Higher Education Quality, Opportunity Costs and Labor Market Outcomes Patrizia Ordine*, Giuseppe Rose**
1. Introduction It is well known that labor markets of OECD countries are characterized by educational mismatch. In these markets overeducation is widespread and its dimensions considerably large1. In the UK and in the United States it is estimated to involve a number of workers that ranges between the 17% and the 42% of the whole employed graduate labor force2. In Germany and in the Netherlands this percentage ranges between 17% and 28%3. In Italy, the share of overeducated workers is around 39% but it changes dramatically across degree subjects4. Figure 1 shows the extent of overeducation in Italy considering degree subjects in two geographical macroareas. It is interesting to note that overeducation is almost equally distributed in macroareas where, on the contrary, the industrial structure is very different. As is well known, * Dipartimento di Economia e Statistica, Università degli Studi della Calabria, Via P. Bucci Cubo 0C, 87036 Rende (CS), Italy. E-mail:
[email protected]. ** Birkbeck College, University of London, UK. We would like to thank two anonymous referees for important suggestions and comments. We also thank participants to the European Association of Labour Economists (EALE) annual meeting, 18th-20th September 2008, Amsterdam. The usual disclaimer applies.
Overeducation arises when there are workers in occupations that require less schooling than they actually have. This definition usually matches with the following stylized facts: i) overeducated workers earn lower wages than workers with similar level of schooling who work in jobs that require the level of schooling they have obtained; ii) overeducated individuals earn more than their co-workers who are not overeducated (Sicherman 1991). 2 For the US see Daly et al. (2000) and McGoldrick and Robst (1996); for the UK see Chevalier (2003) and Sloane et al. (1999). 3 See Daly et al. (2000) and Allen and van der Velden (2001). 4 See Ballarino and Bratti (2006), Di Pietro and Cutillo (2006) and Cainarca and Sgobbi (2007). 1
RIVISTA ITALIANA DEGLI ECONOMISTI / a. XIV, n. 2, agosto 2009
267
Percentage of overeducated by degree subject, Norgh-Center Italy 2004 Medicine Chemistry-Farmacy Architecture Engeneering Agricultural science Geo-Biology Law Psychology Science Economics and statistics Teaching Linguistic Litteracy Sport science Political science 0
20 40 Mean of overeducated, North-Center
60
Percentage of overeducated by degree subject, South Italy 2004 Medicine Chemistry-Farmacy Engeneering Geo-Biology Psychology Architecture Agricultural science Science Law Teaching Linguistic Litteracy Economics and statistics Sport science Political science 0
20 40 Mean of overeducated, South
60
Fig. 1. Overeducation in Italy by subject degree and macroarea. Percentages on total employed graduates.
268
the South of Italy is less developed and less industrialized than the NorthCenter, with a large share of public employment and high rates of unemployment (Brunello et al. 2001). In this paper we consider a possible explanation of the occurrence of overeducation and we explore the determinants of this phenomenon in Italy. The issue is particularly relevant for the analysis of the functioning of existing labor markets and for policy setting since overeducation is potentially costly to the economy, to the firms and to the individuals, considering that resources are wasted on non-productive investments. Relying on Ordine and Rose (2009) we argue that the human capital theory can be consistent with overeducation in a long run equilibrium assuming imperfect labor market, incomplete information and heterogeneous firms and individuals. We model a strategic interaction framework where education raises individuals’ productivity depending on individuals’ ability, education quality and firms’ technology and we show that in presence of individuals with heterogeneous opportunity costs, self-selection into education can be inefficient. In this framework it is crucial the role played by the educational quality since it determines the relevance of ability for the individuals’ decisions on human capital investment. The idea that inefficient educational choices can generate overeducation has been recently modeled in a matching environment by Charlot and Decreuse (2005). These authors show that, since job opportunities increase by schooling, and low ability individuals do not internalize the impact of their choices on firms’ behavior in job creation, overeducation may arise. It is important to note that, although the presence of individuals with different ability is crucial for their results, the authors assume that ability is perfectly observed by firms and education generates the same cost upon all the individuals. In their setup, the authors conclude that «the optimal education policy is thus to set a tax on education to deter low ability individuals from participating to high productivity sector»5. In our paper, we assume individuals’ ability is not observed by firms and the cost of schooling is inversely related to it. We show that, by expanding the job-market signaling game (Spence 1973) to keep into account heterogeneous opportunity costs, pooling equilibria may arise. The idea is that low opportunity costs, by lowering the marginal cost of acquiring education, may destroy a signaling equilibrium if the signaling device (the cost of the effort that an individual must exert in order to acquire the academic skills) is not strong enough. In this sense, the educational quality, determining the role that ability has in acquiring education, is crucial for individuals’ choices. Differently from Charlot and Decreuse (2005), increasing
5
See Charlot and Decreuse (2005, p. 235).
