University education in Italy

Transcription

1 University education in Italy Daniele Checchi (State University o Milano-Bicocca - Italy) File name: ijmp2.doc This version: 22/12/00 Abstract Notwithstanding the low level o tuition ees and absence o other access barriers, Italy is characterised by low educational achievement at university level. We investigate possible reasons or this apparent puzzle and propose a ormal model predicting that amilies invest more in their children the higher are (the expectations on) child s unobservable ability. Since amily income provides an incentive or better student perormance, richer parents internalise this eect by investing more resources in the education o their children. Our empirical analysis does not contradict this theoretical model. Using the Bank o Italy s representative sample o the Italian population (1995), we observe that amily income does not prevent enrolment at the university, whereas unobservable ability (proxied by cultural amily background) is more relevant, especially as it shapes the secondary education choices. Using administrative data on students enrolled in some aculties o the State University o Milan in , we show that students perormance is positively correlated with unobservable ability (proxied here by the marks obtained on leaving secondary school) and with amily income. We take this last evidence as supporting the idea o amily networking: students rom richer amilies tend to go quicker because they have better prospects when they leave university. JEL code: I22, J62 Keyword: Education inancing, human capital investment. Mailing address: Daniele Checchi Facoltà di Economia Università degli Studi di Milano Bicocca Polo Bicocca - U6-356 Piazza dell Ateneo Nuovo MILANO - Italy tel ax daniele.checchi@unimi.it Paper presented at the XIV AIEL conerence (Milan, 7-8 October, 1999). This paper is part o a larger project on educational choices and students perormance, conducted with Andrea Ichino (European University Institute) and Aldo Rustichini (Boston University) and Francesco Franzoni (MIT). The author thanks Giorgio Brunello and Massimo Giannini or comments on an earlier drat. Financial support rom Italian CNR (Progetto Strategico CNR on L'Italia in Europa: governance e politiche per lo sviluppo economico e sociale ) is grateully acknowledged. 1

2 1. Introduction When compared to other OECD countries, Italy is characterised by low investment in higher education. 1 While the ratio o tertiary graduates in university programmes to population at the typical age o graduation was 10.6% in Italy (1994 entire population), the corresponding igure was 12.6% or Germany, 13.7% or France, 27.0% or the United Kingdom and 31.8 or the United States. I we adopt a larger deinition o tertiary education (ranging rom non-university tertiary programmes to Ph.D. programmes), matters are even worse: 22.2% or Italy compares with 24.8% or Germany, 44.4% or France, 61.1% or the United Kingdom and 65.6% or the United States. 2 Various explanations can be put orward to explain these dierences. The most intuitive one is the dierent stage o development. I higher education can be considered as a consumption good providing social status (as in Fersham et alt. 1996), then variations in gross domestic product per capita could account or dierences in graduation rates. However, output dierences are not so pronounced and country ranking correlation according the two variables is extremely low. 3 Thinking o education as a vertically integrated process, we could suspect that educational achievement is already lower at secondary level. The ratio o upper secondary graduates to corresponding population was 76.2% in Italy (1994 entire population), and has to be contrasted with 73.6% or the United States, 88.5% or Germany and 80.8% or France. 4 Thus potential bottlenecks due to shortage o secondary school graduates cannot be invoked as a possible reason or low enrolment rates in Italian universities. Leaving aside any explanation based on access barriers due to high admission ees, because Italian public universities charge very low ees (the Italian average in 1996 was 600 euros), 5 then we wonder about the relative returns to higher education. When compared to other OECD countries, the relative earnings o an Italian college graduate are rather low. Taking the earnings o an upper secondary graduate as a reerence point (100), an Italian college graduate earns 141 i a man (112 i a woman). Corresponding igures are 187 (165) or France, 167 (162) or Germany, 164 (204) or the United Kingdom, and 168 (175) or the United States. Prima acie, low returns to education could represent a possible explanation o lower enrolment rates in Italy: even i attending a university course is relatively cheap, the expected uture gain alls short o current costs, and amilies do not send their ospring to colleges. Though easy, this explanation neglects idiosyncratic dierences among amilies. What are the salient characteristics o the population excluded rom university education? does it constitute a random sample o the entire population? Since the answer is obviously negative, we are interested in studying the determinants at individual level o the choice o attending and completing a university course in Italy. The present paper ocuses on these questions: in a context o cheap access to and low returns or university education, what does govern the educational choices o Italian amilies? We will show that human capital investment theory is not contradicted by our data: amilies plan the educational career o their children starting rom the choice o secondary school, but their investment in higher education is more likely to be successul the higher the child s (perceived) ability. The paper is organised as ollows. In section 2 we propose a standard model o human capital investment with liquidity constraints, that ormalises the idea that amilies invest in children s education whenever the 1 An account o the Italian educational system is in Brunello-Comi-Luciora Data rom OECD 1966 (Table R 12.1, p.181). 3 The Spearman rank correlation coeicient between tertiary education enrolment and output per capita (converted in US 1987 dollars) or OECD countries in 1995 is equal to 0.28 (22 observations). Italian output per capita (15392 US dollars) is lower than US (20716) and France (18068), but exceeds UK (13444) (source: World Bank tables). 4 Data rom OECD 1966 (Table R 11.1, p.175). 5 Table 1 in Brunello et alt A broader discussion o (higher) education inancing o Italy, contrasted with the US system, is in Checchi et alt

