Credit Constraints and Intergenerational Mobility in Education in the Italian Regions

(1)

Credit Constraints and Intergenerational Mobility

in Education in the Italian Regions

Student ID: 22346559

Name: Jack Philip Kleinjan Programme: MSc Public Administration Track: Economics and Governance Supervisor: Max van Lent

Second reader:

Date: 23-12-2020 Word Count: 13,867 Page Count: 44

(2)

Abstract

Italy has a persistent and significant regional divide between the northern and southern regions. This paper examines this puzzling phenomenon by examining differences in intergenerational mobility in education between the regions, and the role of credit constraints in the intergenerational transmission of education between parents and their children. This paper adopts a novel approach to the regional divide by examining educational opportunities between the regions, specifically it answers the question: what is the relationship between intergenerational mobility in education and credit constraints faced by families in the regions of Italy? The results contribute to the literature on the Italian regional disparity by finding that the persistence of educational attainment between generations is large in Italy, especially in the Southern region. Children from families that face credit constraints have a higher intergenerational persistence in educational attainment, with the effect being relatively larger in the North, although families facing credit constraints are more prevalent in the South. These results indicate that there is a stark inequality of (educational) opportunity present in Italy that disadvantages the South of Italy more than the North and is especially detrimental to children growing up in credit constrained families.

(5)

Introduction

Italy has well documented regional inequality between a wealthier North and a poorer South that existed pre-unification (1861) and persists to this day (Felice, 2012). This inequality can be seen in many areas: human capital, social capital, growth, social mobility, and more (Barbieri et al., 2020; Felice, 2012, 2015). A noteworthy difference between the North and South of Italy1_{is that of social} mobility, which can be a gauge for the equality of opportunity present in society. As noted by Checchi and Peragine (2010, pp 429) “equality of opportunity seems to be the prevailing conception of justice in Western liberal societies”. The difference between North and South Italy in this regard can be considered especially galling, and the aim of public policy ought to be to address it.

It is a stylized fact that a worker’s stock of human capital is the major determinant of their productivity and therefore of their earnings. A person’s human capital acquisition is primarily attained via education, although family, personal initiative, and on the job training may also play a significant role (Laing, 2011). These sources of education are not entirely separable; for instance, genetic transmission affects personality traits, aptitudes, and intelligence; families themselves influence their child’s choices; even neighbourhoods can have effects on educational attainments of its residents; ultimately these interact with each other in many complex ways and can be difficult to disentangle from each other (Black & Devereux, 2010; Borghans et al., 2008; Galster, 2010; Sacerdote, 2011).

Relative to less educated individuals, more educated individuals are less likely to be unemployed, earn more, are healthier, and are less likely to be convicted of engaging in criminal enterprises (Laing, 2011). These are among some notable reasons why education is considered a key factor in the intergenerational transmission of earnings, which is the most commonly used measurement of social mobility and a measure of how equal access is to opportunities for new generations (Black & Devereux, 2010).

Parents wish to invest in their children’s education to improve their lifetime earnings, yet in the absence of perfect capital markets, different households can invest different amounts in their children’s education primarily based on how much parents earn (Becker & Tomes, 1986). Better

1_{For clarity this paper follows the convention of the literature to consider the South to be both the official}

NUTS level 1 (nomenclature of territorial units for statistics) region named “South Italy” and “Insular Italy” (i.e. the Islands); the North are the NUTS regions “Northwest Italy”, “Northeast Italy”, and “Central Italy”. Central Italy is inconsistently included or excluded in the literature. To ensure that it doesn’t skew the results significantly the North as considered only Northwest and Northeast Italy have also been run and added in appendix 2.

(6)

educated parents are better situated to invest more into their children’s education than less educated parents (Lee & Lee, 2020). It has been shown that the level of parental education has a significant influence on the education level of children’s education, which, in absentia of government involvement, may perpetuate between generations leading to a stratified and immobile distribution of educational attainment in society (Becker et al., 2018). The correlation between parental education and children’s education is likely to follow the relationships observed between parental and children’s earnings, due to the important role of human capital in determining earnings. If this is the case then the phenomenon of the ‘Great Gatsby Curve’ observed in the literature on intergenerational mobility in earnings is likely to be observed for intergenerational mobility in education; namely, a correlation between high intra-generational inequality and high inter-generational inequality. In the presence of credit constraints by families for investing in their children’s education, poorer families (usually less educated) may not be able to invest their preferred amount into their children’s education. This is for both monetary and non-monetary investments. In the US high income families spend nearly seven times as much money on out of school enrichment activities (Lopez & Caspe, 2014). The non-financial investments (such as time spent) by fathers into their children is correlated with their financial position (Vickery, 1977), notably fathers that cannot provide for their children are likely to drop out of their lives entirely (Christiansen & Palkovitzm, 2001), lower earning fathers are also less likely to reside with their children (Coley & Morris, 2002), their financial contribution has been found to be essential for their active parenting (Gupta et al., 2004). Thereby less educated families lead to lower educational attainment for their children, ceteris paribus. Conversely, richer families may not face such credit constraints, or not require credit and are, thereby, able to invest their preferred amount into their children’s education leading to higher educational attainment, ceteris paribus. A formal model of the intergenerational behaviour of families leading to this intuitive conclusion is presented in the theoretical framework.

This paper addresses two contemporary areas of research: the determinants of intergenerational mobility in educational attainment and the North-South divide in Italy. It is an explanatory and descriptive piece of positive research that examines the role of credit constraints on intergenerational mobility in educational attainment and, in turn, on the regional differences in educational attainment between the North and South of Italy. The regional disparity in Italy is not only a contemporary research interest but also, as noted, a longstanding policy issue, which has yet to be successfully addressed (Felice, 2015). The research design utilised in this paper is to conduct regression analyses using data from the Italian wave of the 2008 European Values Study (EVS) with 1519 observations (EVS, 2016c).

(7)

The structure of the paper is as follows: this first section introduces the topic of intergenerational mobility in educational attainment and the regional divide in Italy. The following section explains the theoretical framework of Becker and Tomes’ model (1979, 1986) and how it applies to the Italian regional divide. The research design of the paper is discussed in the third section. The fourth section examines the methodology. In the fifth section, the analysis is conducted with a discussion of the results in section six including limitations of the analysis, research design, and methodology, as well a discussion of policy implications from the results and theoretical framework. Finally, section seven concludes.