269
the monetary costs of education could prevent an inefficient self-selection into education only if higher tuition fees would actually reflect into a higher instructional quality. We point out the importance that education quality may have in reducing overeducation. Interestingly, we highlight that the role of education quality in the allocation process is stronger in less developed local markets characterized by a few technological firms and high rates of unemployment. Very seldom overeducation has been related to the quality of education. Robst (1995) and McGuinness (2003) represent important exceptions. Di Pietro and Cutillo (2006) try to discriminate the impact of overeducation by gender, using a measure of education quality derived from Italian newspapers’ college ranking. We intend to add to this empirical evidence by exploiting the role of some other relevant variables capturing the interactions between the firms’ behavior and the individuals’ choices in a way illustrated in our theoretical model. In particular, we believe that the outcome in terms of overeducation is not independent on the educational quality supplied by universities since it shapes the individuals’ marginal cost of education and the final decision on human capital investment. We also intend to find evidence of a possible differentiated impact of this variable in segmented local labor markets characterized by different levels of economic development, as it is the case in Italy6. We apply Probit models with selection to individual data from a survey carried out in Italy by the National Statistical Institute on the labor market outcomes of a representative sample of students who completed university in 2001 and were interviewed in 2004. We merge this data set with data from the CIVR that ranks the universities in terms of research and provides data on structural features and accounting information7. We find that education quality strongly determines the probability of working in a job position where the degree is actually needed. Moreover, the impact of education quality is higher in southern regions. The paper is organized as follows. Section 2 contains the theoretical setup and Section 3 describes the equilibria of the model. In Section 4 we illustrate the empirical model, and in Section 5 we comment our results. Some concluding remarks are presented in Section 6.
Spatial aspects of overeducation and the emergence of this phenomenon in the presence of limited spatial flexibility of the workforce are discussed in Büchel and van Ham (2002). 7 CIVR is the Italian commission for the academic research evaluation. 6
270
2. The Model In what follows we analyze the occurrence of overeducation in segmented labor market using a general theoretical framework developed in Ordine and Rose (2009). Consider the following setup where individuals and firms act strategically. We assume that there are two types of individuals with ability qa (with a = h , l and qh > ql) that, before entering the job market decide to acquire a level of education e, with e ∈[0 , ∞), involving monetary and nonmonetary costs. Once in the job market, individuals can obtain a wage w working in a firm. Firms set production on the basis of technology T and employ individuals. Individuals’ ability is unobserved by firms. The share of high ability individuals, indicated by g (with 0 ce (∙ , qh) ∀q
(5)
ce (∙ , lh) > ce (∙ , ll) ∀q.
First, notice that we define the quality of education as the set of scientific and technical skills provided by universities that raise the individuals’ productivity. Consistently with our definition of education quality, we assume that the higher is the quality supplied by an institution, the higher is the cost (in terms of effort) that an individual has to sustain to obtain a given qualification (3). Equation (4) represents the so called single crossing property, which implies that the cost of an additional year of education is higher for low ability individuals than for high ability ones. Equation (5) indicates that the cost of an additional year of education is higher for individuals with high opportunity costs than for those with low opportunity costs. In particular, individuals located in areas with high employment rates, would find more costly to acquire an additional year of education than individuals located in areas where the employment rates are low, since the latter have a lower probability of being employed. Considering the utility function (1) the slope of the indifference curves is given by: (6)
mrs(e , w) = ce (e , q , qa , lb) – we (e , q , qa , T).
Given assumption (4) and (5) we know that high ability individuals with low opportunity costs (qh , ll) have the lowest marginal cost of education. On the contrary, low ability individuals with high opportunity costs (ql , lh) have higher marginal cost of education than other individuals do. Within these boundaries, we have the marginal cost of education for high ability individuals with high opportunity costs (qh , lh) and those for low ability individuals 272
with low opportunity costs (ql , ll). Which of these two costs would be larger crucially depends on which of the following relations holds: |ce (∙ , qh) – ce (∙ , ql)| > |ce (∙ , ll) – ce (∙ , lh)|
(7) or
|ce (∙ , qh) – ce (∙ , ql)| ceq (∙ , qh)
where ceq is the cross partial derivative of the cost function with respect to the level and the quality of education. The implication of (10) is that an increase in the quality of education raises the marginal cost of education of low ability individuals more than those of high ability ones. The net effect of an increase in q, results in the fact that ceteris paribus ability becomes more important than opportunity costs in determining a reduction in the marginal cost of education8. 2.2. The Firms Consider a continuum of firms. Each firm f employs only one worker to produce the final output. Before hiring a worker each firm has to decide the technology to adopt. In particular, the firm can choose between high and low Ordine and Rose (2007; 2008) and Rose (2009) consider a setup where education quality affects individuals’ marginal cost of education in a way that is similar to the one presented here. 8
273
technology. We indicate with T = {HT , LT} the firm’s investment in high or low technology respectively. The cost of technology HT is given by df > 0 for all f. The cost of technology LT is normalized to zero. Following Acemoglu (1997), the average productivity per worker is given by: (11)
y= y(e, q,qa ,T)=e 0 +e(q + e!1{q =qh ,T =HT} )
where e0 is a constant and e > 0. From (11) it appears that high technology is complementary only to high ability workers. As a consequence, firms need a credible signal on individuals’ ability in order to invest in high technology. Equation (11) assumes a scenario where the effect of education on worker’s productivity depends on the quality of education and on the match between high ability individuals and high technology firms. We assume that the wage schedule is given by the workers productivity minus a rent D > 0 that remains within the firm9. We assume that the wage structure is compressed which implies that: (12)
D = D (e , T) with De(∙) > 0.