3 child s expected ability is above a speciied threshold. In section 3 we make use o a representative sample o the Italian population to estimate the determinants o sending a child to college. In section 4 we make use o another sample o administrative data on students rom a speciic university to estimate the determinants o their academic career. Section 5 concludes the paper. 2. A ormal model o educational investment Let us now introduce a ormal model to study the interaction between the investment in children s education by parents and school perormance by children. The model is intended to be applicable to the inal stages o education, where individual students bear some responsibility in their perormance (like going to college and obtaining a degree). We cast the choice in a single period ramework, but extensions to overlapping generation contexts are straightorward. Consider an altruistic parent that has to allocate part o her current income between consumption and investment in her child s education. Raising the investment reduces her own consumption but improves the income opportunity o her child. The parent choice solves the ollowing problem ( C, Y ) = max U ( Y S, Y ), U > 0, U < 0, i = C, Y ; U > 0, i j max U (1) S c S c where U is the parent utility, C is parent consumption, Y is her (amily) income, S is the amount invested to inance the child s education (i.e. payment o admission ees, purchase o textbook, etc.), Y c is the child s uture income. The child s income depends on academic perormance P achieved during university career and (possibly) on amily networking, here proxied by the eect o amily income: children rom richer amilies have access to better jobs, both in terms o earnings and quality o the job. 6 The academic perormance can be thought as the amount o human capital obtained during the university career, which includes both elements o quantity (how many courses attended) and quality (what marks obtained in passing the exams). 7 i ii c ij Y c ( P Y ), Y > 0, Y 0, Y < 0, i = P, Y ; Y > 0 i j = Y,, (2) P Y ii c ij The academic perormance depends on several actors. First, it depends on student ability A that is assumed to be perectly observable at the moment o undertaking the decision to proceed to tertiary education: other things held constant, abler students exhibit better perormance. 8 But perormance is also dependent on student eort E, which is optimally chosen by the student her-sel, taking into account the disutility o eort and the beneit o higher uture earnings. The presence o S captures the positive eect o the resources invested in education. In addition, i we want to take into account the social capital endowment (parents and neighbours educational achievements) that osters the process o learning, we could proxy this variable with amily income Y. 9 6 See or example Montgomery But it could also be interpreted as a signal o unobservable ability that irms take into account when setting their wage policy. 8 We are not interested in going deeper into deining what we mean by ability. We ind convincing the deinition provided by Rubinstein and Tsiddon 1998: ability is everything that contributes to the child s income potential, is in the child at the time he takes his education decision, and cannot be purchased on the market (p.19). Consequently, in their empirical analysis they proxy individual ability with their parents education. 9 In Benabou 1996 amily income drives locational choice, and as a consequence neighbourhood social composition and school inancing. 3

4 ( A, E, S ) P = P, Y (3) P > 0, P > 0, P > 0, P 0, P < 0, i = A, E, S, Y ; P > 0, i j A E S Y ii ij Each student maximises her own utility deined over eort E and uture income Y c subject to the constraint described by equation (3) E ( E, Y ) maxv ( E, Y ( P, Y ) maxv [ E, Y ( P( A, E, S, Y ) Y )] max V = =, (4) c E c V < 0, V > 0, V 0, V < 0, V < 0 E Y EE YY EY The sequence o choices in the present model resembles a Stackelberg game, where the parent acts as a leader, choosing the optimal amount o investment S * under the rational expectation o children behaviour in determining the optimal amount o eort E *. Notice that when the parent makes her choice, her income is taken as predetermined, so that she links her investment on the ability o the child only. 10 Once this choice has been undertaken (that is, when a university has been chosen), the child takes the parent investment as predetermined, and in her turn she bases the optimal amount o eort onto her endowment o ability. To study the sequence o optimal choices we have to start rom the end, and we move to the solution o the problem described by equation (4). Since we want to take into account the possibility o imperections in inancial markets, we could proceed in two ways: either we introduce a (possibly binding) constraint on the optimal amount o investment S *, or we consider that students rom poorer amilies are subject to stronger pressure to complete their curricula. In this case this constraint takes the ollowing orm E ( Y ), < 0 c P P P (5) where P represent the minimum level o perormance a student has to achieve in order to ulil parents expectations. Obviously, the richer the amily, the lighter this type o pressure. 11 I we solve problem (4) under the constraints represented by equation (3) and (5) we obtain Eˆ = Eˆ ~ E = P ( A, S, Y ) i P( A, Eˆ, S, Y ) P ( Y ) 1( P ( Y ) i P( A, Eˆ, S, Y ) < P ( Y ) The signs o the partial derivatives can be studied using the implicit unction theorem. For example, it is possible to show that (6) * E A > 0 i V P + V P EY A Y EA > PE V YY Y + VY Y PP P Thus in general it is impossible to sign these derivatives with certainty, unless choosing speciic unctional orms. Equation (6) can be re-expressed in a more compact orm as * ~ * [ Eˆ, E] = E ( A, S ) E = max, (7) Y 10 Owen and Veil 1997 obtain similar conclusions. 11 It is worth recalling that in Italy student grants (assegno di studio) and ee exemption are conditional on a similar requirement (a minimum number o passed exams per year and a minimum average mark obtained in passing them). 4

5 Given the act that students eort is unobservable, i we reinsert equation (7) into equation (3) we obtain a testable prediction * * * * ( A, E, S, Y ) = P ( A, S ) P = P, (8) Y Student perormance must be correlated with his/her ability, his/her amily income (encompassing the eects o possible liquidity constraints, amily networks and social capital) and the amount o resources invested in education. Equation (7) can be taken as the reaction unction o the ollower (the child), and is taken into ull account by the leader (the parent), when choosing the optimal amount o investment in education S * [ Y S, Y ( P ( A, S, Y ) Y )] maxu, (9) Her optimal choice is thereore given by ( ) S * * = S A, Y (10) Thus a parent inances a child s educational choice based on her perceived ability and present current (amily) income. But these results are too general to provide an intuitive idea o underlying relationships. In order to see how the present model works, let us consider speciic unctional orms. We express equation (3) as P = A α S σ E ε Y β (11) where β measures the intensity o neighbourhood eects in cultural education; setting β = 0 is equivalent to exclude this phenomenon. This ormulation considers individual ability, educational resources, individual eort and social capital (proxied by amily income) as partial substitutes. Equation (2) is re-expressed as θ α σ ε β+ θ c Y = πpy = πa S E Y (12) where θ measures the intensity o the amily networking eect. 12 Taking a linear utility unction or the child yields under the ollowing constraint α σ ε β+θ ( E, Y ) = max Y ηe = max πa S E Y ηe max V (13) E Solving the problem (13) yields the ollowing result c E c E δ ( Y ) = Y P P (14) E * = max [ ] ( ) 1 ~ πε α σ β+θ ε α σ δ β = ε * Eˆ, E max A S Y, A S Y = E η A, S, Y + + ± 1 1 (15) 12 Notice that under the exponential ormulation o equation (11) and (12), the possible impact o neighbourhood eects (during schooling age) is indistinguishable rom social networking (during working age). 5