(8)

Theoretical framework

The seminal contributions of Becker and Tomes (1979, 1986) built a theoretical framework that has since been used as the building blocks for the understanding and analysis of intergenerational mobility. Fundamentally, the model is a child’s human capital accumulation model with altruistic parents who, based on their preferences and constraints, decide how much to invest in their child’s human capital. The model from Becker and Tomes’ 1986 ‘Human Capital and the Rise and Fall of Families’ is presented in the next section, unless stated otherwise their paper is the source used in the theoretical framework.

The Becker-Tomes model with perfect access to capital markets

Parents endow their children. Endowments defined very broadly include genetic traits, familial reputation, connections, and culture which is shown in equation 1.

EQUATION 1.

𝐸_𝑡𝑖 _{= 𝛼}

𝑡+ ℎ𝐸𝑡−1𝑖 + 𝑣𝑡𝑖

The superscript i refers to family, the subscript t to generation. E is the endowment, 𝛼𝑡 is the ‘social

endowment’, the endowment that is common to all members of a generation, h is the degree to which endowments are inherited and are assumed to only be partially inherited (0>h>1) and is given, v is ‘luck’, as well as other non-systematic factors in the transmission process. Parents cannot affect the transmission process of endowments. An important consequence of the assumption that the degree to which endowments are inherited is between zero and one is that endowments in family will regress to the mean; for instance, if the parent has higher than average endowments then the child will inherit only a part (h) and will also be above the average but less than their parent.

Now, assume two periods of life: childhood and adulthood. The adult earnings of the child are given by equation 2.

EQUATION 2.

𝑌𝑡 = 𝛾(𝑇𝑡, 𝑓𝑡)𝐻𝑡+ 𝑙𝑡

Equilibrium in the factor market (𝑇𝑡, 𝑓𝑡), determines the earnings of one unit of human capital

(earnings denoted by 𝛾, human capital by 𝐻𝑡), depending positively on technological knowledge (𝑇𝑡),

(9)

assumption is made that 𝐻𝑡 is chosen so that 𝛾 equals one. The earnings of human capital (𝛾) is

common to all families. Other relevant assumptions are that human capital is simplified to be homogenous and that the accumulation of human capital in childhood is proportional to the amount gained in later life. The human capital of people may be substitutes but each individual forms their own ‘human capital market’ with their individual rates of return depending on the amount invested in them (x) and on aggregate stocks of human capital in the economy (𝑓𝑡). Marginal rates of return to any

person exhibit diminishing returns because the amount of foregone earnings increase and the amount of working life left shortens. The expected adult earnings and adult human capital of a child is determined by the inherited endowments (𝐸𝑡), parental expenditures (𝑥𝑡−1), and public expenditures

(𝑠𝑡−1) on the child’s human capital accumulation.

EQUATION 3.

𝐻𝑡 = 𝜓(𝑥𝑡−1, 𝑠𝑡−1, 𝐸𝑡), 𝑤𝑖𝑡ℎ 𝜓𝑗> 0, 𝑗 = 𝑥, 𝑠, 𝐸.

The marginal effect of the inherited endowments on parental and public expenditures is positive (𝜓𝑗𝐸 > 0). Non-human capital can be bought and sold in efficient markets and the rate of return on

assets is assumed to be equal for all people. It is assumed that parents know the return to investing in their child, because the endowed luck (𝑣𝑡), the social endowment (𝛼𝑡), and public expenditures (𝑠𝑡−1)

are known to them. Furthermore, it is assumed with perfect capital markets that parents can borrow at the asset interest rate and that the debt incurred can become the children’s obligation when they become adults. Another key assumption is that parents maximise their child’s welfare as long as it does not entail a reduction in their own consumption or leisure. Consequently, parents will borrow whatever amount needed to maximise the net income (earnings minus debt) of their child until the marginal rate of return on investment in their child’s human capital (𝑟𝑚) is equal to the marginal rate

of return to investing in assets (𝑟𝑡).

EQUATION 4.

𝑟𝑚 = 𝑟𝑡 𝑜𝑟 𝑥̂𝑡−1 = 𝑔(𝐸𝑡, 𝑠𝑡−1, 𝑟𝑡),

In the case of perfect capital markets, parents can borrow on behalf of their children, who will be obligated to repay the debt when they are adults allowing for parents to invest an optimal amount. Given the positive marginal effectiveness of endowments, well-endowed children have higher expected adult earnings. An increase in the interest rate would reduce the investment in human capital and, therefore, children’s adult earnings. A (sufficiently large) increase in public expenditures would

(10)

raise the accumulation of human capital, the degree to which depends on the assumed substitutability between public and private expenditures. However given that private expenditure cannot be negative, it will lead to at least eventual higher earnings. Crucially, the adult earnings of children will not depend directly on the parent’s earnings or wealth and only indirectly on the inheritability of endowments with the assumption of perfect access to capital markets.

The Becker-Tomes model with imperfect access to capital markets

The assumption of perfect access to capital meant that investment in the family's children was separated from the earnings and resources of parents. Realistically, however, there is likely to be imperfect access to capital. Human capital is not good collateral for lenders because of the presence of moral hazard. Children can default on their debt by working with less effort or choosing a job with lower pay but higher emotional rewards. Furthermore, most societies would not allow parents to incur debts that their children must pay off (Becker & Tomes, 1986 pp S10). The following section shows the consequences for the theoretical model if parents are considered, without assets, who would have to finance their private expenditure on the child’s human capital accumulation by reducing their own consumption. The expenditure of parents on their child (previously shown by equation 4) in such a case depends also on the earnings of parents (𝑌𝑡−1), their generosity to their child (𝑤), and the

uncertainty about the luck of their child (𝜀𝑡−1).

EQUATION 5.

𝑥̂𝑡−1 = 𝑔∗(𝐸𝑡, 𝑠𝑡−1, 𝑌𝑡−1, 𝜀𝑡−1, 𝑤), 𝑤𝑖𝑡ℎ 𝑔𝑌∗ > 0.

Consequently, the formulation reflects that the effect of the child’s endowments is ambiguous (𝑔𝐸∗ ⋚ 0) because better endowment increases the marginal effectiveness of investment in human

capital. Meaning the child has higher expected adult earnings and, thereby, lowering the marginal utility for parents in investing in their child. Higher private expenditure on their child lowers the consumption of parents, which raises their subjective discount rates, what Becker and Tomes (1986) refer to as the “shadow cost of funds” (pp S11). The discount rates will be smaller for parents who have higher earnings or have poorly endowed children. Substituting equation 5 into the earnings-generating equations (equations 2 and 3) yields:

EQUATION 6.