The relation between productivity and wage in the presence of a compressed wage structure is graphically illustrated in Figure 2 where, the higher is the human capital of an individual the higher is the rent that a firm can obtain from him. A compressed wage structure may be generated by many causes. Minimum wages, efficiency wages, bargaining problems and transaction costs represent only some possible sources of wage compression10. Moreover we assume that: (13)
D (e , HT) > D (e , LT) ∀e iff qa = qh
which implies that in the case of a match with a high ability individual, firms are able to capture from him higher rents by investing in high technology than by investing in low technology. This result in terms of wage structure can be achieved by assuming a discrete number of firms that compete à la Bertrand to employ individuals and considering the presence of frictions (as mobility costs) so that, ceteris paribus, some individuals strictly prefer to work for one firm. It is also possible to consider this rent as the result of firms’ collusion in a repeated wage offers game. 10 See Acemoglu (1997) and Acemoglu and Pischke (1998) for a complete and detailed discussion on all the elements that can compress the wage structure. 9
274
y, w y (HT, h) (q + ) w (HT, h) (·) q y (LT,∙) = y (∙, l) (·) e0
w (LT,∙) = w (∙, l)
e Fig. 2. Productivity, wages and rents conditional to firm-individual match.
2.3. The Interaction Process The interaction process consists in the following stages. First, individuals conditional on their ability and on their opportunity costs, choose the level of education e they want to acquire. Secondly, each firm randomly matches with an individual. Firms observe the education acquired by the individual and decide the technology to adopt. Then, production takes place and payoffs realize11. The strategic interaction of this model considers explicitly the externalities generated by low ability individuals that want to signal an ability that they do not have in order to achieve a higher utility. As we discuss in the next Section, the dimension of the «ability effect» with respect to the «opportunity costs effect» is crucial in (partially) re-establishing pooling equilibria. 3. The Equilibria 3.1. A Description of the Equilibria Assume that firms are heterogeneous with respect to the cost that they have to sustain in order to acquire the HT technology. In fact, there is no
See Acemoglu (1997) for another application of random matching between firms and workers in a strategic environment. 11
275
Nature (t = 0)
ihl fh
ihl fl
ill fl
e
e
e
e LT HT
HT
ill fh
LT HT
LT HT
ilh fh
ilh fl e
LT HT
ihh fh e
LT HT
Individual (t = 1)
ihh fl e
LT HT
Firm (t = 2)
e LT HT
LT
u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) u (·) (·) (·) (·) (·) (·) (·) (·) (·) (·) (·) (·) (·) (·) (·) (·) (·) Fig. 3. The signaling game with heterogeneous firms and opportunity costs. Note: ihh, ihl, ilh, ill indicate the type of individual that may arise from the combination of qa and lb. fh and fl indicate high and low cost firms respectively.
reason why firms should have the same cost to obtain the same technology since, for instance these costs may be related to the actual technological endowment of each firm, to the structural characteristic of the specific environment, to spillover effects, etc.12. Assume that there are two types of firms parameterized by df with f = h , l and dh > dl. Indicate with x (0 0 . wx
The quality of education required to avoid overeducation is a function of the industrial structure of the considered economy. The higher is the share of firms that can easily acquire ability-complementary technologies, the higher is the quality of education necessary to avoid overeducation. Ceteris paribus a marginal increase in the quality of education should have a larger impact in reducing overeducation in less developed areas with respect to areas characterized by a large number of firms close to the technological frontier. These are relevant conclusions since, in different industrial environments, education policies may have different impacts in terms of utilization of competencies. 4. Some Empirical Issues The debate on overeducation is based on evidence supplied by recent studies highlighting the incidence of this phenomenon and investigating trends and determinants of its occurrence16. Here, we intend to evaluate to what extent this phenomenon is influenced by the interactions between
16
280
For a survey see McGuinness (2006).