6 By equating Ê and E ~ we obtain the amily income threshold constraint is no longer binding Yˆ above which the perormance 1 β+θε+δ( 1 ε ˆ ) = Yˆ σ, Y η = α πεa S A S (16) which can be represented in igure 1. When Y ˆ < Y the perormance constraint (14) dominates, and thereore the student relaxes and reduces her eort with rising amily income. As soon as the constraint stops biting (at Y ˆ = Y ), the networking eect introduces a positive correlation: a richer amily provides better prospects or uture employment, and thereore induces greater eort in its ospring. [insert igure 1 about here] The threshold Yˆ is lower the higher is the student s ability and the investment in her education, to be determined at the irst stage o the model. By replacing equation (15) in equation (11), we obtain an observable behaviour in student perormance A P = Y δ α S σ Y β+θε πε η ε 1 1 ε i i Y Y Yˆ Yˆ (17) Moving now to the parent choice, we make use o a Cobb-Douglas utility unction ( ) ( ) ω 1 C, Y = max Y S ( Y ) ω When the perormance constraint is binding ( Y ˆ ) maxu (18) S c S Y, we get c arg max whereas when it is not binding ( Y ˆ ) arg max S S Y ω δ θ 1 ω ~ ( Y S ) ( πy Y ) = S = 0 we obtain ε ε ω σ ( ) ( ω) Y S α σ β+θε πε π A S Y θ Y = Sˆ 1 = η ω ( 1 ε) ω + σ( 1 ω) Y = ΩY (19) (20) The optimal amount o investment in a child s education is thereore zero i the amily income is lower than a given threshold and/or the student ability is suiciently low; otherwise it is a constant 6

7 raction o amily income. 13 By taking into account that the income threshold is conditional upon the optimal amount o education, i we replace equation (20) into equation (16) we obtain 1 1 η β+θε+δ( ε) + σ Yˆ = α σ (21) πεa Ω This result tells us that as long as a student is suiciently endowed in terms o ability, her amily will always invest a raction o income in her education. Otherwise, they will not invest any amount because the student s perormance will be insuicient to overcome the perormance constraint. This situation is depicted in igure 2. Notice that the optimal amount o education obtained in equation (20) is independent o student ability. This is strictly due to the speciic technology assumed or student perormance; however i one takes into account that the income threshold is negatively related to student ability by means o equation (21), this inal result conirms the general ormulation indicated in equation (10). [insert igure 2 about here] Notice that when we replace the results o equation (19) into equation (17) we do not observe university perormance or students insuiciently endowed with ability (i.e. students that either perormed poorly at previous stage o education or with low educated parents), because their parents anticipate that they will be unable to over-step the threshold, and in consequently do not invest in them (i.e. they to not pay or their enrolment at the university). Thus our model oers two testable predictions: i) amilies invest more resources in the education o their children the higher is their income, provided that their children are suiciently endowed with ability ; ii) children perorm better at the university the higher their ability, the quality/quantity o resources invested in their education and their amily income (or cultural inheritance and/or networking reasons). Both predictions are not rejected by our empirical analysis, where we now move. 3. The college choice We can study the choice o higher education as a sequential process, characterised by at least our steps: i) the successul completion o a 5-year secondary school (preliminary condition) ii) the decision about whether to enrol at the university; iii) the choice o a university aculty; 14 iv) the subsequent perormance o a student once enrolled at the university. We do not have in Italy longitudinal iles that allow us to trace the school career o a representative cohort o youngsters (like the British National Child Development Survey). Thus we are orced to iner indirect inormation rom resorting to representative samples o the entire population. Among the ew publicly available, we have chosen the Italian population sample provided by the Survey on Household Incomes and Wealth (SHIW) conducted biannually by the Bank o Italy. Using the most recent wave available (reerring to 1995) we can trace out amilies with children aged between 19 and 13 This is an unavoidable consequence o assuming Cobb-Douglas utility unction in equation (18). 14 Once the choice o sending a child to college has been undertaken, and a desired aculty has been selected, there is an additional choice between a public and a private university (i available). Given the limited number o private universities in Italy, we abstract rom this eect. 7