(11)

Where 𝑘𝑡−1includes 𝑠𝑡−1, 𝜀𝑡−1, 𝑎𝑛𝑑 𝑤. The adult earnings of children still depend indirectly on the

transmission of endowments, which now also directly depends on the earnings of their parents. Becker and Tomes (1986 pp S10) remark that the direct relation modelled between the earnings of parents and the adult earnings of children is probably concave, as opposed to linear, because constraints to self-financing private investment in their child’s human accumulation decrease as parents’ earnings increase.

Applying the conclusions of the Becker-Tomes model (1986)

From this theoretical framework hypotheses can be drawn about which variables are of interest and the role they play in the intergenerational mobility of educational attainment:

● Credit constraints (i.e. constraints to self-financing private investment in their child) ● Parental Earnings (𝑌𝑡−1), (for credit constrained families only)

● Parents’ generosity to their child (w), (for credit constrained families only) ● Parental investment in child’s human capital (𝑥𝑡−1)

● Public expenditure (𝑠𝑡−1)

● Returns to investment in human capital (𝑟𝑚)

● Returns to non-human capital (𝑟𝑡)

These variables cross three institutions and the interplay between them will determine the outcomes of intergenerational mobility in education (Neidhofer, 2018). The first institution is the family, which affects children’s human capital accumulation indirectly via the inheritability of endowments, and directly via investment in the child’s human capital (𝑥𝑡−1). In turn, this depends on the parents’

generosity to their child (w), the returns to investing in their child’s human capital (𝑟𝑚), and the return

to non-human capital (𝑟𝑡). The second institution is the market, equilibrium in the factor market

determines the returns to human capital, in turn, determining the incentives faced by parents for investing in their child’s human capital. The third institution is the government which provides the public expenditures (𝑠𝑡−1) for children’s human capital accumulation.

The expectation is that some of these variables will vary by region, specifically (1) the incidence of credit constrained families [table 1], (2) parental earnings (𝑌𝑡−1) [table 2], and (3) parental investment

in their child’s human capital (𝑥𝑡−1). The EVS dataset that this paper uses and discusses in the research

design provides reasonable proxies on two of these factors, namely on the incidence of credit constraints and on parents’ generosity to their child. There is no clear reason to assume that parent’s generosity to their child would differ by region [table 3]. Interestingly, the child’s responsibilities to

(12)

their parents is also considered in the EVS, which theoretically may be a relevant factor about why parents might invest in their child if it is for the purpose of support in their old age (Becker & Tomes, 1986). The respondent’s income is also provided, but not that of their parents, however using this data can still indicate the regional discrepancy in earnings, and as noted in the literature this is a persistent phenomenon and likely reflects the respondents’ parents’ earnings (𝑌𝑡−1) reasonably well. The overall

level of public expenditure on education (𝑠𝑡−1) as a percentage of total government expenditure in

Italy (in 2017) was the second lowest in the OECD, after Greece (OECD, 2020b). The education system is highly centralised, so the public funding is unlikely to differ significantly between the regions (OECD, 2020a). Public and private education are assumed in the theoretical framework to be substitutes, if this is the case then that leaves a larger role for especially parental investments in their child’s human capital (𝑥𝑡−1). It is not clear what the values of this variable is per region, at a national level the share

of private expenditure on tertiary education is one of largest in the OECD in line with the theoretical framework’s expectations given the low level of public expenditure on education in Italy. Private investment can be expected to be lower in the South than in the North of Italy due to lower average earnings in the South. Another salient feature is Italy’s welfare state system which uses a “familiastic” approach whereby transfers are focused on older generations and provides limited support for younger generations; this raises the importance of the family in determining their child’s outcomes (Di Giulio & Rosina, 2007). Return to investment in education (𝑟𝑚) has been found to differ by region,

unexpectedly the results by Brunello et al., (2000) show that returns to education are higher in the South than in the North. Their explanation for this is that most educated labour is employed by the government which is much more important in the South than it is in the North, furthermore the wage differential between private and public sector is much larger in the South too. Return to non-human capital (𝑟𝑡) can be assumed to be approximately equal between regions, because there should be no

barriers to capital movements between the Northern and Southern regions it should entail that arbitrage opportunities should equate the return to non-human capital between the regions.

TABLE 1:THE INCIDENCE OF CREDIT CONSTRAINTS BY REGION

Grown up in credit constrained family? North South

Yes 28.60% 33.30%

No 71.40% 66.70%

This is a response to the question: “parent(s) had problems replacing things” with respondents asked to think back when they were aged 14. The validity of this question as the operationalisation of credit constrained families is more fully considered in the research design section. 14.7% of responses are missing for the South, 14.2% of responses are missing for the North.

(13)

Table 1 shows that the South has a higher incidence of credit constrained families, the difference being 4.7% between the North and South. The important conclusion is that there is a salient difference between the incidence of credit constrained families between the North and South, based on the conclusions of the theoretical model employed in this paper that leads to the expectation that more families in the South probably cannot invest the optimal amount into their child’s human capital accumulation.

TABLE 2:INCOME DISTRIBUTION BY REGION

Income group North South

Low 35.30% 54.50%

Medium 30.00% 26.20%

High 34.80% 19.30%

This is a recoded variable from the EVS based on the response to the annual household income reported by respondents into three possible categories: low, medium, or high. There are 37.2% of responses missing from the South, and 37.0% of responses missing from the North.

Table 2 shows that there is a large and significant difference between income levels of the North and South Italy. Most notably the difference between the percentage of the sample that is in the lowest income group in the North and South is 19.2%. Because this measures respondent’s income it does not show the differences in the variable ‘parental earnings’ (𝑌𝑡−1) that would be the basis for expecting

differences in parents’ abilities to invest in their children between regions. However, the respondent’s income differences is likely to reflect the differences between their parents (Barbieri et al., 2020). Therefore, it is reasonable to assume that there are large differences also in parental earnings (𝑌𝑡−1)

between the regions of Italy.

TABLE 3:PARENT’S GENEROSITY TO THEIR CHILD BY REGION

Parent's generosity (w) North South

Do utmost best for children even at cost to sacrifice their own well-being 81.70% 83.30% Parents have their own life and they should not be asked to sacrifice their

own well-being for that of their children 8.20% 8.00%

Neither statement 10.10% 8.70%

This is a response to the question: “Which of the following statements best describes your view on responsibilities of parents towards their children?”. The South has 2.6% of observations missing or unknown, the North has 2.8% of observations missing or unknown.