the firm’s behavior and the individual’s choices, in a way illustrated in our theoretical model. We believe that the outcome in terms of overeducation is not independent on the educational quality supplied by universities since it shapes individuals’ utility and the final decisions on human capital investment. In particular, we intend to evaluate if the regional environment is relevant in determining the allocation process and to single out the effect of selected explanatory variables. In this respect, we relate the occurrence of overeducation to a proxy of university quality, to the individuals and households attributes and features and to any other variable that may characterize the socioeconomic environments where individuals’ and firms’ decisions are taken. According to the theoretical model illustrated in the previous Sections, we should expect that the probability of finding a job in a position where the university degree is actually required depends on both university quality and firms’ requirements, conditional on individuals’ attributes and opportunities. In order to test the relevance of these variables in the process of the individual allocation into a job, we apply a Probit model with selection to individual data from a survey carried out by the Italian National Statistical Institute on the labor market outcomes of a representative sample of 26,000 students who completed university in 2001 and were interviewed in 2004. We merge this data set with data from the CIVR that ranks the universities in terms of research and provides data on their structural features. The CIVR data refer to the period 2001-2004 hence they do not exactly match to the period of university attendance of the students in our sample. However, we may consider these data as good proxy of the university characteristics as far as they did not change significantly in the years just after the degree completion. We point out that the universities in our sample are mainly public universities charging very similar tuition fees and without binding selectivity criteria17. We already pointed out that in Italy overeducation is diffused homogeneously in different geographical macroareas that are characterized by very different industrial structures and economic development. In order to evaluate to what extent the educational mismatch is driven by different forces in different geographical environments, we estimate two separate empirical models for northern and southern macroareas. Since overeducation is observed only for employed workers, we need to employ a selection model in order to evaluate its determinants. In fact, we have data containing information on college graduates coming from and operating in very different socioeconomic contexts and selection bias may be a serious obstacle to our in A detailed description of the Italian university system and a comparison with other systems is contained in Jongbloed (2004). Interesting features of the Italian system are also illustrated in Perotti (2002). 17
281
ference. It is well understood that individuals are systematically sorted across jobs and operate their choices of accepting a specific occupation on the basis of their characteristics, attributes and opportunities so that in order to evaluate the probability of mismatch we need to employ a selection model where in the first stage we consider the general probability of being employed and in the second stage we estimate the probability of accepting a job where a degree is not needed. A key ingredient of our theoretical model is the university quality. We argue that it influences the chance of a bad match by determining the typology of signaling equilibria arising in the job market. There exists some evidence that there are consistent positive links between research work in the university and students performance in the labor market (Black and Smith 2004). However, there is scarcity of evidence on the relationship between university quality and job matches. McGuinness (2003) uses a proxy for university quality derived from the 2001 Guardian research score and shows that in the UK there exists a high correlation between university research score and the percentage of overeducated. At the same time his results, derived by estimating a Probit model without addressing the selection bias issue, indicate that there is a little benefit, in terms of job match, from being educated at a superior institution. Empirical estimates of overeducation determinants for Italian graduates are presented in Di Pietro and Cutillo (2006). As a proxy for education quality these authors use the CENSIS-la Repubblica newspaper ranking. However, it is possible to cast some doubts on the fact that newspapers rankings really measure university quality as a «productivity enhancing» device. Especially in Italy, the debate focuses on the fact that these rankings only reflect regularity in didactics and they may be exposed to some criticism related to «grade inflation» phenomena (Jacobs and van der Ploeg 2005)18. To the end of capturing the set of scientific and technical skills provided by the universities, we use the investments on technical instruments, equipment, data bases or software for research activities over total receipts set for investments, as our education quality indicator. We refrain from using outcome or structure indicators such as the proportion of students who graduate on time or the average university grade. These measures may be biased by the mentioned «grade inflation» occurrence, since we could observe an inverse correlation between university quality and the number of students who gradu18 Moreover, as we make clear in the next Section, when evaluating the impact of university quality on labor market outcomes, controlling for pre-college ability is crucial in order to alleviate parameters’ bias due to the possible correlation between the student’s ability and the quality of the institution attended. These controls do not appear in Di Pietro and Cutillo (2006) estimates, casting some doubts on the unbiasedness of their results.