8 26, among which some are enrolled at the university. Studying the dierences between the two subsamples, we are able to identiy some conditioning variables in the college choice. The 1995 survey contains inormation regarding individuals and 8135 amilies amilies have cohabiting children, whereas we do not have inormation about children that live elsewhere. Our analysis will be biased whenever the individuals that let cohabitation are signiicantly dierent rom remaining ones with respect to the process under analysis. However we know rom other sources that Italy is characterised by late leaving o amily cohabitation, 15 due to high costs o living and absence o unemployment beneits. Looking at our data, we can iner inormation by looking at amily composition. In a constant population with children not leaving amily cohabitation we should observe a constant number o children at any (average) age o children themselves. Under this assumption when we record a sharp decline, we can take it as evidence o the (average) amily leaving age. By observing igure 3, we notice that this sharp decline occurs in our data set around an average age o children o 29. Thereore we eel justiied in restricting our analysis to amilies with children in an age comprised between 19 and 26, extremes included, and we expect a reduced bias in the sample o these children. [insert igure 3 about here] We have 2022 amilies that include 2748 individuals aged 19-26, 907 o which are recorded as student and have completed a secondary school, and can plausibly be taken as university students; among the remaining 1841 youngsters, 1041 did not achieve a secondary school degree, and thereore have already dropped out o the educational system. Thus we are let with 1707 people that potentially could enrol at the university (they have the right educational credentials), but only a raction (corresponding to 53.1%) that actually did it. 16 In addition to demographic variables (age, gender, number o siblings, region o residence), we know the type o secondary school attained (but unortunately not the inal marks obtained at exit), and we also have some inormation on the amily background (education and occupation o both parents, amily income 17 ). Beore moving to direct estimates on representative samples, we want to examine the indirect aggregate evidence. Table 1 compares the gross amily incomes or the entire SHIW sample, or the subsample o amilies with children who are university student and or the administrative sample rom the University o Milan. Additional inormation is also reported on inal marks at exit o secondary schools. We notice that amilies with children at the university are on average richer by 10 millions o Italian liras (slightly above 5000 euros). In addition students enrolled in the State University o Milan (at least in the aculties under consideration) are indistinguishable rom the rest o the population, especially when looking at the inal marks at exit. I we take this latter variable as a proxy or unobservable ability, we are induced to assume that amily income is a more important determinant o the educational choice, while ability is not. A second aggregate indication is reported in Table 2, where we estimate the returns to dierent types o schools. Since students amilies orm expectations about uture earnings attached to a speciic degree by looking at actual returns in the labour market, looking at the table we notice that completing a generalist secondary school (liceo) without going on with enrolment at the university 15 In 1996 the 98.1% o young people aged was cohabiting with the amily o origin. The same percentage declines to 88.4% or people aged 20-24, 54.1% or people aged and 21.6% or people aged See Istat 1997, pg.224 ss. 16 Translating into a representative cohort, these igure imply that in a cohort o 100 individuals, 37.8 do not complete a secondary school (which in Italy is not yet compulsory), and 33.3 enrolled the university. Corresponding igures or the Italian population are 39 droppers and 40.1 enrolees (see ISTAT 1999). The lower igure or the SHIW sample could be due to children enrolled at the university and not cohabithing with the amily. 17 The SHIW collect inormation about net incomes, whereas or iscal purposes (and or university tuition) people declare gross incomes. Thus we have converted net income into gross incomes using the available inormation on tax intervals, tax marginal rates and tax deduction based on amily composition. 8

9 yields a very limited return. For example, i we annualise the rate o returns reported in the irst column, we ind that attending a proessional (istituto tecnico) or a vocational (istituto proessionale) yields a yearly rate o return o about 6%, whereas attending a generalist school (liceo classico or liceo scientiico) generates a lower return o 5%. 18 However, i a student decides to proceed to the university, the yearly rate o return raises signiicantly, rom the lowest 7.3% or Literature and Philosophy, passing through 11.7% or Economics and Political Sciences up to 13.9% or Law. 19 This situation suggests that the educational career is signiicantly predetermined by the choice o the secondary school undertaken at the age o 14, mainly by the amily. In act, those parents able to inance a college education or their children and/or expecting them to be above a minimum threshold o perormance tend to push them to enrol in generalist schools (licei) even i these schools oer a lower rate o return per se. When their children reach the end o a generalist secondary school, the expected gain o continuing in schooling will be higher than the corresponding gain acing a student leaving technical or vocational secondary school. Thus, the mere existence o a dual curricula 20 in secondary education tends to sel-select the students according to their perceived ability and/or availability o amily inancial resources. I we consider the choice o secondary school attended and completed by children in our sample (see Table 3), we use multinomial logit analysis to ascertain the determinants among the dierent types o secondary school. The only signiicant evidence we are able ind relates to amily educational background: having both parents with a secondary school diploma raises the probability o enrolling at high school, which in turn raises the probability o enrolling at the university. On the contrary, we are unable to detect any signiicant eect o amily income and occupations 21 in conditioning the educational choices o children. Notice however that the sample is biased, because we are considering only students that completed the secondary school, whereas it is common knowledge that dropout rates vary considerably among dierent types o schools. 22 Unortunately we are unable to correct or this bias, since the dataset does not report inormation about students who attended without completing any order o school: as a consequence, among the youngsters without a secondary school diploma we are unable to detect whether they enrolled at any type o secondary school, and which one. We also lack direct inormation about the educational record o these students. Nevertheless, i we accept a wider deinition o unobservable ability, controlling or parents education provides a rough indication o the direction o this selection. 18 Altonji 1993 explains the lower rate o return beore entering college as a result o ex-ante uncertainty on the possibility to complete i. 19 Annualisation is obtained by taking the n 1 root o (1+coeicient reported in Table 2), where n is the expected number o years necessary to achieve the degree. While legal duration is 4 years, eective duration exceeds this amount: or this reason we have taken n = 5. Even i they use net incomes (while here gross incomes are adopted), Brunello et alt. 1999, table 11, report similar values or males (6.5% or any secondary school and 10.6% or any university degree). However, we know that OLS estimates are biased because we cannot control or unobservable ability. When using IV estimates, Brunello et alt ind higher returns (8.6% or secondary education and 13.2% or university education). 20 The Italian secondary education was designed in parallel and non-communicating tracks the ascist period, taking the German system as a reerence model. Additional tracks were subsequently added in the 60 s, but a clear distinction between high schools and the remaining still persist nowadays. 21 In estimating Table 3, we also tested the signiicance o ather and mother occupations, but none o them resulted signiicant, and thereore we have not reported them. 22 In the percentage o boys (girls) ailing during the secondary school was 27.8% (17.3%) in vocational schools, 21.5% (13.0%) in technical schools, 25.3% (12.3%) in teacher training school and 12.5% (6.7%) in high schools. See Gasperoni