Table 3 shows that a large majority of parents in both regions put the well-being of their children before their own. The question does not provide the degree to which parents are willing to sacrifice their own well-being, however it presents the likely possibility that parent’s generosity to their children

(14)

(w) will be high. This will lead to more invested in their child’s human capital accumulation, ceteris

paribus. The differences between the regions are slight, with the South showing a slightly higher

willingness to sacrifice their own well-being. There is however a stronger likeliness for parents to consider this responsibility to run both ways in the South. In response to the question of whether children’s duty to their parents when they need long term care may come at the expense of their child’s well-being 73.7% of the Southern respondents responded affirmatively, versus 62.8% of the Northern respondents. On the basis of the literature this would be interpreted as the South investing in their children as an investment for support at old age, however there are almost no differences according to income groups which is unexpected because it would be most salient for the lower income group (Becker & Tomes, 1986). An alternative interpretation may be the stronger familial ties that exist in the South compared to the North; these ties, referred to as bonding social capital ties, have been posited as contributing to the persistent differences between North and South Italy (Sabatini, 2008). The types of social capital ties also have an effect on educational performance, with bridging ties associated with better educational performance at the community level, and bonding ties associated with worse educational performance at the community level (Menahem, 2011). The effect of bonding capital ties on educational performance becomes more pronounced as socioeconomic status of the community declines (Menahem, 2011), indicating that these may be especially salient in the poorer South.

Literature Review

The following section provides an overview of relevant academic literature to outline and take into account its implications upon the theoretical framework, research design, results, and policy implications.

The Becker and Tomes (1986) model as presented assumes that parents know the return of investment of investing in their child’s human capital. However, in a study by Cunha et al. (2013) the expectations of a group of socioeconomically disadvantaged African-American mothers towards the elasticity of child development to investment is substantially lower (between 4% and 19%) than the estimate by the researchers (between 18% and 26%). Whilst it is not clear whether there are systemic differences between expectations of return of investment to a child’s human capital accumulation according to the income of parents, highlighting a realistic complication to the simplification of the Becker-Tomes model. If poorer parents systematically underestimate the return on investment and richer parents either are more accurate or overestimate the return on investment then this will contribute to lower intergenerational mobility due to the noted variations in educational attainment.

(15)

The assumption of two time periods, childhood and early life, gloss over some important nuances about the timing of investments. The most crucial time for intervention is in early childhood. The flagship study for proponents of early childhood intervention is the experimental study, the Perry Preschool Program, which improved the early family environment of children in poor families with subnormal IQs (Almlund et al., 2011). The treatment group experienced far more successful socio-economic outcomes than the control group, but this emerged not from boosting IQ but rather improved personality traits and motivation of the treatment group (Heckman et al., 2006). Another experimental study, the STAR Project, assigned kindergarteners in the US to different size classes saw higher test scores and significantly higher earnings in adulthood for those placed in better kindergarten classrooms (Chetty et al., 2010). The first six years of life are the most effective time for policy intervention, this critical time in childhood development generates significant differences between children that persist throughout the life cycle, and this nuance is not incorporated in the theoretical framework (Heckman & Mosso, 2014).

The timing of investments also suggests that credit constraints in early childhood play a more important role in determining human capital investment than it does at later ages (Lochner & Monge-Naranjo, 2012). This coincides with most parents of young children being young themselves, who are at the early stages of a career and often face large liabilities in the form of a mortgage, other loans, and do not yet have solid credit histories (Lochner and Monge-Naranjo, 2012). Estimates by Cunha et al. (2010) suggest the optimal investment strategy is to invest more in children at a young age with declining investments with age, which may be hard or impossible to achieve (without credit) since the typical pattern of the lifecycle in earnings starts low and increases with age (Lochner and Monge-Naranjo 2012).

The model assumes the seamless transformation of an individual’s stock of human capital into earnings according to equation 2. However, this relation between human capital and wages does not always play out this way. There are well documented labour market frictions that complicate the determination of wages, most notably discrimination and information asymmetry (Spence, 1973). Discrimination may lead employers to not hire, or pay lower wages, because of (un)conscious prejudices that are not based on productivity differences (Arrow, 1971). Information asymmetry in labour markets means employers face hiring as an investment decision made under uncertainty because they cannot infer a worker’s productivity before hiring (Spence, 1973). Furthermore, there can be non-financial rewards to doing certain jobs, for instance the phenomenon of public service motivation where individuals with strong desire for public service are attracted to government jobs

(16)

(Carpenter et al., 2012). These complicate the earnings determination equation presented in the model.

Although the Becker-Tomes model is instructive and formally explains the causal mechanisms underpinning intergenerational mobility in human capital the literature review shows that the real-life transmission of human capital between generations is more complex and nuanced than the Becker-Tomes model allows. This must be considered especially with regard to interpreting the results of this paper and devising appropriate policy responses to the regional divide in Italy.

(17)

Research Design

Research Question

What is the relationship between intergenerational mobility in education and credit constraints faced by families in the regions of Italy?

Based on the Becker-Tomes model as the theoretical framework it would be expected that for those families that face credit constraints there will be less investment in children’s human capital accumulation because those families will have lower parental earnings and investment comes at the expense of their own consumption and leisure. This leads to the development of the following conceptual hypothesis:

If families are credit constrained they will invest less in the human capital of their child.

This relates to the North-South regional disparity as data from the EVS indicates that there is a higher incidence of credit constrained families in the South compared to the North. Therefore, if the hypothesis is correct then less will be invested in human capital accumulation in the South leading to a persistent difference in educational attainment with negative consequences for the South’s labour productivity and economic growth compared to that of the North.

Data Collection

The EVS can be accessed online via the GESIS data catalogue (EVS, 2016a). The survey is about the values of people throughout the European Union; Italy participated in the 1981, 1990, 1999, and 2008 waves of the EVS. Its selection method is a representative multi-stage random sample of the adult population, 18 years or older, with no upper age limit (EVS, 2016b). The survey is conducted via face to face interviews of a standardized questionnaire. The 2008 wave is used because it contains information on the education of the respondent, and the education of one of their parents, as well as regional information.

Operationalisation

Educational Attainment

The concept used in the theoretical framework is human capital with education being its largest component. The operationalisation of human capital, therefore, uses education as the best indicator of human capital. The dataset contains information on the highest type of educational qualification

(18)

that the respondents and one of their parents have achieved2_{. The level of educational attainment is} then converted into years of education because this makes the data type continuous and is a more intuitive and comprehendible measurement of education. The process by which the educational qualification is converted is by drawing on the minimum length of time required to officially complete the educational track denoted by the qualification, information on which is gathered from Barone and Schizzerotto (2008).