282
ate on time or the average score at the exams. As pointed out in Epple et al. (2006), balance sheets figures may be used to represent university quality. We consider that the share of investments for research activities may provide information on the fact that the university is «research oriented» and we believe it may be appropriate in order to measure quality in the sense that we made clear in the theoretical Section. In our specific case of interest, we estimate a selection Probit model where we associate the occurrence of overeducation to a binary 0-1 dependent variable and we relate it to a set of explanatory variables described as follows. – Basic individual characteristics. We assume that personal attributes may influence the educational choice and the individual’s opportunity costs. Among these variables we include a dummy controlling for the region of residence. Additionally, we use the usual control variables such as age and gender. – Household characteristics. The individual socioeconomic background strongly influences his choices in terms of education and occupation. It is well understood in the economic literature that the cultural and intellectual resources of parents are even more relevant than income and financial constraints in shaping the students’ behavior. Marital status and the presence of sons may also be relevant controls. – Education and ability. In any study of return to education and job match it is essential to control for fields of education. For example, there exists evidence that individuals with a degree in humanities have a higher probability of being overeducated. In Figure 1 it is evident that the share of overeducated is very high among people with a degree in humanities, linguistic, teaching or psychology. We also include variables that should help in controlling for individual pre-college ability. – Job characteristics. The probability of the occurrence of overeducation is obviously determined by the firms’ characteristics in terms of industry, dimension and location. These variables influence the job productivity and its contents in terms of skills. We also include in our empirical model the characteristics of the occupation such as duration and the type of contract. – University quality. We argue that it influences the chances of a good match by determining the typology of signaling equilibria arising in the job market. The variables are fully described in Table 1. However, since we observe the dependent variable only if the individual actually works, we define a selection equation where the dependent variable is associated to a binary outcome which takes the value 1 if the individual is working at the time of the interview, and we relate it to a set of variables determining the probability of 283
Tab. 1. Description of variables in Probit and Selection equations Individual Household Overeducation
Female Age Married Employed Father education Child
Dummy variable for the answer to the question: «Is your degree a required qualification for your job?», Overeducation = 1 if the answer is not, 0 otherwise. Dummy variable indicating the respondent’s sex, Female = 1, 0 otherwise. Respondent’s age at the interview. Dummy variable indicating if the respondent is married, Married = 1, 0 otherwise. Dummy variable indicating if the respondent is working at the interview, Employed = 1, 0 otherwise. Highest grade of years of school completed by respondent’s father. Dummy variable equal to 1 if the respondent has got any dependent child, 0 otherwise.
Education Degree subject
A vector of six 0-1 dummy variables indicating degree subjects: 1) Science = 1 if mathematics, science, chemistry, pharmacy, geo-biology, agrarian; 2) Medicine = 1 if medicine; 3) Engineering = 1 if engineering, architecture; 4) Econ. & Law = 1 if political science, economics, statistics, law; 5) Humanities = 1 if humanities, linguistic, teaching, psychology; 6) Sport Science = 1 if sport science. Technical skill A dummy variable for a group of High Schools: Technical Skill = 1 if Accounting, Teacher training, Vocational; Technical Skill = 0 if Liceo. High school leaving grade Final score by type of high school: H. Sch. grade-lyceum, H. Sch. grade-teacher, H. Sch. grade-accountancy, H. Sch. grade-vocational. University leaving grade Final score. Degree on time Dummy variable indicating if the degree is completed on time (adjusted for course duration), Degree on time = 1, 0 otherwise. Job Employed temporary Firm Size Service sector Public sector University University Quality
Dummy variable indicating if the respondent has a temporary or a permanent contract at the interview, Employed temporary = 1, 0 otherwise. A three level dummy variable for firm size, 50 and 100 employees. A dummy variable indicating if the firm is working in the tertiary sector, Service = 1, 0 otherwise. A dummy variable indicating if the firm ownership is public or private, Public = 1, 0 otherwise. «Investments on technical instruments, equipment, data bases or software for research activities» over «Total revenue set for Investments».
working. We estimate the parameters’ of the two equations simultaneously using the so called averaged log-likelihood function and using the NewtonRaphson algorithm as search direction for the maximum.
284