10 Table 1 - Income and marks comparisons between entire population and university samples Italian population * Italian population with a son enrolled at the University ** Enrolled at State University o Milan*** gross amily income (1995 millions) average amily income (1995 millions) median amily income (1995 millions) standard deviation inal marks at exit o secondary school average marks (max 60) n.a standard deviation 7.16 n.a * Data on incomes rom Bank o Italy sample (1995), reerred to the population with at least one child in the age between 19 and 26. Data on marks rom Gasperoni 1997, tab.4; they are reerred to the academic year ** Data rom Bank o Italy sample (1995), reerred to the population with at least one child in the age between 19 and 26 recorded as student. *** Data rom administrative sources, reerred to the aculties o Economics, Law, political Sciences and Mathematics only: See the Appendix. For comparability, data on marks are restricted to students irstly matriculated in Table 2 - Dierentials in (gross) return or dierent educational degrees (secondary school and university) - percentage increase OLS estimates Completed secondary school: High school (liceo classico or scientiico) (4.75) (3.08) (1.68) High school (liceo artistico) (1.35) (0.69) (0.21) Teacher training (istituto magistrale) (5.82) (3.46) (1.93) Proessional (istituto tecnico) (7.35) (5.75) (3.87) Vocational (istituto proessionale) (5.53) (4.18) (3.66) University degree (BA/BS): Economics and statistics (9.10) (7.54) (4.78) Law (10.5) (8.85) (6.47) Political Sciences (5.90) (4.35) (2.49) Mathematics (7.78) (6.09) (3.78) Agriculture and veterinary science (4.15) (3.96) (2.06) Medicine and surgery (11.1) (9.39) (5.68) Engineering (7.34) (7.00) (3.95) Architecture and planning (3.68) (3.65) (1.29) Literature and philosophy (6.90) (4.14) (1.77) or comparison: return to one year o education (34.76) (25.27) (10.68) Sector o activity no yes yes Job position no no yes R² Controls: experience, experience squared, gender, region o residence, sel-employment, educational dummies or less than secondary education. Excluded case: male, blue collar, without ormal education, working in agriculture, resident in islands. T- statistics in brackets. We can now move to a direct analysis o the determinants o the choice o enrolment at a college. Table 4 reports the maximum likelihood probit estimates o the relative contribution o each variable (evaluated at sample means) in determining the probability o being observed as a student (necessarily a university student) in our sample, conditional on having completed a secondary school. We start by noticing that the most relevant variable seems to be the secondary school o origin: thus sending a child to a high school, i successully completed, almost ensures continuation at university level (contribution in probability between and 0.519). Quite surprisingly, there is no evidence o gender discrimination, even i belonging to a numerous amily reduces the probability o access 10

11 (probably because o limited resources). Finally North-eastern and central regions seem characterised by lower enrolment rates. 23 When we introduce variables measuring the amily background (second column o Table 4), we ind that amily total (gross) income exerts a statistically signiicant positive eect only when we omit any variable measuring unobserved ability o the student (here again captured by the human capital available within the amily). When we add the educational attainment o both parents (third column o Table 4), we ind that having parents that went beyond compulsory education (lower secondary since 1962) helps in accessing the university. As already ound with respect to other countries, having a graduate mother provides great pressure in the children to repeat the university experience. This obviously translates into higher persistence in intergenerational transmission o education. 24 Table 3 - Choice among dierent secondary schools - amily with a student aged in 1995 who has completed a secondary school multinomial logit estimates vocational technical high school art school teacher train. gender (1=emale) 0.458** 0.332*** *** age 0.833** 0.841** 0.814*** 0.827* resident Northeast 0.236*** resident Centre resident South amily background (log) amily income number o children single parent amily ather with compl.lower secondary/vocational ** ather with compl.higher secondary ** ather with university degree 0.226* ** mother with compl.lower secondary/vocational mother with compl.higher secondary ** * mother with university degree * head o household sel-employed Pseudo R² 0.12 χ² Number o cases 1615 Estimated coeicients are transormed to relative risk ratios. *** 99% signiicance ** 95% signiicance * 90% signiicance. Comparison group: other type o diploma. The introduction o some proxy or unobservable ability reduces the explanatory power o amily income. How can we interpret this result? We are tempted to take this as evidence that liquidity constraints do not prevent access to university education, whereas individual ability is more important. 25 But this conclusion seems contradicted by the signiicant (negative) eect o the number o siblings and by the similarly negative eect o having a ather and/or a mother unemployed (ith column o Table 4). Still additional actors play a role in the college choice. In accordance with the theoretical model presented in the previous section, amily inancial resources and unobservable ability are the main determinants. Thereore we would not expect that parental occupations could inluence this choice any urther, since we are controlling or both parental income and education. 23 This is not conirmed by aggregate data: in the percentage o enrolment on student ending the secondary school the previous year was 65.0% in Northwest, 74.3% in Northeast, 84.5% in Centre and 59.7% in South and Islands. See Istat 1997, p See the regressions reported in Checchi at alt As long as parental education and parental incomes are correlated, multi-collinearity could be responsible o the decline in signiicance or amily income. However, taken at ace value, our results imply that parental education is a stronger determinant o a child s schooling than parental income. Our interpretation is also strengthened by the act that the choice o a secondary school (especially in the case o high school ) was conditioned by parental education but not by amily income (see Table 3). 11

12 However, we ind that having a household head holding a managerial or entrepreneurial job ensures the highest probability o enrolling at the university, 26 and having parents with clerical jobs also has an eect, i less powerul. 27 This result cannot be accounted or in the present ramework, and suggests that to a certain extent status attainment could aect college choice. We would have liked to introduce a measure o opportunity costs in attending the university. However, the very same variables that may raise individual employability (and thereore expected orgone income) are potential regressors in determining college choice. Thereore adopting a probit estimate o employment probability within the same sample and inserting the estimated scores in current regression would have produced the same set o regressors. Indirect evidence o the importance o opportunity costs can be obtained by noticing that living in a amily whose household head is sel-employed (where thereore work opportunities are higher) halves the probability o attending the university. On the whole, attending a university in Italy is more likely or students rom educated amilies and/or where the parents hold occupations with some prestige in society. Evidence about potential existence o liquidity constraints is mixed, and some additional role can be played by opportunity costs due to alternative employment availability. 26 The ith column o Table 4 reports the coeicients on ather and mother occupations that survive a stepwise elimination with a p-value threshold o When education becomes a status symbol, it enters the children utility unction directly, and previous conclusions have to be modiied. See Fershtman et alt

