Higher Education Expansion and Unskilled Labour Market Outcomes Veruska Oppedisano London Metropolitan University, Old Castle Street, E17NT, London, UK

Abstract The increasing demand for higher education reduces the supply and changes the composition of unskilled secondary school graduates, and it may therefore a¤ect their labour market outcomes. However, there is little empirical evidence on these e¤ects. This paper analyses a large-scale expansion of higher education supply in Italy, which occurred at the end of the 1990s, to estimate the e¤ects of the policy on the secondary school graduates’ probability of being inactive, employed, unemployed, and on their wages. Robust di¤erence-in-di¤erences estimates show that the probability of being inactive decreases by 4.5 percent, as the policy signi…cantly displaces individuals from inactivity. Those shifting across educational level have middle ability but favourable parental background, and would have worked in the family …rm, or waited for a public competition had the expansion not took place, indicating that a new campus nearby induces mainly those with a low opportunity cost to enrol in university. Lack of signi…cant e¤ects on the labour market outcomes of the workforce provides evidence in favour of the human capital hypothesis. However, the policy may have induced too little variation in the workforce to distinguish between the human capital and signalling theory. JEL Codes: I23 Keywords: Inactivity, Higher Education Expansion, Signalling, Human Capital.

Present contact details: Email: [email protected], +4402073201279. Address: Faculty of Business and Law, London Mteropolitan University, Old Castle Street, E17NT, London, UK.

1

1

Introduction

The proportion of university graduates has signi…cantly risen in all developed countries in the last decade, especially in southern European ones. According to OECD statistics, over the period 1997-2008, the proportion of graduates in the 25-64 age brackets increased by almost 60 percent in Italy, Spain, Greece, and Portugal, and by almost 40 percent in the UK, France, and Norway.1 Moreover, increasing participation in higher education is a major policy objective in almost all European countries. Whilst most recent research investigated the impact of the increased demand for higher education on the labour market outcomes of skilled-university graduates, less attention was placed on those of unskilled-secondary school graduates (Walker and Zhu, 2008; Chevalier and Lindley, 2009; Carneiro and Lee, 2011; Bosio and Leonardi, 2010). The consistent out‡ow of individuals from secondary into tertiary education reduces the supply and changes the composition of secondary school graduates who do not enrol in higher education (unskilled hereafter), and it may therefore a¤ect their labour market outcomes. However, despite the important implications of this policy for individuals likely more disadvantaged than those who bene…t from tertiary education, the empirical evidence on these e¤ects is limited. In part, this is due to the challenge of identifying causal e¤ects of university enrolment on unskilled youngsters’labour market outcomes. Labour markets characterized by higher participation into tertiary education may be those with a lower opportunity cost in the unskilled sector, and unskilled young adults may experience penalties even in the absence of increasing university enrolment. This paper exploits the variation induced by the sharp increase in the supply of universities over a period of just a few years at the end of the 1990’s to evaluate the e¤ects of increasing participation in higher education on unskilled individuals’labour market outcomes. From the beginning of the 1990’s, the government implemented a supply side policy that resulted in a widespread increase in local institutions, homogeneously scattered across the country, and in an expansion of existing universities, which boosted the range of degrees o¤ered. Oppedisano (2011), estimating the e¤ects of this policy on higher educational outcomes, shows that it signi…cantly and positively a¤ected enrolment into higher education, thereby reducing the supply of unskilled individuals. This paper evaluates the e¤ects of the expansion that took place over the period 1995-1998, after which some regions increased their campuses, while others maintained the same universities’provision, by means of a di¤erence in di¤erences estimation strategy. The Italian higher educational system has traditionally been organized at the national level, which guarantees that titles of higher education attainment are legally valid throughout Italy, independently of the institution that issues them. Universities are indeed perceived as substitutes and individuals enrol in the one nearest to their place of residence. Moreover, the Italian political situation at the beginning of the process of expansion o¤ers an ideal setting for evaluating the impact of the program that limits the possible concerns about endogeneity of the policy one might have. The lack of institutional arrangements allowed the dominant party system to implement public 1

Source: Education at a glance, 2010.

2

policies without de…ning clear instructions and objectives. The increase in higher education supply was hardly driven by an economic rationale and the allocation rule was not clearly spelled out. In particular, the planning of the expansion did not take into account the drop in the potential demand for higher education. Rather, the expansion followed an indiscriminate allocation of public funds across Italian regions. The outcomes of interest are the probability of being enrolled in university, the probability of being inactive, unemployed, and employed, and the hourly wage of unskilled individuals, de…ned here as those who completed secondary education but do not enrol in university. The three choices of being employed, unemployed, or inactive are mutually exclusive and add up to one. The purpose of this paper is twofold. First, to understand where the increasing demand for higher education originates from, and its characteristics, in terms of observable talent and the sorting process of the unobservable component. The new ‡ow of enrollees can originate from the population of individuals in paid works or in …nancially unproductive activities, which has important …nancial implications in terms of foregone earnings and therefore the overall implicit …nancial cost of the policy. This provides information on the impact of the policy on youngsters’ probability of being inactive. The fraction of young people in the 20-24 age brackets not in employment, education, and training has increased in the European Community: it passed from a 7.7 value in 2000 to 8 in 2007, and 8.4 in 2010, also as a consequence of the Global Recession.2 This phenomenon is therefore becoming the focus of concern for policy makers. Understanding the e¤ect of a reduction in the monetary cost of accessing higher education on youngsters’inactivity has important policy implications. Second, if the increasing demand for higher education originates from the workforce, …ndings will provide information on whether labour market outcomes are determined under the signalling or the human capital theory. Whilst the signalling model establishes that even though investment in education might be pro…table for single individuals, they are not bene…cial for the society as a whole, the human capital one predicts that an increase in educational attainment raises productivity and economic growth. The controversy in the debate between human capital and signalling has been di¢ cult to resolve because the estimation of the earning equation universally reveals a positive causal e¤ect of years of schooling on earnings, which is consistent with both the human capital theory and the signalling one. The two models have di¤erent predictions on the e¤ects of increasing higher education participation on unskilled labour market outcomes: while under the signalling model the value associated with secondary education may decline if it re‡ects a lower ability composition of secondary school graduates, under the human capital model it may increase if skilled and unskilled workers are imperfect substitute in production, and it should not be a¤ected otherwise. Results indicate that the probability of being enrolled in university increases by 2 percent in regions were new campuses were instituted. The new in‡ow originates from the workforce and from unproductive activities, with the e¤ect on the latter being 2

Source: Eurostat statistics, Indicators of Youth. Statistics before 2000 are not provided.

3

larger and signi…cant. Those shifting educational level as a consequence of the policy have on average middle ability and fathers with a college degree. When looking at the reasons for not enrolling in university, the policy signi…cantly reduces the probability of claiming not to be enrolled because employed in the family …rm or because waiting for a public competition. This suggests that instituting a new campus nearby when the potential demand for higher education experiences a decline induces mainly those with a low opportunity cost to enroll in university. Lack of signi…cant e¤ects in the labour market outcomes of those in the workforce, together with a signi…cant reduction in unemployment in Northern regions, seem to be consistent with the hypothesis of the human capital model. However, as the ‡ow of new enrollees does not signi…cantly a¤ect the workforce, the policy may generate too little variation in the workforce to distinguish between the human capital and the signalling theory. The paper is related to two di¤erent strands of the literature. First, it speaks to the literature that analyzes the impact of the increasing supply of graduates on wages and overeducation level by providing a new focus on unskilled individuals’ labour market outcomes (Walker and Zhu, 2008; Chevalier and Lindley, 2009). Second, the paper is related to the literature that tries to distinguish between the signalling and the human capital theory (Wolpin, 1977; Chevalier et al., 2004; Bedard, 2001; Albrecht and van Ours, 2006; Hämäläinen and Uusitalo, 2008; Brown and Sessions, 1999 and 2006; Hussey, 2012; Miller et al., 2004). Within this literature, the paper is closest to Bedard (2001), who evaluates high school dropouts’responses to the presence of a local university. Lack of information on secondary school dropouts in my data prevents me from applying exactly the same test on the Italian experience. The remainder of the paper is organized as follows. Section 2 illustrates the characteristics of the expansion of the higher education supply in Italy and the data. Section 3 details the identi…cation strategy. Section 4 describes the conceptual framework. Section 5 presents the empirical results and Section 6 the robustness checks. Section 7 contains …nal remarks.