TABLE 4:CONVERTING COUNTRY SPECIFIC EDUCATIONAL ATTAINMENTS INTO YEARS OF EDUCATION

ISCED code (1 digit)

ISCED code (2 digit)

Name of programme (italian) Name of programme (english equivalent)

Years (minimum requirement)

0 0 Scuola materna Pre-school education

(or none)

0

1 1 Scuola elementare Primary school 5

2 2A Scuola secondaria di I grado /

scuola media

Lower Secondary School

8

3 3A Liceo / istitu tecnici / istitu

professionali (5 year)

General upper secondary education

13

3A Liceo artistico Artistic upper

secondary education 12 3C Istituti professionali (3 year) Vocational School (3

year)

11

4 4C Instruzione e formazione

tecnica superiore / alta

formazione artistica e musicale / scuole superiori per la mediazione linguistica, and others

Higher Technical Education and Training

14

5 5A Corso di laurea First degree 16

5A Corso di laurea specialistica Second degree 18

5A Aaster di primo e secondo

livello / corsi di specializzazione / corsi di perfezionamento

Master courses 17

5A (or 5B) Diplomi universitari e scuole dirette a fini speciali

University special courses

15 5B Academie di belle arti / istituti

superio per industrie artistiche

Higher education artistic programmes

16

6 6 Dottorato di ricerca Phd. 21

The advantage of measuring educational attainment in years of education is its intuitiveness. However, it does not measure the actual length of education which would include things like

2_{The International Standard Classification of Education (ISCED) codes provide an international reference for}

(19)

repeating years, breaks in studies, and optional extensions, but only the most direct and shortest duration of years required to complete that level of educational attainment. The expectation is that this measure is both a reliable and accurate indicator of educational attainment and a good

representation of human capital. The representativeness of the years of education in terms of ISCED codes seems consistent for one-digit ISCED codes (the first column in table 1). For the two-digit ISCED codes there is some variation with equivalent educational attainment outcomes such as 5A being possible in the range of 15 to 18 years. This operationalisation is used for both years of education of the respondent and that of their parents. The tables below show the frequencies by which

respondents’ and their parents completed their respective educational tracks. Attendance at the lower secondary level has been compulsory since the 1923 Gentile reform, which raised the compulsory age of education to 14, however participation only increased from the 1950s onwards and has become almost universal only since the 1970s (Barone & Schizzerotto, 2008). Table 5 shows that the South has higher drop-out rates of education than the North does, conversely a higher percentage of respondents in the sample for the South have completed tertiary education than the North. For all of Italy the majority of respondent’s parents have completed less than or equal to eight years of education, 78.8% in the South and 73.2% in the North.

TABLE 5:FREQUENCY TABLE FOR EDUCATIONAL ATTAINMENT BY REGION

Years of education

Child Parent

South North South North

0 7.3 2.5 25.3 15.1 5 11.2 11.6 29.9 37.1 8 23.4 21.9 23.6 21.0 11 5.3 8.3 3.1 4.2 13 34.9 36.6 11.9 16.2 14 2.9 2.8 1.0 1.0 15 1.2 2.8 0.0 0.2 16 1.5 2.3 0.0 0.1 17 1.7 3.2 0.2 0.8 18 10.0 7.9 4.8 4.0 21 0.4 0.2 0.0 0.2

The numbers displayed in the columns are all percentages, except the first column displaying years of education. There are 2.1% of observations missing for the child’s educational attainment in the south, and 2.5% for the south. For the parents’ educational attainment there are 9.6% of observations missing for both regions. The sum of the columns may not exactly equal 100 due to rounding.

(20)

Credit constraints

Credit constraints are operationalised as the response to the question, “parent(s) had problems replacing things”, with either “Yes” or “No” as possible answers (EVS, 2010 pp 83). The context of the question requires the respondent to think back to their circumstances at age 14, which introduces possible human error as this may have been long ago, and memories are not always reliable.

The operationalisation of credit constraint as the answer to the question "my parents had problems replacing broken things" with "yes" or "to some extent" is a notable weakness in the analysis.

TABLE 6:FREQUENCY TABLE FOR CREDIT CONSTRAINT BY REGION

Grown up in credit constrained family? North South

Yes 28.60% 33.30%

No 71.40% 66.70%

This is a response to the question: “parent(s) had problems replacing things” with respondents asked to think back when they were aged 14. The validity of this question as the operationalisation of credit constrained families is more fully considered in the research design section. 14.7% of responses are missing for the South, 14.2% of responses are missing for the North.

(21)

Methodology

The method of analysis utilised in this paper is regression analysis; three main models are analysed and discussed in the sections below.

Model 1: Intergenerational regression coefficient

The first model is a conventional univariate ordinary least squares (OLS) regression with years of education of the respondent (child) as dependent variable. Formally the equation is provided below:

EQUATION 7.

𝑌_{𝑖,𝑗,𝑘}𝑐ℎ𝑖𝑙𝑑= 𝛼𝑗,𝑘 + 𝛽𝑗,𝑘𝑋𝑖,𝑗,𝑘 𝑝𝑎𝑟𝑒𝑛𝑡 _{+ 𝜀}

𝑖,𝑗,𝑘

Where 𝑌_{𝑖,𝑗,𝑘}𝑐ℎ𝑖𝑙𝑑 is the years of education of respondent i, for region j, and age cohort k. 𝑋_{𝑖,𝑗,𝑘}𝑝𝑎𝑟𝑒𝑛𝑡 is the years of education of the parent of respondent i, for region j, and cohort k. 𝛼𝑗,𝑘 is the constant for

region j, and cohort k; 𝜀𝑖,𝑗,𝑘 is the error term for respondent i, region j, and cohort k. 𝛽𝑗,𝑘 is the estimate

of interest, it is referred to as the intergenerational regression coefficient.

The intergenerational regression coefficient shows the effect of one additional year of parental education on the child’s years of schooling. It is, therefore, indicative of the persistence of educational attainment between generations; the larger the intergenerational coefficient is, the less educational mobility there is, and vice versa.

Each cohort is a range of ten years because the intergenerational regression coefficient in education changes over time. The cohorts start at age 26 because earlier ages may still have incomplete education tracks. The descriptive statistics of each cohort by region are provided in table 5. The cohorts are based on the ages of the respondents because the ages of the parents are not provided in the EVS dataset.

The regression is run for each region to examine whether there are significant differences in educational mobility between the North and South of Italy. If there are no differences in intergenerational educational persistence, then the expectation would be for the intergenerational regression coefficients to be the same between different regions for the same age cohorts (𝛽𝑛𝑜𝑟𝑡ℎ,𝑘 =

(22)

in the South than in the North (𝛽𝑛𝑜𝑟𝑡ℎ,𝑘 < 𝛽𝑠𝑜𝑢𝑡ℎ,𝑘) because the effect of the higher incidence of credit

constraints in the South and the lower earnings in the South will outweigh the effect of the lower returns to education in the North and the slightly lower generosity of Northern parents as discussed in the theoretical framework. If this is borne out that would indicate that the circumstances of family background play a larger role in determining educational attainment in the South than it would in the North.