5. Empirical Results Before turning to the discussion of the results of our empirical model, we should clarify some crucial points. First of all, notice that our measure of overeducation is a subjective one since we consider overeducated graduates who affirm that their degree is not a necessary requirement for their job. There exists a substantive literature comparing the outcomes deriving from subjective and objective measures of overeducation (obtained by technical evaluation by professional job analysts of job positions). However, there is no consistent evidence that these different approaches give rise to systematic and significant bias of the incidence or wage effects of overeducation (McGuinness 2006). Further, in any study of returns to education and job match it is essential to control for innate ability. This is extremely relevant in order to disentangle the effect of the individual’s ability from the influence of university quality on university choices and labor market outcomes. We control for innate ability using pre-college grades variables, adjusted for the typology of high school. The use of high school leaving grades in order to control for ability has been suggested in the literature (Dolton and Vignoles 2000; McGuinness 2003) in order to alleviate the selection bias arising from the possible existence of a positive relationship between innate ability and university quality. We also introduce in our estimated equations university leaving grades and the completion of the degree on time, although we know that these variables may obviously be influenced by the characteristics of the specific university attended. In Table 2 and Table 3 we report the results of our empirical models as described in the previous Section. In order to illustrate the relevance of each explanatory variable we report both estimated coefficients and their mean marginal effect. We point out that our models with selection are well specified since we always reject the null hypothesis of no correlation between the residuals of Selection and Probit equations (r in Table 3). First, notice that in each equation we include regional dummies. We know that in Italy there exist huge regional disparities influencing the probability of getting a degree and the probability of finding a job. In Table 3, we assume that the likelihood of being employed (on which depends the likelihood of the phenomenon of overeducation being observed) depends on a set of personal characteristics and on the education received. We see that both sets of variables influence the probability of working. We find that the probability of working increases with the mark obtained at university. However, the concavity of this relationship may be due to the fact that the best students may choose to follow post-graduate courses, delaying their entrance in the job market. Females appear to be strongly penalized in the southern labor 285
Tab. 2. Overeducation Probit models North-Center Coeff. Constant Female Age Married Child Father Education Medicine Sport science Econ. & Law Humanities Mathematics H. Sch. grade-lyceum H. Sch. grade-vocational H. Sch. grade-teacher H. Sch. grade-accounting Univ. leaving grade Degree on time University quality Service sector Public sector Firm size Employed temporary
0.540 (0.338) 0.065*** (0.026) –0.060*** (0.015) –0.161*** (0.027) –0.062* (0.036) –0.032** (0.015) 0.551*** (0.206) 0.281*** (0.086) 0.510*** (0.036) 0.668*** (0.045) 0.422*** (0.048) –0.192** (0.082) –0.184** (0.081) –0.211** (0.082) –0.201*** (0.081) –0.003 (0.009) –0.111*** (0.029) –0.257* (0.146) 0.047* (0.028) –0.338*** (0.033) –0.046*** (0.009) –0.095** (0.041)
South
Marg. Effect – 0.028*** (0.023) –0.023*** (0.006) –0.063*** (0.011) –0.024* (0.014) –0.013** (0.005) 0.217*** (0.079) 0.111*** (0.034) 0.201*** (0.014) 0.261*** (0.017) 0.167*** (0.019) –0.075** (0.032) –0.072** (0.032) –0.082*** (0.032) –0.079** (0.032) –0.002 (0.003) –0.043*** (0.011) –0.102* (0.058) 0.018* (0.011) –0.130*** (0.011) –0.018*** (0.003) –0.037** (0.016)
Coeff. 1.001 (0.064) 0.084 (0.068) –0.085*** (0.030) –0.080*** (0.048) –0.020 (0.086) –0.064** (0.030) 0.0482 (0.381) 0.573** (0.234) 0.775*** (0.070) 0.768*** (0.104) 0.391*** (0.103) –0.322* (0.181) –0.304* (0.170) –0.338** (0.175) –0.304* (0.170) –0.031 (0.095) –0.045 (0.059) –0.769** (0.300) 0.024 (0.053) –0.340*** (0.102) –0.016 (0.017) –0.013 (0.054)
Marg. Effect – 0.034 (0.027) –0.034*** (0.012) –0.032*** (0.020) –0.008 (0.034) –0.025** (0.012) 0.189 (0.139) 0.218** (0.088) 0.300*** (0.030) 0.289*** (0.047) 0.154*** (0.037) –0.128* (0.072) –0.121* (0.068) –0.135** (0.070) –0.121* (0.068) –0.012 (0.051) –0.018 (0.025) –0.307** (0.120) 0.009 (0.021) –0.134*** (0.036) –0.006 (0.007) –0.005 (0.022)
Notes: i) The dependent variable is a latent variable equal to 1 in case of negative answer to the question: «Is your degree a required qualification for your job?», 0 otherwise; ii) «University quality» is investments on technical instruments, equipment, data bases or software for research activities over total revenue set for investments; iii) Robust standard error in parenthesis; iv) Regional dummies included in all equations; v) *** 1% significant, ** 5% significant, * 10% significant; vi) North-Center and South are defined as in the ISTAT classification of Italian regions.
286
Tab. 3. Selection equations for overeducation Probit models North-Center Coeff. Constant
South Coeff.
1.300*** (0.045) –0.121*** (0.025) 0.170*** (0.027) 0.067* (0.036) 0.215*** (0.027) –0.025*** (0.004) –0.001*** (0.000) –1.577*** (0.043) –0.083 (0.093) –0.499*** (0.041) –0.380*** (0.041) –0.695*** (0.043) –0.569*** (0.149)
0.721*** (0.094) –0.257*** (0.036) 0.091** (0.039) 0.040 (0.071) 0.148*** (0.036) –0.028*** (0.006) –0.001*** (0.000) –1.244*** (0.060) 0.114 (0.197) –0.565*** (0.055) –0.411*** (0.065) –0.534*** (0.059) –0.683*** (0.252)
r LR test (r = 0)
–0.924 21.96*** c 21
–0.857 5.42*** c 21
Observations Censored
13,583 3,978
6,193 3,191
Female Married Child Technical skill Univ. leaving grade Univ. leaving grade2 Medicine Sport science Econ. & Law Humanities Mathematics University quality
Notes: i) The dependent variable is a latent variable equal to 1 if the respondent is working at the interview; ii) «University quality» is investments on technical instruments, equipment, data bases or software for research activities over total revenue set for investments; iii) «University leaving grade2» is the variable squared; iv) r is the correlation index between the residuals of the Selection and the Probit equations; v) Robust standard error in parenthesis; vi) Regional dummies included in all equations; vii) *** 1% significant, ** 5% significant, * 10% significant; viii) North-Center and South are defined as in the ISTAT classification of Italian regions.