2 2.1

Institutional background and data Institutional background

Since the Italian constitution was established, the university system has traditionally been organized by central approval. This centrally organized procedure has always guaranteed that titles of higher education attainment were legally valid throughout Italy, independent of the institution which issued them. A substantial process of university expansion started up at the beginning of the 1990’s. This expansion was driven by two broad rationales: (i) the necessity to spread the accessibility to university homogeneously across the territory in order to balance funds allocation and university sites nationwide; (ii) the need to decongest overcrowded universities, which at that time exceeded forty thousand students enrolled. The Italian political situation at the beginning of the process of expansion o¤ers an ideal setting for evaluating the impact of the program which limits the possible 4

concerns about endogeneity of the policy one might have. The lack of institutional arrangements allowed the dominant party system to implement public policies without de…ning clear instructions and objectives. In fact, the increase in higher education supply was hardly driven by an economic rationale and the allocation rule was not clearly spelled out. As emerges from a series of ministerial documents that evaluate the process of expansion of the higher education supply, the implementation did not follow any cost-bene…t analysis, the evaluation of the evolution of potential demand for higher education, possible job opportunities and existing infrastructures. In particular, the planning of the expansion did not take into account the declining rate of university enrolment, re‡ecting a lower demand for higher education and the demographic decline in the number of young adults in the university age population.3 Rather, the increase followed an indiscriminate allocation of public funds across Italian regions. The Ministry of Education and Research assessed "...with respect to the development and rebalancing of university premises prevailed - at least for the most part - a non selective ‘all over the place’approach, inspired by a barely incremental purpose...".4 The amount of resources allocated to this program was around 1.5 thousand millions of euro in six years.5 Resources were fully provided by the central government without any …nancial involvement of the local and regional administration.6 Because of delays on resources assignment, some campuses, whose institution was forecast at the beginning of the 1990’s, were e¤ectively established at the end of that decade. The program triggered the birth of a number of smaller campuses. Besides, all universities increased the supply of degrees, either by creating new departments or by o¤ering new courses within existing departments. Over the period 1995-1998, which is where I place my focus on due to data availability, the number of public campuses increased from 69 to a total of 80.7 Figure 1 depicts the territorial distribution of the expansion. Eight regions increased their supply through the institution of a new campus (Val D’Aosta, Piemonte, Lombardia, Trentino Alto Adige, Marche, Molise, Puglia and Sicilia). In the Emilia Romagna region, three new campuses were instituted. In the Campania region, the total number of campuses did not change because of the closure of the second university, which opened a subsidiary campus nearby. The remaining eleven regions maintained the same number of universities.8 I de…ne as "treated regions" those 3

These documents can be downloaded at http://www.vsu.it/ website, under the "Publications and Documents" section. 4 Ministry of Education and Research (1997) "Veri…cation of universities’development plan 1986-90 and 1991-93", doc. 4/97, pg. 10. 5 Amount in euro, indexed at prices 2004. 6 The centralized …nancing structure of the higher educational system in Italy rules out the possibility that local government’s investments in higher education crowd out resources in other sectors of the Public Administration, thereby a¤ecting the supply of jobs in the public sector and indirectly unskilled individuals’labour market outcomes. 7 In this process of expansion private universities were also founded. However, changes in the supply of private universities ruled by private enterprises are left out of consideration because procedures di¤erent from those applied for public universities were applied and because other dimensions, such as family wealth, a¤ect the choice of entering private universities. 8 Regions are the …rst order administrative subdivision of the Italian state. There are 20 regions

5

where the number of campuses changed over the period 1995-1998, while the remaining ones are "control regions".9 To understand the potential impact of this reform, it is useful to brie‡y describe how access to the Italian higher educational system is regulated. After …ve years of non-compulsory secondary education, secondary school graduates have a legal right to study in any …elds in the university sector, regardless of whether graduation was obtained in an academic, technical, or vocational secondary school. Universities do not screen diploma’s holders to select admission, with the exception of some courses de…ned by the law (Medicine, Veterinary, Dentistry, Pharmacy), whose places are awarded on the basis of a general selection procedure.10 The higher educational system is public, and partly …nanced by tuition fees, whose average yearly amount was e 546 in 1999.11 Another important characteristic concerns the quality of the supply of higher education. As the value of a degree is established by the law, all degrees are equal and there is no incentive for universities to compete for quality and prestige. Universities are indeed perceived as substitutes and often chosen on the basis of the proximity to the place of residence.12 These characteristics make new universities likely to attract individuals who reside nearby.13

2.2

Data

I use data collected from the 1995 and 1998 "Survey on School and Work Experiences of Secondary School Graduates", a cross-section of a representative sample of secondary school graduates interviewed three years after graduation (in 1998 and 2001). The data contain a wide range of information on the school curriculum and on the postschool experiences, either in university or in the labour market. Moreover, information on personal characteristics, family background, region of residence during secondary school and year of enrolment is available. The estimation sample includes 34,149 observations, 16,417 of which belong to the in Italy. Their size ranges from almost 10 million inhabitants in Lombardia to 125 thousand in Val D’Aosta. In 1998, the ‡ow of young adults with a diploma from secondary school, the population of interest in this paper, amounted to 64 thousand in Lombardia, and to less than 800 thousand in Val D’Aosta. 9 Some of the new universities are located at the borders between treated and control regions. This would constitute a problem if individuals in control regions moved to treated regions to bene…t from the policy. However, only 3.5 percent of those living in a control region enrol in the neighbouring treated region where a new university is located close to the borders. 10 None of the new universities opened courses that require a selection procedure. 11 Source: Ministry of Education and Research, Statistical Department. 12 Only 15 percent of students enrol in a university placed in a region di¤erent from the one of residence (Source: ISTAT). 13 Depending on the degree of substitutability between skilled and unskilled labour, unskilled labour market outcomes may be di¤erently a¤ected by the composition of degrees provided. In the present analysis, the composition of degrees supplied can be neglected, as the interaction between the higher supply of graduates in a speci…c degree and unskilled workers will exert its e¤ects on labour market outcomes only in the medium term.

6

1995 cohort and the rest to the 1998 one. The 1995 and 1998 repeated cross sections of individual data are pooled and information matched with regional-level data on the supply of higher education, both in 1995 and 1998 (years in which secondary school graduates were interviewed). Information about the regional supply of higher education, including campuses, faculties and degrees courses, is taken from the annual ISTAT report "Statistics of Higher Education". Table 1 shows baseline summary statistics for treatment and control groups conditional on not being enrolled in university (except for the enrolment variable) before the policy implementation to assess whether the program creates comparable groups.14 The third column presents average di¤erences between treatment and control groups with standard errors in parenthesis. There are no statistically signi…cant di¤erences between individual observable characteristics of the treatment and comparison group, except for: age, slightly higher in control group; the fraction of individuals whose fathers have a university degree, slightly lower in control regions; the proportion of youngsters with a technical school degree, lower in control region; the fraction of graduates from professional secondary school, higher in control region; the unemployment rate of not enrolled, slightly higher among individuals living in control regions; …nally, the logarithm of hourly wage, slightly higher in treated regions. It is worth bearing in mind that the regression analysis controls for all pre-policy observable characteristics. Slightly more than half of respondents are female; average respondent’s age is about 22.5; roughly three fourths are composed of youngsters whose parents have a primary education, one fourth of youngsters whose parents obtained a secondary school diploma and 2 percent have college graduated parents. Marks at the end of secondary school show a prevalence of individuals who obtained a low D mark. More than 50 percent of students attained the diploma in a technical secondary school, 30 percent in a vocational secondary school, and 10 percent in other secondary schools. Only 4 percent of not enrolled students obtained their degree from a high school. Roughly 50 percent of students enrol in university. 25 percent of unskilled young adults are unemployed, almost 70 percent are employed, and the rest inactive. For those employed, the average logarithm of hourly wage amounts to e 1.57. Given that the empirical speci…cation controls also for region …xed e¤ects, it is not necessary that higher education expansion is unrelated to regional characteristics. It is useful, however, to understand the determinants of the expansion across regions. Table 2 presents statistics for treatment and control regions in 1995. In particular, it shows the GDP growth rate, the overall and the youngsters’unemployment rate, enrolment rate and number of campuses per 10,000 19 year-olds. The reported statistics show that the two groups of regions are similar along a list of economic dimensions. The only notable di¤erences are enrolment and the ratio of campuses coverage to the population of 19 year-olds, both lower in treated than in control regions. This conforms well to intuition, since government investments were higher in regions with lower enrolment and few higher educational institutions, but also because the local political pressure for 14

Individuals resident in the Campania region are considered as treated for this descriptive analysis.

