP 7.0 for Windows. 10 Tests results can be obtained on request by the authors. 11 This model is preferred to the restricted model estimated imposing δ = 0 . 9
12
occupation, unskilled and no degree; the father works in the agriculture sector and the mother in the construction sector. As some economic sectors of the parent activity result to be insignificant, we don’t consider these dummy variables in the final model. Therefore in the final model the reference cases is characterized by individuals with father not working in manufacturing, commerce or transportation in the south of Italy, and by a mother who does not work in the other private services in the Centre of Italy. Looking at the estimation results we notice that the effects of time dummies are positive, and tend to increase until 1997, but while time dummies for both northern and central regions are highly significant, those for southern regions are jointly highly insignificant. Regional dummies result to be jointly significant and have a positive effect (with respect to Trentino-Alto Adige), implying a remarkable spatial heterogeneity. We differentiate the effects of the activity sectors of the father/mother with respect to North, Centre and South as the production structure is differently characterized in these areas. As we can see from Table 5 the marginal effect of this variable on the probability of going to school is very low and significant only for some sectors. The result is probably due to the fact that we are yet controlling for other more relevant family variables, such as the educational and professional levels of parents and household composition. As far as gender is concerned, it is interesting to notice that only females living in the North of Italy have a significant increase in the probability of further education with respect to males. Compared with the base group, the presence of only one parent in the family increases the probability to continue studying, whereas the missing of both parents reduces such probability. A large number of children in the family have a negative impact, whereas the presence of a large number of other persons has also a negative but negligible effect. Eventually, the age of both the father and the mother have a significant but less pronounced effect on the education choice. The indicator of social background used in this study is based on the professional and employment status and on the educational background of parents. A higher professional level of the father increases the probability to continue studying, whereas if the father is not in employment a decrease in this probability emerges. We obtain the same result for the mother, but the effects on the choice are statistically insignificant. Dummy variables for the educational level of both parents are highly significant. The estimated marginal effects are very high and as expected, they increase with the level of education.
13
As far as the effects of the regional labour market are concerned, we can see that both skilled/unskilled structure and total unemployment rate in the 15-19 band are significant in the general model, whereas only the former variable is significant in the final model at the 5% significance level. In our model the skilled/unskilled structure variable is defined as the difference with respect to Italy in the skilled/unskilled ratio. Therefore for those regions having a ratio below (above) the Italian mean value the effect on the probability is negative (positive). In section 3 we have seen that this difference is positive for the central regions and negative for the other regions in the 90ies. The annual change in unemployment and in employment are not significant in both models, but it is interesting to notice that while the total unemployment rate has a negative effect on the probability of going to school, the annual change in unemployment has a positive effect. We will show in the following sections that the annual change in employment results to affect significantly the choice of both males and females but with the opposite sign. Therefore in aggregate the two effects tend to neutralize theirs self. Other studies have emphasized that there is a different behaviour of males and females in the education choice (see Rice, 1999; Bradley and Taylor, 1996). So a better understanding of the variables affecting the probability to continue studying and of their effects can be obtained by analysing separately the sub-samples by gender.