Model 2: Testing for the effect of credit constraints

The second model tests for the effect of coming from a family that is credit constrained. The operationalisation of credit constrained is as a dummy variable that is included in the multivariate regression equation below:

EQUATION 8.

𝑌𝑖,𝑗𝑐ℎ𝑖𝑙𝑑= 𝛼𝑗+ 𝛽𝑗𝑋𝑖,𝑗 𝑝𝑎𝑟𝑒𝑛𝑡

+ 𝜀𝑖,𝑗

The regression equation is not run for the age cohorts because missing responses from the question entail that there are not enough observations for reliable regression analysis. The descriptive statistics for the regressions are presented in appendix 1.

The expectation is that credit constraints have a significant negative effect on years of education. Meaning, that the results indicate a coefficient with a negative sign and statistical significance. Interpreting the dummy is not as intuitive as the intergenerational regression coefficient. Therefore, in model three, the credit constraint dummy variable is used as the selection variable for the regressions, meaning the size of the differences in the intergenerational regression coefficient can be measured. There is no expectation of differences in the magnitude of the estimated coefficients between the regions because although credit constrained families are more prevalent in the South there is no reason to expect that being credit constrained has a different size effect between regions.

Model 3: Intergenerational elasticity of education with and without credit constraints

EQUATION 9.

𝑌𝑖,𝑗,𝑙𝑐ℎ𝑖𝑙𝑑 = 𝛼𝑗,𝑙+ 𝛽𝑗,𝑙𝑋𝑖,𝑗,𝑙 𝑝𝑎𝑟𝑒𝑛𝑡

(23)

Where the subscript 𝑙 denotes the credit constraint (𝑙 = 0,1), when 𝑙 is equal to one that denotes the respondent is from a credit constrained family. The measure used to analyse the effect of coming from a credit constrained family is by taking the difference between the regression coefficient of not coming from a credit constrained family from the regression coefficient of coming from a credit constrained family (𝛽𝑗,0− 𝛽𝑗,1). For the same reasons as in the second model, there are not adequate observations

to also conduct an analysis of the separate age cohorts, as can be seen in the descriptive statistics in appendix 1.

The expectation is that respondents from credit constrained families will have higher intergenerational elasticity of education than those from non credit constrained families. Relating this to theoretical framework this is because credit constrained families cannot invest the optimal amount in their child’s human capital accumulation because they cannot borrow unlimitedly against their child’s future earnings. Instead with no access to credit their child’s human capital accumulation will depend on whether parental earnings in combination with parental generosity to their child can match the optimal investment in their child’s human capital in the perfect access to capital markets case. This is likely to fall short because it is especially families where parental earnings are low that likely face binding credit constraints.

Robustness checks

Several other model specifications are tested as robustness checks, these are presented briefly in the appendices. In appendix 2 the results are provided for not including Central Italy in the “North” macro-region. In appendix 3 the results are run using weighted least squares (instead of OLS) using the EVS’s weighting variable as weight. In appendix 4 only male respondents are used because of the historical differences in labour force participation of women potentially affecting their educational attainments.

(24)

Results

TABLE 7:RESULTS FOR MODEL 1 Cohort

All of Italy North South

Constant Beta Constant Beta Constant Beta

26-75 7.681 0.501 8.123 0.445 7.019 0.593 (0.169) (0.020) (0.213) (0.025) (0.276) (0.034) 26-35 10.949 0.265 11.401 0.231 10.449 0.302 (0.406) (0.039) (0.550) (0.052) (0.606) (0.060) 36-45 9.201 0.425 9.471 0.362 8.990 0.489 (0.380) (0.048) (0.515) (0.065) (0.568) (0.070) 46-55 9.200 0.401 9.472 0.366 8.655 0.477 (0.345) (0.046) (0.427) (0.055) (0.590) (0.084) 56-65 7.408 0.562 8.097 0.485 6.403 0.669 (0.424) (0.064) (0.537) (0.077) (0.693) (0.126) 66-75 4.961 0.603 5.259 0.593 4.343 0.593 (0.443) (0.073) (0.536) (0.089) (0.815) (0.135)

All results are significant at p<0.001, standard errors are in parentheses.

The intergenerational regression coefficient shows the effect of one additional year of parental education on the child’s years of schooling. On average, there are large effects observed across the regions and cohorts although two notable observations are clear: (1) there is a persistent decline in the intergenerational regression coefficient over time, (2) the intergenerational regression is persistently larger in the South than in the North for the same age cohorts. The findings show that the circumstances of family background play a larger role in determining educational attainment in the South than it does in the North of Italy. This matches the prediction made in the methodology (𝛽𝑛𝑜𝑟𝑡ℎ,𝑘 < 𝛽𝑠𝑜𝑢𝑡ℎ,𝑘) for every single age cohort, except the 66-75 age cohort where they are equal

(𝛽𝑛𝑜𝑟𝑡ℎ,66−75 = 𝛽𝑠𝑜𝑢𝑡ℎ,66−75). Putting aside the latter age cohort these result fit with other findings

relating to the North-South divide, Ballarino et al. (2014) report that social background influences the transition between secondary education levels with a negative impact in South Italy. The importance of parent’s educational attainment was also found by Brunello & Checchi (2005), their results also found these to be have a larger effect in regions and cohorts with poorer family background. The constant term reflects the increases in minimum education in Italy over time. The coefficient also measures some of these cross-generational differences, including the increased participation in schooling and increases in minimum compulsory attendance of education (Checchi et al., 2013). This limits the interpretability of the coefficient as an accurate measure of a child’s educational attainment conditional on their parent’s educational attainment.

(25)

TABLE 7:RESULTS FOR MODEL 2

Variable All of Italy North South

Constant 8.327*** 8.760*** 7.331***

(0.210) (0.256) (0.363)

Years of education parent 0.474*** 0.409*** 0.589***

(0.022) (0.027) (0.038)

Credit constrained family dummy -0.915*** -1.128*** -0.420

(0.231) (0.281) (0.401)

Standard errors are in parentheses, asterisks denote significance, *** denote p<0.001, the significance of the credit constrained dummy variable in the regression for the South is 0.295.

The results of model 2 show that there is a significant effect of coming from a credit constrained family for the results of all of Italy, including the North. This matches results from Lee & Lee (2020) and numerous theoretical expectations (such as Herrington, 2015; Solon, 2014) as well as those of the theoretical framework employed in this paper. However, results of the Southern macro-region are insignificant and, therefore, inconclusive. The South has a smaller coefficient for the credit constraint dummy than the North; its standard error is almost as large as the coefficient. For a more interpretable measurement of the effect of coming from a credit constrained family are the results from the third model.