7

expansion of higher education is likely to be stronger in areas where higher education was rationed.

3

Identi…cation strategy

To estimate the e¤ects of the expansion of the higher education supply on unskilled youngsters’labour market outcomes a di¤erence in di¤erences strategy is applied, exploiting that the supply shocks to higher education were larger in some areas than others. The main empirical strategy consists in comparing unskilled individuals’labour market outcomes before and after the expansion, from regions where a new university was instituted (the treatment group) and regions with no change in the supply of higher education (the control group). An individual’s treatment is jointly determined by his year of graduation from secondary school and his region of secondary school attendance. The young people who left secondary school in 1995 did not bene…t from the program, since the higher education expansion only came into force between 1996 and 1998, whilst individuals who terminated secondary school in 1998 were fully exposed. A second source of variation arises from the expansion of higher education supply across regions. Evaluating the enrolment decision and the labour market outcomes according to the supply of higher education in the region of secondary school could downward bias the coe¢ cient estimate of the policy because migration introduces measurement errors. On the contrary, assessing enrolment and labour market outcomes according to the supply of higher education in the region where university is attended or a job found could give positively biased estimates because of endogenous selection. To rule out bias induced by endogenous migration, the outcomes of interest are evaluated on the basis of the exogenous supply of higher education in the region of secondary school.15 The basic idea behind the identi…cation strategy can be illustrated using a simple two-by-two table. Table 3 shows di¤erences of outcomes’ means, computed at the regional level, from 1995 to 1998 by control and treatment groups.16 The table provides an illustration of the identi…cation strategy. A list of outcomes of individuals who have no exposure to the program is compared to those of individuals who are exposed to the program. Outcomes of interest are the following: enrolled, a dummy equal one if the individual is enrolled in higher education; inactivity, a dummy equal one if the individual is not looking for a job; employment, a dummy equal one if the individual is employed; unemployment, which takes the value of one whenever the individual is unemployed; 15

Province of residence may be a more precise variable for assigning treatment status. However, information on the individual’s province of residence is available only at the ISTAT in Rome. As results are not qualitatively di¤erent when using region of residence, I prefer to assign treatment on the basis of the region rather than the province of residence. 16 The average values reported in this Table di¤er from Oppedisano (2011). This table reports population weighted means value obtained at treatment and control groups level; the table in Oppedisano (2011) reports average outcomes …rst de…ned at year and regional level, and then collapsed with population weight.

8

…nally, the logarithm of hourly wage. The labour market outcomes refer to unskilled individuals, secondary school graduates who do not enrol into higher education. The simple di¤erences in Table 3 suggest that higher education expansion led to a smaller decline in enrolment rates in treated regions. This decline re‡ects the lower demand for higher education, which had started since 1994.17 As for labour market outcomes, higher education expansion led to decrease the fraction of those who were inactive, decrease employment and wages, and increase unemployment among unskilled young adults. However, none of the di¤erence is statistically signi…cant. Changes in individual characteristics, background variables and labour market conditions between 1995 and 1998 could o¤set the e¤ect of the policy on the other outcomes of interest. In the regression analysis, by controlling for these other sources of variations, the e¤ect of the expansion can be assessed more precisely. The di¤erence in di¤erences between treated and control groups can be interpreted as the causal e¤ect of the policy, under the assumption that in the absence of the higher education expansion, the trend of the variables of interest would have not been systematically di¤erent between control and treated regions. To provide evidence in favour of the parallel trend hypothesis, Figure 2 shows the trend of the unemployment rate of youngsters in the 15-24 age brackets, by treated and control regions over the period 1985-1995, while Figure 3 and 4 show the trend of the employment and inactivity rate for youngsters in the same age group. Regional level data are available only from 1993 for the inactivity rate. Figure 5 shows the time trend of the percentage of young adults in the 19-24 age group enrolled in university by control and treatment regions.18 Apart from the unemployment trends, which seem slightly converging over time, the other outcomes show parallel trends, supporting the parallel trend hypothesis. Although di¤erences in pre-determined factors at the regional level are picked up by regional …xed e¤ects, one may worry about the determinants of the higher education expansion being systematically related to underlying trends in unskilled individuals’ labour market outcomes. As always in policy evaluation using non-experimental data, one cannot completely guard against such omitted variables bias. The Italian political situation at the beginning of the process of expansion and the assessment made by the Ministry of Education provide arguments in favour of the hypothesis that the policy was somewhat randomized. Yet, to increase the con…dence in this identi…cation strategy, a placebo test is performed. As no survey data are collected before 1995, I use data collected after the reform in 2001, when the expansion of higher education was almost completed. The placebo test pretends that the higher education expansion took place in the post reform period. Signi…cant e¤ects in the placebo test would suggest that the estimated e¤ects of the expansion re‡ect di¤erential time trends, rather than true policy impacts. 17

Ministry of Education and Research (1998) "The evolution of the demand for Higher Education: students, graduates, and equivalent students", doc. 4/98. 18 The trends in the …gure look constant and not decreasing because 19-24 year-olds are used as reference group. When only the percentage of 19 year-olds enrolled is considered, the trends are declining for both treated and control regions.

9

Individuals who delay university enrolment could be mistakenly considered as untreated, despite they took advantage of the program; in that case, estimates on enrolment would su¤er a downward bias. Similarly, an erroneous value of the program could be assigned to those who graduated in 1998 and delayed enrolment, since higher education expansion had continued after 1998. In that case, the e¤ect of the policy would be overestimated. To avoid these problems, individuals who entered higher education years other than 1995 and 1998 are dropped from the estimation sample. The estimation sample is further reduced by dropping the observations of employed individuals who started their job while enrolled in secondary school, since their post-graduation experiences might not be comparable with those of the rest of the sample. The research design could be invalidated in presence of lack of compliance at the regional level. This may occur if after assignment, some regions assigned to treatment do not institute the university or all the faculties forecast. The realization of the development plans that included changes in the provision of universities’supply occurred after 1990, when funds were assigned. A speci…c commission was deputed to report to the Ministry of Education about the realization of the development plans. The produced documents show that the actual plan realized in 1998 coincides with the content of the development plans, ruling out lack of compliance at the regional level.19 As noticed by Heckman et al. (1998) and Angrist (1995), policies that reduce the monetary cost of education might, as an indirect e¤ect in the long run, change the equilibrium in the market of skills. If so, the policy might a¤ect all workers’ labour market outcomes regardless of their region of residence. However, the limited mobility of labour in Italy limits concerns one may have on results being biased by this spillover e¤ect. According to the ISTAT (ISTAT, 2003), in Italy more than 80 percent of individuals with live parents reside in the same municipality as their mothers or fathers, about 7 percent of the people live in a municipality within 16 km from their parents, and only 8.2 percent of the citizens reside abroad or at a distance greater than 50 km. Another point concerns the set of time varying regional controls to be included: the variation of the regional unemployment rate and of the number of students who successfully …nished secondary school between 1995 and 1998. The …rst variable controls for possible changes in labour market opportunities that might be correlated with educational choices and labour market outcomes, whilst the second for variations of the total potential supply of secondary school graduates, including university enrollees. Estimates rely on the identi…cation assumption that there is no omitted time-varying and regional speci…c e¤ect that might be correlated with the program. This assumption will be violated if the allocation of other programs was correlated with the establishment of new campuses. Along with the new campuses’set up, the Legislator spurred the expansion of existing universities by allowing the institution of new departments. Identi…cation is achieved by controlling for this second source of expansion, which may be a substitute for the one under analysis. A concern is that while treatment regions expanded higher education on the external margin (by building more campuses), control 19

See note 3.