4.2 Males analysis
In Table 6 we present the estimates of the selected final model and the estimated odds ratios for the sub-sample of males. The final model has been selected starting from the general model presented in Table 5. We proceed eliminating insignificant variables as well as replacing the aggregate regional labour market variables with that of the males. We obtain results qualitatively similar, but the significance of the male regional variables results to be improved. As in the aggregate analysis, we obtain a better fit of the model by introducing regional labour market variables as well as correcting for heteroscedasticity. In the model for the variance term only the dummy for the central regions results to be significant. Notice that in Table 6, there are two columns for the odds ratios: one for the males living in the Centre of Italy and one for the others. As we can see estimated odds ratios are higher (lower) for males not living in the central regions when the variable affects positively (negatively) the probability of going to school. This result is due to the positive coefficient of the
14
dummy Centre in the model for the variance. It means that, for example, a young boy with a skilled father has a probability to proceed in further education 1.423 times the probability of a young boy with an unskilled father. This is true if they do not live in the Centre of Italy, whereas the odds ratio reduces to only 1.276 if they live there. As far as temporal and regional heterogeneity is concerned, we notice that both regional and time effects are significant for the north, the centre and the south of Italy. Moreover, looking at the odds ratios for these variables we can see that regional heterogeneity is more pronounced than temporal heterogeneity. While the maximum value of the odds ratios for the temporal effects is obtained in 1999 for the Central regions and is equal to 2.432, higher odds ratios are obtained for most of the south regions. Finally with respect to the model estimated on the total sample we notice a greater difference in the pattern of time effects for the three macro-regions. The results for the effects of family background are very similar to those for the aggregate sample. As mentioned in the previous section an interesting result emerges as far as the local labour market variables are concerned. Contrary to aggregate analysis, the coefficient of the annual change in male employment is negative and significant at the 1% level. The same result is obtained using the annual change in total employment. We finally notice that the rate of youth unemployment for the males is significant at the 5% level.
4.3 Females analysis In the case of young females we expect to observe some differences with respect to the male results, as we have outlined in the section 3. We notice that compared to the aggregate analysis only the professional status of the father is significant in the variance specification and enters with a positive sign12. As we can see from the Table 7 time effects are no more significant. This result supports the assertion that, as far as educational choice is concerned, female behaviour is less variable in time. On the contrary regional fixed effects result to be of high importance in determining the choice of further education. We can see from Table 7 that the majority of odds ratios for regional effects are greater than two. 12
In the case of a regressor affecting positively (negatively) the probability of further education a young girl with an unskilled father will have a higher (lower) odds ratio than a girl with a skilled father. Actually, also in the model without heteroscedasticity the group of girls with an unskilled father will have odds ratios higher (lower) than the
15
The parents employment sector highlights some minor differences by gender. The household composition with only the father present is not significant and the effect of the presence of only the mother is higher for the females than for the males. In the female results the university degree of the father has a higher relative effect on the probability to continue studying than in the case of males. The opposite is true for the educational level of the mother. As regards the regional labour market variables we specify the final model in the following way. We estimate the model both with the aggregate regional variables and with the females regional variables. In the case of youth unemployment rate we observe that only the total rate is significant at the 5% level. Both the annual variation in total unemployment and in female unemployment is not significant. Finally, the professional structure and the annual change in employment are significant in both models, but the fit improves using the variables for the female group. As in the male analysis, the sign of the unemployment coefficient is negative, whereas the effect of the annual variation in employment is positive for the females. Therefore a cyclical component, that is pro-cyclical for females and anti-cyclical for males, seems to be present in the decision of the Italian youths undertaking further education after compulsory school.
4.4
Probabilities and elasticity
In this section we report the results on the predicted probabilities and the elasticity with respect to the regional labour market variables for both the male and female analysis. In Table 8.1 and Table 8.2 we compute the predicted probabilities for ten different household profiles (see appendix for the definition of the profiles) for males and females respectively. In Table 9.1 and Table 9.2 we compute the elasticity for the same set of profiles. Calculations for both males and females are performed for the year 1999. We set to zero the dummy variables inserted in the final model to take account of the significant parents’ sector of activity and we set to the sample mean values the parents’ age variable. Finally for the regional labour market variables we use the respective regional values in the year 1999. Profiles from (1) to (10) in Table 8.1 and Table 8.2 are ordered with respect to the value of predicted probability, from the lowest to the highest one. Profile (1) and profile (10) correspond roughly to the case of the “poorest families” and “richest families” respectively.
group of girls with a skilled father. Adding a model for the variance has the effect of decreasing (increasing) further the odds ratio for the group of girls with a skilled father.