13 Table 4 - Determinants o enrolment at the university - ML probit estimates (t-statistics in parentheses - constant included - coeicients report the probability eect a discrete change o a dummy variable rom 0 to 1 or one point increase in a continuous variable) # obs : Depvar: sonuni sonuni sonuni sonuni sonuni Demographic gender emale=1 (-1.22) (-1.18) (-0.61) (-0.70) (-0.26) age (-7.66) (-7.98) (-7.51) (-7.23) (-7.02) regio North-east (-1.59) (-1.66) (-1.11) (-0.91) (-1.18) regio Centre (-2.23) (-2.08) (-2.01) (-2.01) (-2.02) regio South (-0.97) (0.22) (-0.28) (-0.47) (-0.31) regio Islands (0.24) (0.95) (0.49) (0.42) (0.40) Secondary school degree vocational (-0.95) (-0.98) (-0.67) (-0.67) (-0.71) technical (0.66) (0.58) (0.62) (0.71) (0.58) high school (liceo) (7.16) (6.95) (6.34) (6.39) (6.32) art school (lic.art.) (2.18) (2.08) (1.88) (1.94) (1.70) teach.trng (magistr) (1.37) (1.26) (1.13) (1.23) (0.79) Family background (income and parental education) log amily income (2.12) (-1.22) (-1.02) (-2.66) n.children (-2.69) (-2.78) (-2.83) (-2.46) single parent (0.03) (0.55) (0.76) (-0.70) ather lwr sec./vocat. (-0.64) (-0.48) (-0.40) 13

14 ather high sec./shrt dgr (2.60) (2.85) (2.48) ather college (1.12) (1.29) (0.68) mother lwr sec./vocat. (0.42) (0.28) (0.12) mother high sec./shrt dgr (3.69) (3.38) (2.42) mother college (4.83) (4.75) (3.51) household head selemployed (-5.10) Family status (parental occupations) th white collar low (1.51) th teacher (1.49) th manager (2.83) th entrepr (3.09) th selempl (2.93) th am.irm (1.62) th unemployed (-1.94) th retired but working (2.12) mth white collar low (1.37) mth unemployed (-2.32) mth housewie (-3.06) mth retired but working (-1.19) mth retired no working (-1.06) R² =================================================================== Excluded case: male student, living in north-western Italy, with other type o diploma obtained rom secondary school, with both parents with primary or without ormal education, and with both parents employed as blue collars. 14

15 Once a amily has reached the decision to send a child to the university, the next step is to choose among available aculties and universities. 28 From this stage on, we move to an administrative sample o students enrolled at the State University o Milan, attending certain aculties: Law, Economics, Political Sciences and Mathematics. 29 Since administrative reasons prevented us rom analysing the entire population, we have chosen these aculties to include all social disciplines available in the city (a Faculty o Sociology opened in 1998), adding a scientiic aculty which has some overlap in research areas. Going back to Table 2 (reerring to the entire Italian population), expected earnings are highest or the Faculty o Law, whereas a degree in Economics and/or in Political Science provide similar pay, leaving the Faculty o Mathematics at the bottom o the distribution. Having said that, it is not clear according to what criterion students are sorted to more proitable degrees. Even without inormation on the entire university population, we try to iner some inormation rom available data. We start by noticing that, according to the marks obtained at the completion o the secondary school taken as a proxy o unobserved ability, the above the average marks students enrol in the aculties o Economics and Mathematics, whereas below the average marks students tend to attend the Faculty o Political Science. With a national average mark o 44.5 (expressed in sixties), students irst enrolled in in the Faculty o Mathematics exceeded that average by 9.6%, ollowed by the Faculty o Economics (+6.6%) 30, the Faculty o Law (-1.9%) and the Faculty o Political Science (-2.8%). I we resort to multinomial logit analysis and take our data set as the universe o available choices within the State University o Milan (as it is shown in Table 5), we discover that amily inancial resources (as measured by total income and sel-employment o household head) are almost irrelevant in the choice among these aculties. Previous schooling perormance seems more important: as already remarked rom simple descriptive statistics, students with better marks tend to go to the aculties o Economics and Mathematics; there is also a clear preerence rom students exiting the high school (liceo classico) or the aculty o Law. Age does not seem to identiy a precise pattern o choice, whereas women tend to go to the aculty o Law. Even i this evidence is limited, since we are not including other choices available within the city o Milan, it strengthens the previous result that previous educational choices at the secondary level and unobservable ability are more important determinants than ease o access to inancial resources in shaping the educational career o prospective graduates. 28 The reader has to keep in mind that each university does not contain all aculties, and sometimes to choose a speciic university implies to move (or to commute) to a dierent city. We do not have aggregate inormation about this phenomenon, but we believe that the local availability o given aculties may condition the inal choice o aculty. 29 The main dierence with the previous SHIW sample is that, apart rom demographic and secondary school inormation on students, we lack inormation about parental education. Conversely we know the average mark obtained at the end o secondary school: thereore we are orced to change our proxy o unobserved ability. As long as people with learned parents achieve better results at school (because educated parents value education more, put more pressure or schooling on their children, have more books available, and obtain more respect or their children rom teachers), the two variables are related. Gramsci was well aware o the correlation between the two variables: In many amilies, especially among intellectuals, children receive rom the amily a training which integrates (ormal) education. In other words they 'breath rom the air' knowledge and model roles that acilitate their educational achievements." (Gramsci 1975, p our translation). Analogously: "Consequently selection in school avors children rom those amilies that already possess dominant cultural advantages" (Shavit-Blosseld 1993, p.30). 30 The reader has to keep in mind that students enrolled at the Faculty o Economics o State University could represent a selselected sample, because o the existence o admission tests partially based on the marks obtained at the completion o the secondary school. However occasional inormation indicates that in several circumstances students that eectively enrolled belonged to the bottom o the admission list. This is due to reusal to accept an oer when accepted at the same by private universities oering courses in economics (Università Commerciale L.Bocconi and Università Cattolica del Sacro Cuore). 15