10

regions may have expanded the internal margin by increasing the number of departments, and therefore the number of available seats. Data show that over the period 1995-98 control regions increased the number of departments by 5.7%, while treatment regions by 22%. This small increase in departments in control regions, together with the demographic decline of the number of eligible students, suggests that control regions should not have experienced an increase in university enrolment during the period under consideration. Moreover, the assumption will be violated if the allocation of other governmental programs was correlated with the allocation of new campus. In 1997, a series of legislative initiatives named after the Labour Minister (“Pacchetto Treu”) introduced interim workers in the Italian institutional setting. The instrument was aimed at offering ‡exible labour supply in the aftermath of temporary shocks. The reform may a¤ect the labour market outcomes of individuals in the sample. However, the reform was implemented nationwide, and should be controlled for by the cohort …xed e¤ect. To account for the possibility that the new atypical contracts spread across regions in a way correlated with the expansion of the higher education supply, the speci…cations control also for the fraction of new ‡exible contracts signed at the regional level.20

4

Empirical framework

The e¤ect of the expansion of higher education on labour market outcomes of unskilled individuals is …rstly examined through a multinomial logit model, which can be regarded as an approximation to a reduced from of a four-state structural model for being enrolled, inactive, employed, or unemployed. The chosen speci…cation, following Flinn and Heckman (1983), distinguishes the status of unemployed from that of inactivity, as the distinction appears empirically relevant for young adults below the age of 25 (Tano, 1991). The reduced form model captures changes in one outcome relative to the base outcome. k stand for the utility of alternative k to individual i resident in region j at Let Vijt the end of secondary school in period t. The probability that the individual will choose alternative k is given by: exp(ajk + tk + k Pj Tit + k Xi + k Rjt ) h Vijt ; 8k = 1; ::H) = XH exp(ajh + th + h Pj Tit + h Xi + h Rjt ) h=1 (1) where aj is the region of secondary school …xed e¤ect, t a cohort of graduation …xed effect, and Pj denotes the intensity of the program in the region of residence. The average treatment intensity corresponds to the number of new university sites for every 1,000 k k Pijt = Pr(Vijt

20

Unfortunately, there are no o¢ cial statistics on the number of such contracts at a local level. Administrative sources estimated more than 2 million contracts in 2001. The fraction of atypical contracts at the regional level is computed using a survey item which asks the type of contract a worker has signed. The variable, taking value of one if an atypical job is held, is averaged out at the regional level.

11

college graduates. The average value of Pj in treated regions 0:073. Tit is a "treatment dummy" which takes the value of one for individuals belonging to the 1998 cohort and zero otherwise; the coe¢ cient k measures the e¤ect of the treatment on the treated conditional on not being enrolled in university. Xi is a vector of individual variables related to family background and past schooling career. Rjt is a vector of regional speci…c time varying controls which include: the change in the regional unemployment rate from 1995 to 1998, the number of students who successfully …nished secondary school, the number of departments instituted, and the fraction of ‡exible contracts signed at the regional level. The coe¢ cients for outcome H;unemployment, are set to 0 in the estimation. Once the increase in enrolment is accounted for by an equivalent change in the supply of workforce or inactive youngsters, it is possible to explore whether the policy exerts price e¤ects on employment, unemployment probability, and wages. Changes in the supply of unskilled individuals may produce price e¤ects on the returns of secondary education, whose sign di¤ers according to the theory at work. Under the signalling hypothesis, higher participation into tertiary education may imply a lower return to secondary education. As more campuses are provided, the movement of marginally more talented individuals from the group of unskilled secondary school graduates into that of university enrollees reduces the unskilled individuals’average ability, which is re‡ected in a lower price associated with their educational level.21 On the contrary, under the human capital hypothesis, there are no penalties that translate into lower wage and employment or higher unemployment as educational choices of unskilled secondary school graduates do not change. This holds under the assumption that workers with di¤erent educational levels are perfect substitutes for each other. If they are imperfect substitutes in production, one would expect a decrease in the supply of unskilled individuals to drive up (down) their wage and the employment (unemployment) probability. A linear-probability equation for each outcome is also estimated by least squares. To explore if individuals are nonrandomly selected out of higher education, labour market outcomes are also estimated with the Heckman selection model, which is speci…ed as follows: Pr(Sijt = 0) = (aj + t + Pj Tit + Xi + Rjt + Zjt ) E(Yijt jSijt = 0) = aj + t + Pj Tit + Xi + Rjt + ("s ) 21

(2) (3)

The underlying assumption is that imperfect capital markets may prevent some students from entering university if they do not live near a university and their parents lack the …nancial resources to board them in another city. In practice, some fraction of the population, unconstrained, will choose to attend university even if distant universities are the only option, whereas another proportion of the population, constrained, will choose to attend only if a university exists in their local area. When a new university is set up, those who enrol are individuals marginally more talented, previously constrained. Di¤erently from the perfect capital market hypothesis, this assumption makes theoretically possible that marginal entrants will raise the average ability of university graduates. However, both hypotheses imply that the average ability of secondary school graduates will decline.

12

where equation 2 is the participation equation out of higher education, whilst equation 3 is the outcome equation which includes the non-linear term that accounts for self selection in the previous stage. indicates whether the sorting process between educational levels is characterized by self-selection. The Heckman model is estimated with the two step procedure. Zjt is the variable excluded from the main equation to identify the parameters of interest without excessive reliance on functional forms. The chosen variable should a¤ect the choice of not enrolling in university without directly in‡uencing the individual’s labour market outcomes. Excluded variables usually used in the literature are indicators of the higher education local supply, capturing the fact that students who grew up in an area without a university or their preferred degree face higher costs of education (Card, 1995 and Di Pietro and Cutillo, 2008). In this framework, I use the change over time of academic faculties as a proxy that re‡ects the change in variety of academic alternatives provided to potential students.22 The lower is the variation of locally supplied faculties, the higher is the probability that an individual will not …nd a course tailored to her abilities and interests, and that she will not enrol in university. In turn, faculties’variability at the time the enrolment decision was made is not expected to a¤ect unskilled employment, unemployment and wage. The change in number of faculties is poorly correlated with the change in number of campuses (0.08) and in number of degrees courses (0.23), which are controlled for in the outcome equation. Therefore, this variable represents a good exclusionary restriction. Finally, "ijt is a zero-mean stochastic error term, clustered at the region level to account for correlation of errors within region. In the next section, …rst the e¤ect of the policy on enrolment is estimated using least squares to show evidence of the magnitude of the out‡ow from unskilled to higher education. Next, reduced form multinomial logit results are presented for the labour market outcomes, and …nally least squares and Heckman estimates for each outcome.

5

Results

5.1

University enrolment

The …rst set of results is presented in Table 4, showing linear probability estimates of the coe¢ cient of the treatment dummy on university enrolment. The estimation of the enrolment outcome replicates that in Author (2011) and shows the magnitude of the out‡ow from secondary to tertiary education.23 The dependent variable Yijt takes the value of one if the individual is enrolled in university and zero otherwise. The new campus dummy is normalized to account for heterogeneous cohort size at regional level, by dividing it for the number of individuals who obtained a secondary degree in that region in 1998 (in thousands). The baseline speci…cation in column 1 includes controls for regional …xed e¤ects, cohort of graduates’dummy, and program intensity interacted 22

Faculties are intended in this context as the number of divisions in a given university. The sample used di¤ers because in Oppedisano (2011) missing values were imputed, whilst in this paper only observations for which all controls are not missing are used in the estimation sample. Results are quantitatively similar between the two procedures. 23

13

with the treatment dummy. The coe¢ cient of the treatment variable turns out to be positive and statistically signi…cant at the 1 percent level in the …rst two columns. Controlling for changes in regional variables and other government programs does not change estimates, indicating that the bias for omitted programs may be very low. The estimated coe¢ cient in the speci…cation with the full set of controls indicates that the likelihood of entering university increased by nearly 12.5 percentage points for a new university instituted every 1,000 secondary school graduates.24 Adjusting the e¤ect of the policy for the intensity of the program in treated regions (0.073) and the average enrolment rate in 1995 (0.50) yields a 2 percent increase in enrolment in treated regions. The third column of the table shows the interactions between the reform and personal characteristics. The speci…cation chosen is the one which includes the whole set of regional time varying controls. Results show that the expansion mainly bene…ts middle ability individuals whose parents have secondary and college education: there is a positive and signi…cant e¤ect of the program on enrolment of students awarded with a B and a C mark at the exit of lower secondary school; students whose fathers have college and secondary education increase their probability to enter higher education by 11.5 and 6.9 percentage respectively.