16
In Table 9.1 and Table 9.2 we can see that as one moves on the same region from profile (1) to profile (10) the elasticity with respect to regional labour market variables decreases (in absolute value). This implies that the performance of the regional labour market tends to be less important for richer families than for poorer ones as far as educational choice is concerned. Moreover we notice that, among the regional labour market variables, youth unemployment rate has the highest values of elasticity. A high level of regional heterogeneity emerges very clearly from the Tables. For example in Table 9.1, column (1), the elasticity with respect to male youth unemployment rates ranges from –0.16 in Trentino-Alto Adige to –1.23 in Campania; in Table 9.2, column (1), the elasticity with respect to total youth unemployment rates ranges from –0.146 in Trentino-Alto Adige to –0.902 in Campania. Comparing males and females elasticity with respect to the unemployment rate we notice that it presents lower values for females than for males in all Italian regions except Toscana, Marche and Lazio. Looking at the skilled/unskilled structure gender play a role in determining the sign of the elasticity. Regions Friuli-Venezia Giulia, Molise and Sardinia show an opposite elasticity sign by gender. The regions Lazio and Sicily present high and positive elasticity for both genders. The negative elasticity is higher for males in Sardinia, Veneto and Basilicata and in the female case in the regions Marche, Veneto and Toscana. For all profiles females are less influenced by the professional structure. The annual change in employment is of minor importance in the decision of further education, but at least in the year 1999, the elasticity is (in absolute term) higher for the females than for the males.
5. Conclusion In this paper we have shown that the performance of regional labour market are of importance in the education choice of Italian young people during the 90ties. Analysing the information contained in the QLFS, supplemented with the regional labour market data, we find that local unemployment rates do influence negatively the decision to invest in further education; whereas the annual variation in unemployment has a negligible effect on youth choice. This finding do not support the hypothesis that higher rates of unemployment, by reducing opportunity cost of education, increase the participation rate of further education. This decision seems also to depend from the regional employment perspectives. When we take account of gender, the variation of
17
employment interacts negatively by reducing the probability of staying at school for males, the opposite is true for females. A cyclical component, that is pro-cyclical for females and anticyclical for males, seems to be present in the decision of the Italian youths undertaking further education after compulsory school. A better professional structure in the local labour market influences positively the choice of education, particularly for males. For political purposes the results of our analysis are of deep interest. These results support the view that the educational system has to match the professional structure and the conditions of the local market in order to hit the target. Italy presents some problems in this respect. For example, looking at the results for the region Veneto it emerges that, while this region has one of the highest output growth in Italy, the skilled/unskilled structure, ceteris paribus, tends to dissuade youths from further education. Insofar as a high level of human capital is required to preserve the economic leadership, the situation outlined for Veneto and some other Italian regions alike, can be critical for the future and needs a careful attention.
18
Appendix The ten household profiles used in section 4.4 are defined as follows: (1)
both parents in the family, father is not employed, father is unskilled, both parents have no education
(2)
only the mother in the family, mother is employed, mother is unskilled, mother has no education
(3)
only the father in the family, father is employed, father is unskilled, father has no education
(4)
both parents in the family, father is employed, father is unskilled, both parents with compulsory school
(5)
only the father in the family, father is employed, father is unskilled, father with a university degree or higher
(6)
only the father in the family, father is not employed, father is skilled, father with a university degree or higher
(7)
only the mother in the family, mother is employed, mother is unskilled, mother with a university degree or higher
(8)
both parents in the family, mother is employed, father is skilled, mother with compulsory school and father with a university degree or higher
(9)
both parents in the family, father is employed, father is skilled, father with compulsory school and mother with a university degree or higher
(10) both parents in the family, father is employed, father is skilled, both parents with a university degree or higher.