TABLE 8:RESULTS FOR MODEL 3

Region (1) (2) (3) All of Italy 0.633 0.474 0.159 (0.049) (0.025) North 0.532 0.375 0.157 (0.060) (0.030) South 0.663 0.561 0.102 (0.082) (0.041)

(1) is the regression when only credit constrained families are selected, (2) is the regression when only non credit constrained families are selected, (3) is the difference between column 1 and 2, (𝛽𝑗,0− 𝛽𝑗,1). The results show

the intergenerational regression coefficient, all results are significant at p<0.001, standard errors are in parentheses.

The results of the third model show that the intergenerational regression coefficient is larger for the regressions of respondents from credit constrained families. The absolute numbers of the intergenerational regression coefficients remain larger for the South in both the regressions yet, the difference between coming from a credit constrained family or not is relatively larger in the North, as

(26)

shown by the third column in the table. This fits with the findings from the second model, which showed the effect of the credit constraint dummy variable to be statistically insignificant in the South.

Using the dummy as a selection variable shows the difference between intergenerational persistence in the education of credit constrained and non credit constrained families. The results confirm the hypothesis that the intergenerational elasticity of education is larger in credit constrained families. A search of the relevant literature did not find a paper that employs a similar method of testing the effect of coming from a credit constrained family that these results could be compared to. The relative effect to the rest of the macro-region of coming from a credit constrained family is larger in the North with a 0.157 difference between the elasticities of credit and non credit constrained families. However, the

absolute number remains higher in the South of Italy in both groups. This means that although the

North of Italy does not have as strong a correlation as the South between parental and their child’s education levels, the effect of being from a credit constrained family relative to the rest of the region’s population is larger.

(27)

Discussion

The modelling of intergenerational mobility generally follows that of a markov process between generations that regress to the mean over time depending on the size of the average persistence between generations (Becker & Tomes, 1986; Solon, 1999). The results of this paper show a large role of familial educational background on the educational outcomes of the next generation, importantly the role of family background has been consistently decreasing over time meaning that regression to the mean will occur more rapidly and will continue doing so if progress is maintained. The results also showed regional differences, with family background playing a more important role in the South compared to the North. If regression to the mean occurs at a national level then it can be expected that eventually North and South will experience full convergence in educational attainment. However, if regression to the mean occurs at regional level then convergence between North and South may not converge. Furthermore, recent theoretical work by (Becker et al., 2018) show that regression to the mean may not occur due to for instance skill-biased technological change that lead to a convex relationship between high earning families and their children’s earnings. The results from this paper indicate that regression to mean would occur more slowly for credit constrained families, which are generally lower earning, than the rest of population. Based on the theorical model this would be the result of suboptimal investment in their children’s human capital acquisition. The result is that persistence in educational attainment is likely to require more generations to regress to the population mean.

The differences between North and South in educational attainment may be even larger than the results of this paper reflect. The drop-out rate from education is strongly influenced by the quality of the school system and the higher rate drop out rate in the South may reflect a lower quality

education system (Hanushek et al., 2008). If the quality of the educational system is worse in the South than the same qualification may actually not represent the same quality of human capital (Odoardi & Muratore, 2019).

Limitations

The lack of observations that can be utilised for the third model means that the analysis cannot be conducted as granularly and, resultantly, does not capture how the effect of a credit constrained family, in the regions, change over the long term. Therefore, convergence between the regions cannot be observed by this analysis. Furthermore, any heterogeneity of the results within the regions cannot be examined.

(28)

Results for Italy may well be unique, and not very applicable to other contexts. Furthermore, the relationship between education and intergenerational mobility in earnings may operate differently in the country. This, for example, may be seen with surprising findings in Italy, namely a sizable intergenerational correlation remains after controlling for human capital in a paper by Barbieri et al. (2020), which is consistent with findings by Corak (2013) of Italy being an outlier with regards to higher returns to schooling being associated with lower intergenerational earnings mobility as shown visually in the graph below. Furthermore, the findings by Franzini and Raitano (2019) highlight that skill premia account for only a small part of rising intra-generational wage inequality in Italy since the 1990s.

Figure 1: “Higher Returns to Schooling are Associated with Lower Intergenerational Earnings Mobility” (Corak 2013, pp 10)

Source: Corak (2013)

This figure illustrates the puzzling finding that the returns to education are very low in Italy and, yet, it still has been estimated to have a very high intergenerational earnings elasticity.

This paper has remarked upon the crucial role of human capital in earnings determination and the related intergenerational transmission of earnings. Nevertheless, it is not realistic to claim that all pay differences are a result of differences in productivity (and by extension worker’s human capital). Furthermore, well established labour market frictions complicate the relationship between human capital and wage determination that could complicate any inferences drawn from the results of this paper for the wider social mobility literature.

(29)

A common limitation for studies involving human capital and credit constraints, including this paper, is the imposition of ad hoc constraints on credit (Lochner and Monge-Naranjo 2012). The operationalisation of credit constraints is problematic for any strong inferences on the subject. The answer to the question: “my parents had problems replacing broken things" with "yes", or "to some

extent" is simply uncertain whether it could be reliable or accurate. As explained in the research design

section, the operationalisation is functionally useful for the analysis but any interpretation of the results of this paper must be aware that the accuracy of the role of credit constraints in the intergenerational mobility in educational attainment is contingent on the accuracy and reliability of the operationalisation of it.

The linearity of the relationship between parental educational attainment and that of their children is not a given. Noting however that linearity is an approximate fit for the intergenerational earnings in mobility for other immobile countries like the US and UK and not for Nordic countries (Bratsberg et al. 2007). The theoretical framework utilised in this paper illustrates that the relationship could be concave if poorer families are more credit-constrained than richer families. Therefore, the model may have been mis-specified leading to the OLS regression not being the best unbiased estimation method.

There is a significant amount of ‘noise’ that is captured by the intergenerational regression coefficient,

β. Changes in the mean education between parents’ generations and that of their children are

captured alongside institutional changes such as increases in the age of compulsory education.

Further avenues of research

The actual variable of interest from a policy perspective is the child’s education conditional on their parent’s. To that end, the regression coefficient is not the most accurate, or the most useful, measurement for the development of policy. Checchi, Fiorio, and Leonardi (2013) developed a decomposition of the correlation coefficient of education “to account for the different intergenerational mobility of subgroups of the population” (page 229). This is a natural extension for the analysis that has been conducted in the paper and would provide a more detailed and accurate picture of the relationship between parental education and their children’s education in Italy and, thereby, allowing better tailored policy development.