16 Table 5 - Choice among some aculties within a state university - irst enrolment in multinomial logit estimates Economics Law Mathematics gender (1=emale) ** 0.177*** age *** ** * resident Northwest resident Northeast ** resident Centre Secondary school degree Completed high school (liceo classico) *** 0.572* * Completed high school (liceo scientiico) *** Completed accounting school (ragioneria) *** *** Completed vocational school (tecnico/proess.) *** ** inal mark at exit secondary school 0.102*** 0.015*** 0.107*** amily background (log) amily income * * head o household sel-employed (log) amily income*household head sel-employed * Pseudo R² χ² Number o cases 7750 Estimated coeicients are transormed to relative risk ratios. *** 99% signiicance ** 95% signiicance * 90% signiicance. Comparison group: Political Sciences. 4. Academic perormance Once the crucial decision to enrol at the university has been undertaken, the students open up to academic lie. As expected rom previous results about irrelevance o inancial resources in attending a university course, enrolment ees or public universities in Milan are rather low and correlated with amily income: in admission ees ranged between Italian lire (687 euros or the lowest amily income) and lire (1885 euros or the highest amily income). But enrolling is just the irst step o a long ladder. It is widely known that the Italian university is alicted by two problems: the dropout rates and the excessive duration o a student s academic career. Table 6 estimates the extent o the irst phenomenon in our sample by going back to the original number o new enrolments in each year and observing how many student have survived 2, 3 and 4 years later. The dierence between original enrolment and actual enrolment can be taken only as a rough measure o drop-out rates, because a student could have changed aculty, or transer rom other universities to higher years is possible. We cannot go beyond 4 years because this is the minimum oicial time length to complete each course, and students could disappear rom the sample just because they graduate. However this is an exception rather than the norm, as can be seen rom oicial sources: the average length o time spent obtaining a degree in the State University (reerring to the academic year ) is 6.8 years or the Faculty o Law, 7.5 or the Faculty o Political Sciences and 6.8 or the Faculty o Mathematics. 31 National surveys indicate that only one tenth o the students complete their undergraduate studies on time. I more than hal o the students enrolled does not complete their curricula, the surviving one constitutes a sel-selected sample. The extent o sel-selection can be judged by observing representative statistics or year o enrolment, as in Table 7. But no clear pattern emerges rom descriptive statistics: potential sel-selection does not appear based on amily income nor on students 31 We do not have comparable igure or the Faculty o Economics o State University, since it opened in and by the time these data were collected (April 1997) no student has yet completed his/her career. A survey conducted at the national level in 1995 and reerred to graduates in 1992 report an average duration or a degree in economics o 6.25 years (Istat 1996). 16

17 unobservable ability (here proxied by inal mark at the exit o secondary school and by the average mark obtained in passing the university exams). Taking individual data on enrolment o two subsequent years, we trace out those students who did not re-enrol, and we can explore the determinants o this choice. Because o lack o data, this could be done only or the students enrolled in the Faculty o Economics with reerence to students who enrolled in and did or did not in The probit model reported in Table 8 indicates that amily income and secondary school education (both in terms o type and marks obtained at exit) are statistically non-signiicant with respect to this choice, whereas age and academic perormance within the university are the only signiicant variables. When a student is getting older, is unable to pass a suicient number o exams per year and/or is unsatisied with the marks obtained, then she becomes more likely to drop out o the university. Given the act that the sel-selection o the sample is mainly due to poor perormance, we think it appropriate to concentrate on the analysis o the determinants o academic perormance. Table 6 - Estimated dropout rates or year o enrolment Economics Law Pol.sciences Mathematics enrolment at 1 year 40.60% 44.63% 53.50% 62.91% enrolment at 2 year 34.85% 35.99% 51.09% 61.56% enrolment at 3 year 34.10% 25.62% 36.72% 53.04% enrolment at 4 year 11.56% 3.89% 3.26% 10.81% Note: it reports the ratio between students with "active enrolment" status and the irst-year enrolment o the corresponding cohort. Table 7 - Descriptive statistics or year o enrolment Economics Law Pol.sciences Mathematics amily income (millions It.liras 1995) enrolment at 1 year enrolment at 2 year enrolment at 3 year enrolment at 4 year inal marks at exit scnd school (max 60) enrolment at 1 year enrolment at 2 year enrolment at 3 year enrolment at 4 year Average marks at university (max 31) enrolment at 1 year enrolment at 2 year enrolment at 3 year enrolment at 4 year

18 Table 8 - Probability o dropout - State University Faculty o Economics ML probit estimates (t-statistics in parentheses - constant included - coeicients report the probability eect a discrete change o a dummy variable rom 0 to 1 or one point increase in a continuous variable) # obs : Depvar: dropout dropout dropout gender emale=1 (-1.31) (-1.33) (-0.89) log(age) (5.98) (5.95) (6.91) regio North-east (0.85) (0.87) (0.50) regio South isld (0.64) (0.59) (0.84) Secondary school degree hgh school (lic.class)(-0.38) (-0.31) (-0.45) hgh school (lic.scien) (0.09) (0.16) (-0.18) technical (ragion.) (0.37) (0.43) (0.00) vocational (tecnica) (0.58) (0.63) (0.10) log(marks) exit scnd (-0.71) (-0.71) (1.23) Family background (income) log amily total income (-0.38) (-0.22) Academic perormance log exams per year (-9.77) log average marks per exam (-2.26) R² ======================================== The problem is how to deine a precise measure o students perormance. According to the Italian tertiary education system, a student obtains a BA degree when she has passed a predetermined number o exams, obtaining in each o them at least a minimum mark o 18 (out o 30), and deending a inal dissertation (tesi di laurea). There are a minimum number o years o enrolment, which are 4 or 18