19
References Bradley, S. and Taylor, J. (1996), Human Capital Formation and Local Economic Performance, Regional Studies, Vol. 30.1: 1-14. Bratti, M. (2000), Oltre la scuola dell’obbligo. Un’analisi empirica della scelta di proseguire nell’istruzione post-obbligo in Italia, Atti del XV convegno Aiel, Ancona 2-3 ottobre 2000. Card, D. and Krueger A. (1992), Does School Quality Matter: Returns to Education and the Characteristics of Public Schools in the United States, Journal of Political Economy, 100 (1): 1-40. Card, D. (1999), The casual effect of education on earning, in Handbook of Labour Economics (eds. Ashenfelter O. C. and Card D., North Holland, Amsterdam. Checchi, D. (1999), Istruzione e mercato, Il Mulino, Bologna. Gourieroux, C. (2000), Econometrics of qualitative dependent variables, New York: Cambridge University Press. Green, W. (1993), Econometric Analysis, 2nd edn, New York: Macmillan. Heckman, J., Layne-Farrar, A. and Taylor, L.L (1996), Human Capital Pricing Equations with an Application to Estimating the Effect of Schooling Quality on Earnings, The Review of Economics and Statistics, 78(4): 562-610. Meco, I. and Sestito, P. (2000), La condizione sociale e professionale dei giovani in Italia, Note Lavoro 1/2000, Ministero del Lavoro e della Previdenza Sociale. Isfol (2000), Rapporto Isfol 2000, Roma. Rice, P. (1999), The impact of local labour markets on investment in further education: Evidence from the England and Wales youth cohort studies, Journal of Population Economics, 12: 287-312.
20
Table 1: Descriptive statistics and main characteristics of the sample Variable
University degree and higher 6.3 5.9 Upper secondary diploma 24.5 23.5 Compulsory school 63.4 65.3 No education 3.8 2.8 Family’s characteristics Other persons 11.0 10.0 Number of members 4.3 4.3 Number of children 2.2 2.3 Parents not in household 2.0 1.8 Mother only present 7.0 7.6 Father only present 2.0 2.4 Age of mother 41.9 41.6 Age of father 43.2 43 Economic sector of father (*) Agriculture 5.1 5.7 Manufacture 21.0 20.4 Construction 9.7 9.5 Commerce and Transportation 19.2 19.2 Other private services 5.0 5.0 Public services 20.6 20.0 Economic sector of mother (*) Agriculture 3.3 3.6 Manufacture 6.1 6.5 Construction 0.6 0.3 Commerce and Transportation 8.3 9.2 Other private services 2.2 2.5 Public services 18.4 17.9 Professional status of father (*) Not working 12.5 13.1 Skilled worker 31.7 30.3 Unskilled worker 48.9 49.4 Professional status of mother (*) Not working 59.6 58.0 Skilled worker 19.1 18.5 Unskilled worker 19.8 21.5 Note: (*) Indicate relative proportion in the sample
Mean values Male Female 7968 8161
Observations (16.129) Time dummies (*) 1993 1994 1995 1996 1997 1998 1999 Regional dummies (*) Piedmont Lombardy Trentino Alto Adige Veneto Friuli-Venezia Giulia Liguria Emilia Romagna Toscana Umbria Marche Lazio Abruzzo Molise Campania Puglia Basilicata Calabria Sicily Sardinia Parent’s education Father University degree and higher Upper secondary diploma Compulsory school No education Mother