5.2

Multinomial model

Reduced form results for the labour market outcomes of unskilled are presented in Table 5, showing multinomial logit estimates of the treatment coe¢ cient on enrolment, inactivity and employment, using unemployment as base outcome. In the …rst two columns, the coe¢ cient of the treatment variable is positive negative and statistically signi…cant for enrolment, while it is negative and statistically signi…cant at 5 percent level for the inactivity outcome. It indicates that given a one unit increase in the number of campuses, the risk of being enrolled relative to unemployed would be 0.578 times more likely, while the risk of being inactive relative to unemployed would be 0.950 times less likely. The coe¢ cient for employment, although negative, is not signi…cant. These results indicate that the expansion of higher education reduces employment, unemployment, and inactivity relative to enrolment, with the proportional decline in inactivity being larger than the decline in employment or unemployment. The last two columns of the table show the interactions between the reform and personal characteristics. Results indicate that the expansion reduces by a factor of 0.637 the relative risk of being enrolled for individuals with a B mark relative to those with an A mark. It reduces by a factor of 4.678 the risk of being inactive for children of parents with secondary education. The risk of being employed declines by 0.817 for individuals with a B mark at the exit of lower secondary school and by 1.594 for 24

The e¤ect of the policy on enrolment is 9-8 percent when treatment is assigned according to the province of residence (Author, 2011), whilst 12.5 percent when it is assigned according to the region. The higher value displayed with treatment area de…ned at a higher geographical level is due to possible externalities associated with the policy, thereby accounting for social interactions as well (see Miguel and Kremer, 2004).

14

children of college educated parents.

5.3

Probability of being inactive

The outcome of interest in this section is the probability of being inactive, conditional on not being enrolled in university. The dependent variable Yijt takes the value of one if the individual is inactive and zero otherwise. In Table 6, the …rst two columns present least squares estimates of the treatment variable, and the treatment variable interacted with individual characteristics. The third and fourth speci…cations present Heckman estimates without and with the interactions with individual characteristics. The coe¢ cient of the treatment variable turns out to be negative and statistically signi…cant at the 1 percent level in the speci…cations without interactions in columns 1 and 3. The estimated coe¢ cient in column 1 indicates that the probability of being inactive decreased by 4.9 percentage points for a new university instituted every 1,000 secondary school graduates. The estimated negative e¤ect could be biased due to the omitted selection term, which appears positive and signi…cant in the Heckman model. Indeed, the estimated coe¢ cient of the inactivity outcome in the Heckman model yields an overall e¤ect of the treatment variable of 8.3 percentage points, which is almost double the e¤ect of the policy without the self-selection correction. The magnitude of the coe¢ cient of the treatment dummy indicates a large e¤ect on the probability of being inactive. Adjusting the coe¢ cient for the average new campus intensity in treated regions (0.073), the average inactivity rate in 1995 (0.067), and the fact that the coe¢ cient is applies to only the half of the population not enrolled in university, yields an e¤ect of the policy of 4.5 percent on inactivity in treated regions. This result shows that the policy signi…cantly displaces individuals from …nancially unproductive activities. This result is consistent with the …ndings in Dearden et al. (2008), who evaluate the impact of a conditional cash transfer program implemented in the UK to encourage 16 to 18 year-olds staying in full time education. The authors …nd that the monetary subsidy displaces mainly individuals from unpaid activities into education. The coe¢ cient is positive and signi…cant. This means that secondary school graduates’probability of being inactive is higher than the probability that would have been observed for the average member of this sample had she chosen to enrol in university. The result is intuitive for the inactivity outcome: among those who choose not to enter the labour force, those who do not enrol in university are less unobservable talented than those who choose to enter higher education. In the selection equation the excluded variable is signi…cant in a¤ecting the non enrolment choice at the 1 percent level. Change in the number of faculties has a signi…cant negative impact on the probability of not enrolling in university, consistent with the intuition that the higher supply of faculties implies a higher probability of …nding a course tailored to personal interests. The null hypothesis of independent equations is rejected in the …rst speci…cation, but it is not when individual controls and time varying variables are included. 15

The interaction analysis helps in further assessing the average ability composition of individuals attracted to tertiary education because of a reduction in its monetary cost. Columns 2 and 5 of Table 6 show that those who reduced the probability of being inactive are youngsters whose fathers have college and secondary education and, to a lower extent, primary. They present similar characteristics to those who entered higher education as an e¤ect of the policy, providing additional evidence in favour of the hypothesis that the out‡ow from secondary to tertiary education originates from the pool of inactive youngsters. As shown in Author (2011), new enrollees face a higher probability of being retained in the university system, although their academic performance worsens. Ultimately, their labour market outcomes at the end of university should reveal the e¤ectiveness of the policy, but they are not available in these data. In the attempt to understand which individuals who were inactive entered higher education, I consider how the reasons why people decided not to look for a job were a¤ected by the policy. The reasons advanced for inactivity include: studying, lack of interesting jobs, waiting for a public competition, engagement in a stage, working in the family …rm, personal reasons, military service, and other motivations. The reasons signi…cantly and negatively a¤ected by the policy are the following: working in the family …rm, waiting for a public competition, and personal reasons, which corroborates the idea that a new campus nearby induces mainly those with a low opportunity cost to enrol in university.

5.4

Employment and unemployment probability

If the consistent out‡ow of youngsters from secondary school into tertiary education draws upon a similar population as inactivity draws upon, the direct e¤ect of the policy on employment, unemployment, and wages may be limited. Table 7 shows the e¤ects of the expansion on employment probability estimated using least squares and the Heckman model. The dependent variable Yijt takes the value of one if the individual is employed, zero otherwise. Results show that the treatment variable is close to zero and not signi…cant in the two speci…cations without the interaction terms. When controlling for self-selection out of higher education, lambda turns out to be negative but not signi…cant, indicating that there is not substantial di¤erence in unobservable ability between employed out of higher education and those enrolled. The interaction term analysis in columns 2 and 5 indicates that students marked B at the exit of lower secondary school face a lower probability of being employed, but the coe¢ cient is not signi…cant in the Heckman model. Results for unemployment (not reported) are very similar: there is no signi…cant e¤ect of the policy on the probability of being unemployed, nor any particular pattern of self-selection for those who chose not to enrol in university. The interaction term analysis does not indicate speci…c e¤ects due to personal characteristics.

16

5.5

Hourly wage

Table 8 shows the e¤ects of the expansion on wages estimated using least squares on the sample of not enrolled and the Heckman model on the whole sample. The dependent variable Yijt is the logarithm of hourly wage. Results show that the treatment variable is positive and non statistically signi…cant in all speci…cations, consistent with results from the employment and unemployment regressions. Lambda is negative and not signi…cant, suggesting that the unobservable wage component is not statistically di¤erent between employed individuals who are enrolled in university and those who are not. Considering selectivity, the excluded variable is signi…cant in a¤ecting the non enrolment choice at a 1 percent level. The interaction term analysis indicates that none of the coe¢ cients is statistically signi…cant. This result, together with those on employment and unemployment, shows that the expansion of higher education supply did not a¤ect the labour market outcomes of young adults in the workforce, lending support for the human capital theory. Indeed, as the educational choices of those who do not enrol in university do not change, no penalty translates on unskilled labour market outcomes. However, as the new in‡ow of enrollees mostly originates from the pool of inactive individuals, the policy may not provide enough variation among unskilled in the workforce to assess which theory is at work.