19 most degrees (including the present under analysis), but not a maximum. I we take the number o exams corresponding to each degree course and we multiply them by either 18 or 31, 32 we obtain respectively the minimum or the maximum result a student can obtain through her academic career. By dividing the latter value with the minimum number o years o university attendance, we obtain the supremum in the range o an indicator o student perormance, the inimum being zero (since potentially there is no time limit to complete the academic career). For each student it is thereore possible to deine the perormance by taking the sum o the marks obtained in passed exams and dividing it by the number o enrolment years, that is p mi perormance= i = 1 (22) n where p is the number o passed exams, m i is the mark obtained in the i-th exam and n is the number o years o active enrolment. Let us notice that the perormance indicator can be obtained as the product o two other indicators o perormance, namely the average mark obtained in passed exams and the speed at which a student is undertaking the exams. In symbols perormance = average mark speed = p mi =1 p p n (23) i It is important to distinguish between the two components, because there is an implicit trade-o between the two. A student can decide to dedicate one year to prepare each exam, obtaining high marks (and consequently cumulating a high value o average mark), but this produces a very long stay at the university to complete the career (a very low value in speed). At the opposite extreme, a student can decide to devote the minimum amount o time required just to pass an exam with the minimum mark (a very low value in average mark), thus being able to get through a higher number o exams (a high value in speed). 33 The descriptive statistics or these variables are reported in Table 9. It can be noticed that students rom dierent aculties exhibit similar values in average mark, but dierent values in speed: as a result o combining these two pieces o inormation, students rom the Faculty o Law who experience the lowest cumulated drop-out rate also exhibit the highest perormance index. They are then ollowed by students rom the Faculties o Political Sciences, Mathematics and Economics. Table 9 - Descriptive statistics o student perormance Economics Law Pol.sciences Mathematics #exams required to complete the career minimum # years to complete course theoretical maximum o perormance average perormance (last 5 years) average mark (last 5 years) average speed (last 5 years) But what are the determinants o students perormance? In Tables 10 we estimate ordinary least square models predicting perormance in each aculty, based on previous student achievements and amily inancial resources. 34 Older students perorm better, probably due to sel-selection (they are 32 It is possible to pass an exam with the mark "30 cum laude": we have coded this outcome as An alternative way to look at the same concept is to think o perormance as an index o productivity in the production o a vertically dierentiated product, where average mark is the quality and speed is the quantity. 34 Estimates o average mark and speed determinants are available rom the author upon request. 19

20 the survivors o the process o dropping out). We also ind corroboration o previous evidence on unobservable ability : students with better educational records (higher marks at the exit o secondary school, attendance o high school) perorm better, whereas they are at some disadvantage when they are resident in a region that is ar away. But a more interesting result is the evidence on inancial resources: amily income inluences perormance positively in the case o attending the aculties o Law and Political Sciences, 35 whereas it is irrelevant in the other cases (Economics and Mathematics). 36 The eect o amily income on perormance is all due to the eect on speed and not on average mark, indicating that students rom richer amilies tend to go aster in their academic career. 37 It is not easy to provide interpretations o the evidence on inancial resources. Prima acie, it seems that liquidity constraints do not bite, because otherwise we should have ound that students rom poorer amilies would have had a higher speed (even at the cost o a lower average mark). However, our theoretical model presented in section 2 oers a rationale or this evidence: whenever we introduce a perormance threshold, we do not observe students below the threshold because amilies do not invest. I we consider the high size o dropout, we are let with sel-selected students that passed the threshold. In the spirit o the model, when neighbourhood eects and/or social networking are present, the higher the amily income the higher the investment in education, and the higher is student s perormance (see equation (17)). Thus observing a positive correlation between amily income and perormance could be taken as evidence in support o either a networking hypothesis (students rom richer amilies go quicker because they have better employment prospects) or a neighbourhood hypothesis (students rom richer amilies go quicker because they live in a better educated environment). An alternative way to rame the latter hypothesis is thinking o amily income being correlated with parental education. Since students would receive more support rom their amilies o origin the higher the educational achievements o their parents, which would be richer on average, the positive correlation o perormance with income would be a case o spurious correlation in the absence o inormation on the educational achievements o parents. This interpretation is however contradicted by the evidence on average grade, where in no case do we ind any positive eect o amily income: had a cultural background played any role, we should have expected a positive eect o amily support on the average marks obtained in passing exams. In addition, in previous interpretation we have considered the average marks at exit o secondary school as an alternative proxy to parental education or unobservable ability o students. Since in Table 10 we are controlling or the ormer variable, the amily income eect can be taken as an independent eect. Thus we are let with a networking explanation: other conditions being equal, students rom richer amilies ace better employment prospects, and thereore ace a higher opportunity costs by protracting their university career. This creates an incentive to reach a higher speed in passing exams, irrespective o the marks obtained. This explanation is not contradicted by the absence o eect within two aculties. In the case o the Faculty o Economics we can argue that these students could represent a sel-selected sample (see ootnote 33), since students with better prospects have already 35 In the case o the Faculty o Law this eect is lighter in the case o sel-employment o the household head: parents seem to put less pressure on the shoulders o their o-springs when amily income exceeds millions o Italian liras ( euros). 36 A positive and signiicant eect o amily income is ound in a sample o students enrolled to a private university oering courses in economics. Thus the absence o eect or the students enrolled at the Faculty o Economics in the State University could be attributable to sel-selection: students rom poor amilies, who have chosen to enrol a aculty o economics, sel-select into a State University. 37 Unortunately we do not have inormation on whether the students are ull-time or part-time students. Anecdotal evidence tells us this phenomenon could be rather widespread: in a survey conducted among the students o the Faculty o Political Sciences during 1996 (910 cases), 18% o the sample sel-declared "ull-time worker", 19% "part-time worker" and an additional 34% "occasional worker". Had we had this inormation or our samples, we could have tested an alternative version o the liquidity constraint model: students rom poorer amily have to work in order to inance their studies, and thereore less time is devoted to the academic career. On the contrary, students rom richer amilies are implicitly exempted rom work and thereore dedicate more time to study activity. 20