15.8 15.3 13.7 14.2 13.8 13.6 13.5
15.4 14.8 14.8 13.7 14.2 13.7 13.5
6.4 9.4 4.5 4.5 2.3 2.3 3.7 4.8 1.8 2.3 8.4 2.5 3.3 10.8 8.1 3.3 5.8 10.3 3.9
6.7 10.1 4.0 5.6 1.9 2.2 3.9 4.9 2.1 2.3 8.0 2.7 3.1 10.9 8.7 3.2 5.8 9.7 4.3
7.2 23.6 59.4 2.6
6.7 23.1 60.5 2.2
Table 2: Gender preference for further education North Further education
M
Centre F
M
South F
M
F
YES
68,8
75,2
71,8
77,0
72,3
71,3
NO
31,2
24,8
28,2
23,0
27,7
28,7
21
Table 3: Father’s level of education by region and by economic sector Centre Agriculture
Bank and Commerce Construction financial market
Manufacture Public services
1 PhD
22,2
2 University degree 2nd level 3 University degree 1st level
8,5
2,8
9,0
40,0
57,5 60,0
1-3
0,0
22,0
8,1
2,7
8,5
58,7
4 Upper secondary diploma 5 secondary s. vocational 6 Compulsory school 2nd 7 Compulsory school 1st 8 No education
1,5
16,8
22,1
4,2
24,1
31,3
1,9
3,8
34,3
4,8
31,4
23,8
4,0
2,8
30,1
10,3
30,6
22,2
9,4
2,0
26,0
16,2
34,6
11,9
7,7
7,7
7,7
38,5
38,5
0,0
Tot. by sector
4,1
8,3
25,0
9,2
27,7
25,8
North 1 PhD
0,0
4,0
4,0
4,0
4,0
84,0
2 University degree 2nd level 3 University degree 1st level
1,3
25,7
7,9
4,8
15,9
44,4
0,0
16,7
5,6
0,0
16,7
61,1
1-3
1,1
23,7
7,5
4,5
15,1
48,0
4 Upper secondary diploma 5 secondary s. vocational 6 Compulsory school 2nd 7 Compulsory school 1st 8 No education
2,2
17,1
22,9
7,4
30,4
19,9
3,6
6,6
30,2
7,5
35,9
16,2
4,1
2,2
28,6
9,9
40,0
15,2
11,3
1,6
21,4
20,2
37,0
8,5
17,4
0,0
21,7
26,1
26,1
8,7
5,5
7,1
23,9
11,6
34,9
17,0
Tot. by sector
South 1 PhD
8,3
8,3
8,3
75,0
0,9
14,8
4,7
3,0
5,8
70,8
4,2
4,2
0,0
12,5
4,2
75,0
1-3
1,0
14,2
4,6
3,4
5,8
71,1
4 Upper secondary diploma 5 secondary s. vocational 6 Compulsory school 2nd 7 Compulsory school 1st 8 No education
3,3
11,4
20,7
5,1
18,0
41,5
3,5
2,0
24,7
4,0
33,3
32,3
6,4
2,0
29,4
12,9
19,2
30,0
17,4
1,3
24,2
24,1
19,3
13,8
40,0
0,0
13,3
25,3
14,7
6,7
8,7
4,9
23,7
13,1
18,2
31,5
2 University degree 2nd level 3 University degree 1st level
Tot. by sector
22
Table 4: Sector of parent’s employment and education decision Sector of employment of parents
% of youth of 15 on total sample % of youth in FE 7.1 64,7 Agriculture 22.9 73,2 Manufacture 9.4 66,8 Construction 23.1 75,2 Commerce 6.6 80,8 Bank and finance 30.9 78,0 Public services
23
Graph 1: Skilled/Unskilled Ratio Differences with respect to Italy by macroregions Males
0.08 0.06 0.04 0.02 0.00 -0.02 -0.04 -0.06 -0.08 -0.10 -0.12 1993
1994
1995 NOR TH
1996
1997
CENTRE
1998
1999
SOUTH
Females
0.50 0.40 0.30 0.20 0.10 0.00 -0.10 -0.20 1993
1994
1995 NORTH
1996
1997
CENTRE
1998
1999
SOUTH
Total
0.08 0.06 0.04 0.02 0.00 -0.02 -0.04 -0.06 -0.08 -0.10 1993
1994
1995 NORTH
1996 CENTRE
199 7
1998
1999
SOUTH
24
Graph 2: Total Unemployment Rate of Youths 15-19
25
Graph 3: Variation in Unemployment Rate by Gender (in %)
26
Graph 4: Employment Variation by gender
27