market. Institutional quality affects significantly the probability of working in both macroareas. In Table 2, the Probit for overeducation show a series of interesting results. First of all, we notice that the variable capturing university quality in the form of the share of investments for research strongly influences the probability of being overeducated (the parameters are negative and significant at the 5% and 10% level for South and North respec287
tively). At the same time we point out that the size of the coefficient and its marginal effect is higher in the southern area of the country19. This finding is robust to many controls undertaken in the empirical analysis20. Interestingly, overeducation decreases with age and this could indicate that in some cases its occurrence is a temporary phenomenon. Individual’s ability, controlled by high school grades, is an important determinant of job matches. Considering that in Italy teaching, evaluation and effort requirements for course completion strongly change across the specific high schools i.e., lyceum, vocational, teacher-training and accounting, we build four interaction dummies in order to control for the high school final marks depending on the specific high school followed. These variables are all significant at the 5% level in the North and at 10% level in the South. On the other hand, the final college graduation marks are not significantly related to the probability of being overeducated. The completion of the degree on time is completely irrelevant in the South while it appears to be strongly significant in the Northern area. We suspect that this may reflect signaling mechanisms working differently in geographical sub-markets. The individual’s socioeconomic background determines both the probability of working and the probability of working in a firm where the degree obtained is a necessary qualification. The parameters’ estimates significantly show that people coming from more educated families reduce their chances of ending overeducated. Controlling for fields of education is compulsory for studying the phenomenon of overeducation. Indeed, as expected in the majority of cases the dummies are significant with respect to the excluded engineering field. Once selection is taken into account it appears that in the North a degree in Medicine reduces the probability of having a good match with respect to the engineering field. In this respect, it is important to note the negative and significant parameters associated to the Medicine degree in the selection equations. This may be due to the fact that usually students with a degree in Medicine are involved in three years of specialization courses; hence most of them were not in the job market at the time of the interview. Firms’ characteristics also influence the likelihood of overeducation. In the North the occurrence of it is more likely if the firm is small while in both macroareas, skill mismatch is reduced if the firm belongs to the public sector. Overall, our findings highlight that high quality universities may contribute to deter19 The intervals of confidence associated to the relative point estimates do not intersect each other. 20 Robustness checks making separated regressions by gender and excluding graduates in medicine from the sample confirm our main results and are available by the authors on request.
288
mine the individuals’ labor market outcomes in terms of job matches and this effect is stronger in less developed areas. 6. Conclusions In this paper we take the issue of overeducation and we highlight the relevance of studying this phenomenon in the presence of segmented labor markets. We show that if we consider heterogeneous opportunity costs of education, equilibria where individuals with different working ability acquire the same educational level may arise. These equilibria can be characterized by overeducation conditional to firms’ technological endowments. The occurrence of this phenomenon is strongly related to the education quality since it determines the role that ability has in the individuals’ schooling choice. We argue that the larger is the share of innovative firms, the higher must be the quality of education in order to avoid overeducation. We take the case of Italy to find evidence of determinants of overeducation in a segmented labor market. Using a sample of Italian graduates we find that having a degree from research-oriented universities significantly reduces the probability of being overeducated. At the same time, the geographical environment influences the overeducation phenomenon and, in line with our theoretical model, the impact of education quality is higher in less developed regions. We figure out that policies on education may have different impacts in term of utilization of competencies in different industrial environments. Our policy indications point out the importance that instructional quality may have in reducing overeducation and the need to consider interactions with the socioeconomic context in refining policy measures to promote efficient selection into education and appropriate allocation in the labor market. References Acemoglu D. (1997), Training and Innovation in an Imperfect Labour Market, in Review of Economic Studies, vol. 64, pp. 445-464. Acemoglu D. - Pischke J.S. (1998), Beyond Becker: Training in Imperfect Labor Markets, in Economic Journal, vol. 109, pp. 112-142. Acemoglu D. - Aghion P. - Zilibotti F. (2006), Distance to the Frontier, Selection and Economic Growth, in Journal of the European Economic Association, vol. 4, pp. 37-74. Allen J. - van der Velden R. (2001), Educational Mismatches versus Skill Mismatches: Effects on Wages, Job Satisfaction and on-the-Job Search, in Oxford Economic Paper, vol. 53, pp. 434-452.