6

Robustness checks

This section reports results from a few speci…cation tests. The …rst test concerns the time trend hypothesis. The Di¤-in-Di¤ approach identi…es the e¤ect of the expansion from the common time trend assumption in the treatment and control group in the absence of the reform. A concern is that the e¤ects on unskilled individuals’labour market outcomes may re‡ect di¤erential time trends in the outcome of interest between treated and control regions, rather than a true policy impact. When examining this graphically (Figure 2, 3 and 4), it emerges that the pre-expansion trends of the outcomes look parallel for the treatment and the comparison group. To provide further evidence on this point, a placebo test is implemented. The only data available to perform this test is the 2001 "Survey on School and Work Experiences of Secondary School Graduates", as no data were collected before 1995. Speci…cally, the same regressions are estimated using the cohort of secondary school individuals graduated in 1998 and 2001.25 The estimates control for the little expansion of higher education supply that occurred between 1998 and 2001 and the interaction between the treatment status and the intensity of the program occurred between 1995 and 1998.26 Di¤erential secular time trends in 25

Since 2001 universities had to comply with the Bologna process, which introduced the 3+2 reform of the higher education system. This does not undermine the falsi…cation strategy employed as the reform a¤ected universities in all regions. Its e¤ect is thus captured by the time dummy …xed e¤ect. 26 New campuses were instituted in Piemonte, Liguria, Emilia Romagna, and Campania over the period 1998-2001.

17

treatment and cohort regions should cause this e¤ect to be signi…cantly di¤erent from zero. The results in Table 9 show that the treatment e¤ect is not signi…cant for any of the considered outcomes. To consider further whether results are driven by time varying regional speci…c e¤ects related to the evolution of the labour market, the baseline regressions were modi…ed to include controls for the change in the fraction of individuals employed in the di¤erent sectors of the economic activity (agriculture, industry, services) from 1995 to 1998. These controls do not a¤ect the magnitude and the signi…cance of the coe¢ cients of interest in the labour market outcomes.27 Binary outcomes are also estimated using a probit model. The estimate is slightly higher for the enrolment probability as the marginal e¤ect of the expansion is 17 percentage points, whilst it is the same for the probability of being inactive. Unemployment and employment probabilities show non signi…cant treatment coe¢ cients. To make sure that results are not driven by secular changes between Northern developed regions and less developed Southern ones, the interaction between the dummy for Northern regions is included among the controls. Northern regions include Piemonte, Valle D’Aosta, Lombardia, Trentino-Alto Adige, Veneto, Friuli-Venezia Giulia, Liguria and Emilia-Romagna. The interaction does not show any di¤erential pattern in Northern regions as for the enrolment, inactivity, and employment outcomes. However, in the unemployment equation, the interaction between the treatment and the Northern dummies indicates that relative to Southern and Central regions, Northern ones subject to the policy experience a reduction in unemployment, compensated by a parallel increase in employment. The e¤ect is signi…cant at 10% level in the unemployment equation, while it is not signi…cant in the employment equation. The result is consistent with the human capital hypothesis when skilled and unskilled workers are imperfect substitutes. In Southern regions, these e¤ects are not signi…cant. The di¤erential pattern between Northern and Southern regions re‡ects economic disparities between these two areas of Italy. The reduced supply of unskilled labor drives up employment in more developed and competitive Northern regions, while it does not exert any e¤ects in regions characterized by a lack of job opportunities, as seen for young people in the Southern regions. Location decisions based on unobservable characteristics may bias the estimates. Families may sort themselves nonrandomly into areas where universities are provided, leaving estimates on labour market outcomes likely to be biased. To address this issue, treatment and control are assigned on the basis of the individuals’region of residence at the end of secondary school. Coe¢ cients estimates may be downward biased because of measurement errors introduced by migration. Estimates that exclude individuals who relocated to …nd a job (around 3 percent of the sample in both surveys) are not statistically di¤erent from estimates including individuals who changed location. Overall, the e¤ects of the expansion of higher education on the labour market out27

The remaining results in this section are not reported here, but are available from the author upon request.

18

comes of unskilled secondary school graduates show fairly similar estimates across the alternative speci…cations.

7

Conclusion

A program that expands the supply of higher education increases enrolment in universities for more groups in the society and their labour market outcomes once they achieve graduation. However, the evaluation of such a program has to take into account also the e¤ects of the policy on the labour market prospects of secondary school graduates who do not bene…t from the expansion. The out‡ow of youngsters from secondary school to university a¤ects both the supply and the composition of unskilled secondary school graduates, depending on the observable and unobservable characteristics of those shifting from secondary to tertiary education. The new intake of enrollees may originate from the workforce or from the population of inactive youngsters, with di¤erent e¤ects in terms of foregone earnings. If the intake mainly originates from the workforce, the labour market outcomes of unskilled may be positively a¤ected if they re‡ect the scarcity of unskilled individuals in the labour market, or negatively, if they re‡ect the lower signalling value associated with having a secondary degree. Using the expansion of the higher education supply that occurred in Italy over the period 1995-1998, the paper analyses the e¤ects of the program on unskilled individuals’labour market outcomes by means of a di¤erence in di¤erences estimation strategy. Results indicate that the policy signi…cantly increases university enrolment in regions where the supply of higher education expanded, thereby reducing the supply of secondary school graduates. The policy displaces individuals from the workforce and from unproductive activities, with the e¤ect on the latter being signi…cant. As the foregone earnings for these individuals are zero, this result suggests that the implicit …nancial cost of the policy is lower. This e¤ect, although consistent with that found by Dearden et al. 2008 in the UK, needs to be read in the context of an expansion of the higher education supply that took place while the demand for higher education declining, partially of the lower number of young adults in the university age group. This …nding indicates an important potential instrument for policy makers concerned with the increasing number of European youngsters neither employed, nor in education, although a more precise assessment of this policy requires the evaluation of the labour market outcomes of individuals who bene…ted from the policy, which is left for future research. When looking at the characteristics of those shifting across educational level, it appears that they have middle ability but favourable parental background, and would have worked in the family …rm, or waited for a public competition had the expansion not took place. This provides additional evidence that a new campus nearby induces mainly those with a low opportunity cost to enrol in university. Lack of signi…cant e¤ects on the labour market outcomes of the workforce suggests that the human capital hypothesis is at work. The signi…cant reduction in unemployment in Northern regions goes in the direction of corroborating this result. However, in the context of a policy that draws

19

the new in‡ow of enrollees mostly from the pool of inactive, the policy may induce too little variation in the workforce to distinguish between the human capital and signalling theory.

20

I acknowledge helpful comments by the associate editor, McKinley Blackburn, and two anonymous referees, by seminar participants at UCL, RES Conference, the Petralia Workshop in Applied Economics and the International Workshop on Applied Economics of Education. I acknowledge the generous support of the Marie Curie Excellence Grant FP-6 51706 "Youth Inequalities" while based at the Geary Institute and the Marie Curie Intra European Fellowship for Career Development 251668 while based at UCL.