The phenomenon captured by the Great Gatsby Curve of greater intra-generational inequality leading to greater inter-generational inequality is likely to occur both between and within countries. Chetty et al.'s (2014) findings for commuting zones across the US are consistent with the Great Gatsby Curve.

(30)

The phenomenon may be part of the explanation of persistent regional differences between North and South Italy.

The findings from Chetty et al. (2014) also pointed toward the role of family stability on intergenerational mobility in earnings (areas with higher levels of mobility are correlated with greater family stability). These could be used to develop a research design that either follows the research design of Chetty et al. (2014) for other countries or to examine the relationship between family stability and the intergenerational transmission of education. Chetty et al. (2014) find it to be a significant result using both micro and macro analysis (so at both the individual and community level). The EVS waves contain many variables relating to family life, including stability, that could be used to explore the role of family stability. Chetty et al. (2014) also measure whether children were raised by one or both parents as their variable for family stability, given the extent to which the EVS asks questions about family life this analysis could be extended to incorporate more aspects of family stability and, hopefully, becoming able to examine the causal mechanisms by which family stability affects intergenerational mobility in earnings and education.

An extension to the Becker-Tomes model in their 1986 paper “Human Capital and the Rise and Fall of Families” includes the effect of family size on intergenerational mobility in earnings. Their conclusion is that more children reduce the amount invested in the human capital of each one when investments must be financed by the family (i.e. credit constrained families). Therefore, it would follow that larger families (i.e. more children) at the bottom end of the income distribution that do not have access to perfect capital markets would experience lower intergenerational mobility in education (and earnings) relative to smaller credit constrained families. A search of the literature indicates that this hypothesis has not yet been empirically established and it would be worthwhile to establish this finding due to both the potential consequences for policy and the appropriateness of the Becker-Tomes model. The EVS contains detailed family information albeit the educational attainment of siblings is not included and, therefore, a different data source is likely to be required to carry out this research. A possible research design compatible with the EVS dataset would be a comparison of the intergenerational mobility in education of individuals from similar backgrounds but with different numbers of children. The data strictures are likely to be onerous requiring data on parental education, their children’s education, the number of children, and ideally each child’s educational attainment.

(31)

Policy Implications

The results of the analysis indicated that respondents from credit constrained families experienced a higher persistence of educational attainment between generations. Furthermore, there are large significant differences between the macro-regions of Italy. The inequality of opportunity that these results represent provides plentiful evidence that policy intervention is warranted. This paper next provides possible policies that could be implemented to provide regional convergence and reduce the intergenerational persistence in educational attainment.

The findings from the literature review point to several important aspects of investment in human capital. First, the timing of investments was found to be crucial with the first six years of life offering especially high returns of investment that are compounded by complementarities between early and late investments. Second, whilst early childhood was the best time to invest this is when parents generally face lower earnings and more binding credit constraints. Third, parents may not be aware of or estimate incorrectly the return to investing in their child’s human capital acquisition. These three findings inform the following policy recommendations.

High quality universal child-care is provided in the low intergenerational elasticity countries such as seen in Nordic countries and Canada. A study of the Norwegian system found it to lead to significant increased social mobility, especially for the lower and middle segment of the income distribution. The effect on later life earnings for children from high income families was negative as a result of the policy (Havnes & Mogstad, 2015). The Italian childcare system is currently a mix of private (subsided) and public childcare facilities with, on average, the most expensive childcare being found in the North of Italy and the cheapest in the South. In Italy, children younger than three mostly receive childcare from non-formal arrangements (family and friends) as seen by 74.9% being non-formal childcare arrangements in 2016 and 25.1% in formal childcare with more prevalent non-formal arrangements in the South and Islands of Italy (Bulgarelli, 2018). There are differences across regions in financing of institutions, requirements for employment at institutions, accreditation for childcare provision ages 0-3; between ages 3-6 these are run at the national level (Bulgarelli, 2018). Having established the impact that early age differences have on later life a useful policy to implement to improve social mobility and equalize it across regions is to move authority and governance of childcare provision of ages 0-3 to also be at the national level. This would be enacted with the aim of promoting care at equal levels between regions. Further subsidizing or removal of barriers to getting children into formal childcare would also benefit the lower income segment of the population and the Southern region and Islands that currently have lower levels of enrolment in formal childcare. This policy is on the one hand feasible because its

(32)

aim of equalising equality of opportunity is politically palatable, however, it is an expensive policy to implement. Informal childcare has the advantage of not further burdening the Italian government’s stressed finances and the opportunity costs involved may be low if the child carers are for instance grandparents who are out of the labour force. Nonetheless, the benefit of formal child care is the provision of roughly equal quality childcare which prevents locking in early life differences.

The Becker-Tomes model (1986) assumes approximate perfect information, specifically, parents know the return on investment of investing in their child’s human capital accumulation. The literature review presented a paper by Cunha, Culhane, and Elo (2013) that showed that this assumption does not hold true for (at least) a sample of socioeconomically disadvantaged African-American women who underestimated investments in their child’s development. As argued by the authors, correcting for mis-informed parental expectations can improve the outcomes of children. Policy intervention can alter expectations, for instance Field et al. (2009) found an uptake in iodized salt (prevents early onset brain damage) after parents learnt of its benefit. Improving information and awareness of parents about the returns of investing in their child’s human capital should lead to parents investing closer to the optimal amount into their children’s human capital accumulation.

The generosity of the education system is an important factor for intergenerational mobility in educational outcomes. As noted previously Italy’s public expenditure on education in 2017 was the second lowest in the OECD (OECD, 2020b). The theoretical framework discusses how public and private expenditure on education are substitutes. The more generous Italy’s education system is (𝑠𝑡−1) the

less circumstances of family background (through parental investment in education: 𝑥𝑡−1) will play

into educational outcomes. Increasing the public expenditure on education will reduce the role of family background in educational outcomes, and the differences between the regions as the South generally has less fortunate family origins.

Furthermore, the structure of the Italian welfare state reinforces the control of parents over their child’s choices by supporting older generations over younger generations due to the “familiastic” design of the welfare state (Di Giulio & Rosina, 2007). Changing the welfare state type to a more Nordic model that aims at universal coverage and a pivot from supporting older generations to ensuring support and investment and younger generations would much improve educational mobility and likely would lead to at least some regional convergence due to more generous support of poorer households. However, such a change to the Italian welfare state system may not reflect the values of the Italian population and their low trust in government running effectively (Andretta, 2018).