289
Black D. - Smith J. (2004), How Robust is the Evidence on the Effects of College Quality? Evidence from Matching, in Journal of Econometrics, vol. 121, pp. 99-124. Ballarino G. - Bratti M. (2006), Fields of Study and Graduates’ Occupational Outcomes in Italy During the 90s. Who Won and Who Lost?, Department of Economics University of Milan, Working paper #2006-17. Brunello G. - Lupi C. - Ordine P. (2001), Widening Differences in Italian Regional Unemployment, in Labour Economics, vol. 8, pp. 103-129. Büchel F. - van Ham M. (2002), Overeducation, Regional Labour Markets and Spatial Flexibility, IZA DP n. 424. Cainarca G. - Sgobbi F. (2009), Quanto paga studiare? La relazione fra istruzione e retribuzione in Italia, in Economia Politica, vol. 26, pp. 53-84. Charlot O. - Decreuse B. (2005), Self Selection in Education with Matching Frictions, in Labour Economics, vol. 12, pp. 251-267. Chevalier A. (2003), Measuring Overeducation, in Economica, vol. 70, pp. 509-531. Daly M. - Buchel F. - Duncan G. (2000), Premium and Penalties for Surplus and Deficit of Education: Evidence from the United States and Germany, in Economics of Education Review, vol. 19, pp. 169-178. Di Pietro G. - Cutillo A. (2006), University Quality and Labour Market Outcomes in Italy, in LABOUR: Review of Labour Economics and Industrial Relations, vol. 20, pp. 37-62. Dolton P. - Vignoles A. (2000), The Incidence and the Effects of Overeducation in the UK Graduate Labour Market, in Economics of Education Review, vol. 19, pp. 179-198. Epple D. - Romano R. - Sieg H. (2006), Admission, Tuition, and Financial Aid Policies in the Market for Higher Education, in Econometrica, vol. 74, pp. 885-928. Jacobs B. - van der Ploeg F. (2006), Guide to Reform of Higher Education: A European Perspective, in Economy Policy, vol. 47, pp. 535-592. Jongbloed B. (2004), Tuition Fees in Europe and Australasia: Theory, Trends and Policies, in Higher Education: Handbook of Theory and Research, edited by John C. Smart, Kluwer: Dordrecht. McGoldrick K. - Robst J. (1996), Gender Differences in Overeducation: A Test of the Theory of Differential Overqualification, in American Economic Review Conference Proceedings, vol. 86, pp. 280-284. McGuinness S. (2003), University Quality and the Labour Market, in Applied Economics, vol. 35, pp. 1943-1955. McGuinness S. (2006), Overeducation in the Labour Market, in Journal of Economic Surveys, vol. 20, pp. 387-418. Nelson R. - Phelps E. (1966), Investment in Humans, Technological Diffusion and Economic Growth, in American Economic Review, vol. 56, pp. 69-75. Ordine P. - Rose G. (2007), Students’ Mobility and Regional Disparities in Quality and Returns to Education in Italy, in Giornale degli Economisti ed Annali di Economia, vol. 66, pp. 149-176. Ordine P. - Rose G. (2008), The Supply of Education Quality in a Spatial Model with Asymmetric Moving Costs, in Research in Economics, vol. 62, pp. 197-214.
290
Ordine P. - Rose G. (2009), Overeducation and Instructional Quality: A Theoretical Model and Some Facts, in Journal of Human Capital, vol. 3, pp. 73-105. Perotti R. (2002), The Italian University System: Rules vs. Incentives, in Report on Monitoring Italy 2002, ISAE: Rome. Robst J. (1995), College Quality and Overeducation, in Economics of Education Review, vol. 14, pp. 221-228. Rose G. (2009), Higher Education Reforms and Signaling Equilibria, in Journal of Economic Policy Reform, vol. 12, pp. 75-90. Sicherman N. (1991), «Overeducation» in the Labor Market, in Journal of Labor Economics, vol. 9, pp. 101-122. Sloane P. - Battu H. - Seaman P. (1999), Overeducation, Undereducation and the British Labour Market, in Applied Economics, vol. 31, pp. 1437-1453. Spence A.M. (1973), Job Market Signaling, in Quarterly Journal of Economics, vol. 87, pp. 355-374. Abstract: Jel Classification: C24, C73, J24, I20. This paper studies educational choices in a signaling setting in segmented labor markets. We show that in the presence of heterogeneous working ability imperfectly correlated with schooling costs, equilibria characterized by overeducation may arise. The quality of education is crucial in determining the extent of this phenomenon conditional on labor market features. To find evidence of the main implications of the model we use data of Italian graduates merged with the CIVR data set and disaggregated by geographical areas. We find that overeducation is strongly determined by university quality and by other variables that characterize the individual’s socioeconomic background. In particular, the impact of education quality on educational mismatch appears to be higher in less developed areas. Keywords: Overeducation, Perfect Bayesian Equilibrium, Probit, Selection Models.
291