21

References Albert, J., van Ours, J. (2006) Using Employer Hiring Behavior to Test the Educational Signaling Hypothesis. Scandinavian Journal of Economics, 108(3), pp. 361-372. Angrist, J. (1995). Economic returns to schooling in the West Bank and Gaza strip. American Economic Review, 85(5), 1065-1087. Bedard, K. (2001). Human capital versus signalling models: university access and high school dropouts. Journal of Political Economy, 109(4), 749-775. Bosio, G., Leonardi, M. (2010). The impact of Bologna process on graduate labor market: demand and supply. Giornale degli Economisti, 69(3), 29-66. Brown, S., Sessions, G. (1999). Education and employment status: a test of the strong screening hypothesis in Italy. Economics of Education Review, 18(4), 397-404. Brown, S., Sessions, G. (2006). Evidence on the relationship between …rm-based screening and the returns to education. Economics of Education Review, 25(5), 498509 Card, D. (1995. Using Geographic Variation in College Proximity to Estimate the Return to Schooling. NBER Working Paper n. 4483. Carneiro, P., Lee, S. (2011). Trends in quality-adjusted skill premia in the United States, 1960-2000. American Economic Review, 106(1), 2309-49. Chevalier, A., Harmon, C., Walker, I., Zhu, Y. (2004). Does education raise productivity or just re‡ect it? Economic Journal, 144(499), 499-517. Chevalier, A., Lindley, J. (2009). Overeducation and the skills of UK graduates. Journal of Royal Statistics Society: Series A (Statistics in Society), 172(2), 307–337. Dearden, L., Emmerson, C., Frayne, C., Meghir, C. (2008). Conditional cash transfers and school dropout rates. Journal of Human Resources, 44(4), 827-857. Di Pietro, G., Cutillo, A. (2008). Degree ‡exibility and university drop-out: The Italian experience. Economics of Education Review, 27(5), 546-555. Flinn, C. J., and Heckman, J. J. (1983). Are Unemployment and out of the Labor Force Behaviorally Distinct Labor Force States? Journal of Labor Economics, 1, 28-42. Hämäläinen, U., Uusitalo, R. (2008). Signalling or human capital: evidence from the Finnish polytechnic school reform. Scandinavian Journal of Economics, 119(4), 755-775. Heckman, J., Lochner, L., Taber, C. (1998). General-equilibrium treatments e¤ects: a study of tuition policy. American Economic Review, 88(2). 381-386. Hussey, A. (2012). Human capital augmentation versus the signaling value of MBA education. Economics of Education Review, 31(4), 442-451. ISTAT. Parentela e reti di solidarieta’. Indagine multiscopo sulle famiglie. Rome 2003. Miguel, E., Kremer, M. (2004). Worms: identifying impacts on education and health in the presence of treatment externalities. Econometrica, 72(1), 159–217. Miller, P. W., Mulvey, C., Martin, N. (2004). A test of the sorting model of education in Australia. Economics of Education Review, 23(5), 473-482 Oppedisano, V. (2011). The (adverse) e¤ects of expanding higher education: evidence from Italy. Economics of Education Review, 30(5), 997-1008. 22

Tano, D. K. (1991). Are Unemployment and out of the Labor Force Behaviorally Distinct Labor Force States? Economics Letters, 36, 113-117 Walker, I., Zhu, Y. (2008). The college wage premium and the expansion of higher education in the UK. Scandinavian Journal of Economics, 110(4), 695–709. Wolpin, K. (1977). Education and screening. American Economic Review, 67(5), 949-958.

23

Figure 1: Changes in university supply between 1995-1998

Figure 2: Time series of the unemployment rate of 15-24 year olds, by control and treated regions before the policy. Source: ISTAT 24

Figure 3: Time series of the employment rate of 15-24 year olds, by control and treated regions before the policy. Source: ISTAT

Figure 4: Trend of the inactivity rate of 15-24 year olds, by control and treated regions before the policy. Source: ISTAT

25

Figure 5: Time series of the percentage of young adults in the 19-24 age group enrolled in university, by control and treated regions before the policy. Source: ISTAT

26

Comparison 11 regions Mean (s.d.)

Treatment 9 regions Mean (s.d.)

ComparisonTreatment Mean (s.e.)

0,542 (0,498) 22,597 (0,980) 0,013 (0,113) 0,012 (0,108) 0,223 (0,416) 0,169 (0,375) 0,764 (0,425) 0,819 (0,385) 0,071 (0,257) 0,154 (0,361) 0,325 (0,469) 0,450 (0,497) 0,034 (0,181) 0,527 (0,499) 0,329 (0,470) 0,110 (0,313) 0,065 (0,247) 0,672 (0,469) 0,263 (0,440) 1,569 (0,286) 3,690

0,540 (0,498) 22,464 (0,936) 0,019 (0,135) 0,013 (0,115) 0,226 (0,418) 0,183 (0,387) 0,756 (0,430) 0,803 (0,498) 0,069 (0,253) 0,150 (0,357) 0,336 (0,472) 0,445 (0,497) 0,041 (0,199) 0,557 (0,497) 0,298 (0,458) 0,104 (0,305) 0,070 (0,254) 0,692 (0,462) 0,239 (0,426) 1,596 (0,296) 4,566

0,002 (0,011) 0,134 (0,021)*** - 0,006 (0,003)** - 0,001 (0,005) - 0,002 (0,009) - 0,014 (0,009) 0,008 (0,009) 0,016 (0,009) 0,002 (0,006) 0,004 (0,008) - 0,011 (0,010) 0,005 (0,011) - 0,007 (0,004) - 0,029 (0,011)*** 0,031 (0,010)*** 0,006 (0,007) - 0,005 (0,005) - 0,020 (0,010) 0,024 (0,010)** - 0,026 (0,009)***

Sample of not enrolled Female Age Father with college degree Mother with college degree Father with secondary degree Mother with secondary degree Father with primary education or lower Mother with primary education or lower Junior school mark A Junior school mark B Junior school mark C Junior school mark D High school Technical secondary school Professional secondary school Other secondary school Inactivity status Employment status Unemployment status Lof of hourly wage Observations All sample College enrolment

0,502 0,493 0,008 (0,500) (0,500) (0,008) Observations 7,600 8,817 Note: Observations are weighted; *** signi…cant at 1 percent; ** signi…cant at 5 percent

Table 1: Baseline descriptive statistics 27(1995 survey) - Individual level data

Comparison 11 regions Mean (s.d.) 0,018 (0,016) 11,48 (3,275) 31,86 (9,430) 37,300 (17,008) 1,121 (0,764)

Gdp growth 1994-95 Unemployment rate Unemployment rate 15-24 Enrolment rate (on pop 19 years old) N. of campuses every 10,000 19 years old

Treatment 9 regions Mean (s.d.) 0,023 (0,018) 11,56 (6,262) 29,31 (13,500) 35,687 (14,160) 0,775 (0,763)

ComparisonTreatment Mean (s.e.) - 0,005 (0,007) - 0,080 (2,235) 2,550 (5,208) - 1,614 (7,228) 0,346 (0,341)

Source: ISTAT

Table 2: Baseline descriptive statistics (1995 survey) - Regional level data

28

Treatment 1

0

Di¤erence

0,367 (0,006) 0,493 (0,005) -0,126 (0,021)

0,353 (0,005) 0,502 (0,006) -0,149 (0,017)

0,014 (0,007) -0,009 (0,008) 0,023 (0,025)

0,045 (0,002) 0,069 (0,004) -0,024 (0,006)

0,049 (0,003) 0,065 (0,004) -0,016 (0,009)

-0,004 (0,006) 0,004 (0,010) -0,008 (0,011)

0,790 (0,005) 0,692 (0,007) 0,098 (0,029)

0,805 (0,006) 0,672 (0,008) 0,133 (0,024)

-0,015 (0,070) 0,020 (0,076) -0,035 (0,037)

0,165 (0,004) 0,239 (0,006) -0,074 (0,028)

0,145 (0,005) 0,263 (0,007) -0,118 (0,020)

0,020 (0,065) -0,024 (0,081) 0,044 (0,034)

1,659 (0,005) 1,596 (0,006) 0,063 (0,022)

1,659 (0,005) 1,569 (0,007) 0,090 (0,012)

0,000 (0,032) 0,027 (0,034) -0,027 (0,025)

Enrolled (all sample) Cohort 1998 Cohort 1995 Di¤erence Inactive (not enrolled) Cohort 1998 Cohort 1995 Di¤erence Employed (not enrolled) Cohort 1998 Cohort in 1995 Di¤erence Unemployed (not enrolled) Cohort 1998 Cohort in 1995 Di¤erence Log of hourly wage (not enrolled) Cohort 1998 Cohort in 1995 Di¤erence

Note: Means and standard errors clustered at regional level in brackets.

Table 3: Means of outcomes of interest by cohort of graduation and treatment

29

(1) 0.125*** [0.012]

Treatment

(2) 0.125*** [0.012]

Junior school mark B*Treatment

(3)

0.089** [0.031] Junior school mark C*Treatment 0.098* [0.049] Junior school mark D*Treatment 0.078 [0.086] Father college degree*Treatment 0.115** [0.041] Father secondary educ*Treatment 0.069** [0.031] Father primary educ*Treatment 0.029 [0.064] Controls yes yes yes Observations 30,774 30,774 30,774 R-squared 0.365 0.365 0.365 Note: Linear Probability Model estimates for the whole sample. Robust standard errors in brackets; * signi…cant at 10 percent;** signi…cant at 5 percent;*** signi…cant at 1 percent. Standard errors clustered at region level. Regressions include year and region …xed e¤ects. Individual controls include: gender, parental education dummies, past scores dummies, type of sec. school dummies. Regional controls include: variation of regional unemployment between 1995-98, change in the number of sec. school graduates between 1995-98, variation in the number variation in the number of departments between 1995-98, variation in the fraction of ‡exible contracts signed at regional level between 1995-98.

Table 4: Probability of college enrolment

30

31

Inactivity (2) -0.950** [0.408]

Employment (3) -0.15 [0.261]

Enrolment (4)

Inactivity (5)

Employment (6)

Table 5: Multinomial logit estimates for enrolment, inactivity and employment relative to unemployment

-0.637* 3.219 -0.817** [0.342] [2.731] [0.359] Junior school mark C*Treatment 0.671 0.158 -0.139 [0.825] [4.366] [0.929] Junior school mark D*Treatment 0.471 2.618 0.14 [0.312] [2.937] [0.364] Father college degree*Treatment -0.69 -7.732 -1.594** [0.495] [5.322] [0.720] Father secondary educ.*Treatment 0.091 -4.678* -0.282 [0.193] [2.806] [0.262] Father primary educ.*Treatment 0.769 -2.535 0.529 [0.452] [2.752] [0.406] Observations 30,774 30,774 30,774 30,774 30,774 30,774 Note: Multinomial logit estimates for enrolment, inactivity and employment relative to unemployment. Robust standard errors in brackets;* signi…cant at 10 percent; **signi…cant at 5 percent;*** signi…cant at 1 percent. Standard errors clustered at region level. Regressions include year dummy, region …xed e¤ects and the whole set of controls. Individual controls include: gender, parental education dummies, past scores dummies, type of sec. school dummies. Regional controls include: variation of regional unemployment between 1995-98, change in the number of sec. school graduates between 1995-98, variation in the number of departments between 1995-98, variation in the fraction of ‡exible contracts signed at regional level between 1995-98.

Junior school mark B*Treatment

Treatment

Enrolment (1) 0.578*** [0.183]

32

0.192*** [0.062] 0.076* [0.040] 0.101** [0.045] -0.348** [0.138] -0.184*** [0.039] -0.128*** [0.039]

(2)

0.159*** [0.042]

Outcome Eq (3) -0.083*** [0.030]

-0.451*** [0.124] -0.792*** [0.276] 30,774

Sel Not Enrolm. (4)

0.198** [0.099] 0.084 [0.088] 0.111 [0.086] -0.385** [0.171] -0.228** [0.091] -0.171** [0.086] 0.159*** [0.042]

Outcome Eq (5)

-0.451*** [0.124] -0.792*** [0.276] 30,774

Sel Not Enrolm. (6)

Table 6: Probability of being inactive

Observations 17,511 17,511 30,774 30,774 R-squared 0.024 0.025 Note: Linear Probability estimates on the sample of not enrolled (1)-(2) and Heckman estimates on whole sample (3)-(6). Robust standard errors in brackets;* signi…cant at 10 percent; **signi…cant at 5 percent;*** signi…cant at 1 percent. Standard errors clustered at region level. Regressions include year dummy, region …xed e¤ects and the whole set of controls. Individual controls include: gender, parental education dummies, past scores dummies, type of sec. school dummies. Regional controls include: variation of regional unemployment between 1995-98, change in the number of sec. school graduates between 1995-98, variation in the number of departments between 1995-98, variation in the fraction of ‡exible contracts signed at regional level between 1995-98.

Change n. of faculties

Treatment

Lambda

Father primary educ.*Treatment

Father secondary educ.*Treatment

Father college degree*Treatment

Junior school markD*Treatment

Junior school mark C*Treatment

Junior school mark B*Treatment

Treatment

(1) -0.049*** [0.014]

33

-0.132* [0.071] -0.093 [0.154] -0.055 [0.098] 0.14 [0.141] 0.105 [0.077] 0.06 [0.097]

(2)

-0.021 [0.078]

Outcome Eq (3) 0.003 [0.052]

-0.451*** [0.124] -0.792*** [0.276] 30,774

Sel Not Enrolm. (4)

-0.133 [0.192] -0.094 [0.170] -0.057 [0.167] 0.145 [0.330] 0.111 [0.174] 0.065 [0.165] -0.021 [0.078]

Outcome Eq (5)

-0.451*** [0.124] -0.792*** [0.276] 30,774

Sel Not Enrolm. (6)

Table 7: Probability of employment

Observations 17,511 17,511 30,774 30,774 R-squared 0.148 0.148 Note: Linear Probability estimates on the sample of not enrolled (1)-(2) and Heckman estimates on whole sample (3)-(6). Robust standard errors in brackets;* signi…cant at 10 percent; **signi…cant at 5 percent;*** signi…cant at 1 percent. Standard errors clustered at region level. Regressions include year dummy, region …xed e¤ects and the whole set of controls. Individual controls include: gender, parental education dummies past scores dummies, type of sec. school dummies. Regional controls include: variation of regional unemployment between 1995-98, change in the number of sec. school graduates between 1995-98, variation in the number of departments between 1995-98, variation in the fraction of ‡exible contracts signed at regional level between 1995-98.

Change n. of faculties

Treatment

Lambda

Father primary educ.*Treatment

Father secondary educ.*Treatment

Father college degree*Treatment

Junior school mark D*Treatment

Junior school mark C*Treatment

Junior school mark B*Treatment

Treatment

(1) -0.002 [0.042]

34

-0.009 [0.103] -0.11 [0.124] -0.045 [0.082] -0.712 [0.470] -0.011 [0.070] 0.109 [0.094]

(2)

-0.046 [0.056]

Outcome Eq (3) 0.029 [0.046]

-0.509*** [0.142] -0.807** [0.324] 23,798

Sel Not Enrolm. (4)

-0.008 [0.159] -0.109 [0.140] -0.044 [0.138] -0.743 [0.609] 0.003 [0.150] 0.124 [0.136] -0.053 [0.056]

Outcome Eq (5)

-0.509*** [0.142] -0.807** [0.324] 23,798

Sel Not Enrolm. (6)

Table 8: Log of hourly wage

Observations 10,535 10,535 23,798 23,798 R-squared 0.121 0.122 Note: Ordinary Least Squares estimates on the sample of not enrolled (1)-(2) and Heckman estimates on whole sample (3)-(6). Robust standard errors in brackets;* signi…cant at 10 percent; **signi…cant at 5 percent;*** signi…cant at 1 percent. Standard errors clustered at region level. Regressions include year dummy, region …xed e¤ects and the whole set of controls. Individual controls include: gender, parental education dummies, past scores dummies, type of sec. school dummies, experience, and squared experience. Regional controls include: variation of regional unemployment between 1995-98, change in the number of sec. school graduates between 1995-98, variation in the number of departments between 1995-98, variation in the fraction of ‡exible contracts signed at regional level between 1995-98.

Change n. of faculties

Treatment

Lambda

Father primary educ.*Treatment

Father secondary educ.*Treatment

Father college degree*Treatment

Junior school mark D*Treatment

Junior school mark C*Treatment

Junior school mark B*Treatment

Treatment

(1) 0.016 [0.039]

(1) (2) (3) (4) (5) Enrolment Inactivity Employment Unemployment Hourly wage Treatment 1995-98 0.076 -0.579 -0.076 0.124 0.4 [0.093] [0.344] [0.121] [0.117] [0.282] Controls yes yes yes yes yes Controls: Rjt yes yes yes yes yes Observations 32,693 18,442 18,442 18,442 12,346 R-squared 0.389 0.15 0.117 0.11 0.128 Note: Linear Probability Model Estimates (1)-(4) and Ordinary Least Squares (5). Robust standard errors in brackets. Standard errors are clustered at region level.

Table 9: Outputs estimated for the 1998 and 2001 cohort - Placebo test

35