Equality of Opportunities for Young Italian Workers - Società Italiana di [PDF]

WORKING PAPER No 704

Società italiana di economia pubblica

febbraio 2016

 

EQUALITY OF OPPORTUNITIES FOR YOUNG ITALIAN WORKERS Francesco Scervini, IUSS Pavia Agnese Peruzzi, IRPET Enrica Chiappero, Università di Pavia

JEL Classification: D63, J21, J60 Keywords: Intergenerational equality of opportunity, Labour market, Job contracts, Inequality

società italiana di economia pubblica c/o dipartimento di scienze politiche e sociali – Università di Pavia

Equality of opportunities for young Italian workers∗ Francesco Scervini†1 Agnese Peruzzi2 Enrica Chiappero3 1

Istituto Universitario di Studi Superiori, Pavia, Italy Regionale Programmazione Economica della Toscana, Italy 3 University of Pavia, dsps, Italy

2 Istituto

Abstract The duality of the labour market in Italy is a dramatic social and economic issue. Not only the youth unemployment rate is one of the highest in Europe (39.7% in 2014), but also the quality of job contracts is very low for young, compared to elder cohorts. Moreover, Italy is perceived as a low-mobility country, where individual outcomes depend significantly on initial conditions and parental background. This paper exploits the heterogeneity of job contracts to investigate whether and how circumstances beyond individual control affect the labour market outcomes of young labour force, not only as the probability of employment, but also as the type of job contract achieved. Results confirm that the role of predetermined conditions is very strong, in particular gender and the region of origin, but also parental labour status. The relative effects and significances of circumstances changed their magnitude over time. Jel codes: D63, J21, J60 Keywords: Intergenerational equality of opportunity, Labour market, Job contracts, Inequality. ∗

The research leading to these results has received funding from the European Union’s Seventh Framework Programme for research, technological development and demonstration under grant agreement n. 320136 (Project Title ‘Social Innovation – Empowering the Young for the Common Good’, Project Acronym: SocIEtY). This article reflects the authors’ views and the European Union is not liable for any use that may be made of the information contained within. † Corresponding author. [email protected]

1

1

Introduction and related literature

The strong deterioration in the labour market situation for young people during the last few years is a common feature to many European countries, but this is particularly the case in Italy, where the crisis has exacerbated the country’s long-standing structural weaknesses (Garibaldi and Taddei, 2013). Indeed, labour market opportunities and conditions for youth, already gradually deteriorated, worsened further in the aftermath of the 2008-2009 recession, to the point that young people’s disengagement from the labour market has emerged as one of the most striking social and economic issues plaguing the Italian economy (OECD, 2015).1 The adverse trends of youth unemployment and employment rates occurred alongside a decline in the youth participation, as captured by the high levels of discouragement and inactivity among young Italians, with more than two million people aged 1529 (26% of the total in 2013) in the NEET (neither in education, employment or training) condition. This greater vulnerability, widely investigated in the economic literature (see for example Scarpetta et al., 2010; International Labour Office, 2013), is in part attributable to the Italian labour market peculiarities and macroeconomic conditions – both long-standing and recent. As for the peculiarities of the labour market, Italy underwent a significant process of deregulation over the past decades, as a way to enhance labour market flexibility and reduce unemployment levels. Overall, the reform process aimed at introducing gradual and partial changes of the institutional framework and consisted mainly of policies promoting flexibility ‘at the margin’ – attenuating employment protection legislation (for temporary contracts and other new contractual forms such as atypical contracts), while maintaining stringent existing rules on permanent contracts. More specifically, the Italian labour market deregulation, whose first antecedent dates back to 1984, was more effectively enhanced in the late 1990s and early 2000s through two major labour market laws: the so-called riforma Treu in 1997 and the riforma Biagi in 2003. While the riforma Treu introduced some kinds of temporary and atypical job contracts (Ichino et al., 2008) and provided incentives for part-time work, the riforma Biagi further deregulated the use of atypical work arrangements, such as temporary agency work (staff-leasing) and part-time work, and introduced new forms of atypical work arrangements such as on-call jobs (lavoro intermittente), job sharing and occasional work (lavoro a progetto). These reforms led to a severe issue of duality in the Italian labour market (Boeri 1

Youth employment rate declined from 25% in 2008 to 16.6% in 2013, while the unemployment rate peaked in January 2014 at the dramatic figure of 39.7% (Eurostat statistics).

2

and Garibaldi, 2007; Ballarino et al., 2014), where a large number of atypical contractual arrangements characterized by low social security protection coexist with standard employment contracts (Scarpetta et al., 2010; Chung et al., 2012) and where the costs of flexibility have been endured mainly by workers with atypical contracts, basically new entrants (Heyes, 2011; Berton et al., 2012). As for the macroeconomic conditions, the two major recent economic crises (the 2008-2009 financial crisis and the sovereign debt crisis in 2011) have worsened Italian long-standing and structural weaknesses, such as its weak growth2 and population ageing, thus producing a deep recession, with strong negative effects especially on the young generations (Bruno et al., 2014). Indeed, as widely recognized, youth tend both to be highly concentrated in sectors and industries particularly sensitive to business cycle fluctuations (Banerji et al., 2014), and to be disproportionately present among those holding part-time jobs and temporary or atypical contracts (OECD, 2010). As mentioned above, those represent the categories of workers which have suffered the highest cost from the flexibility introduced with the recent reforms. However, the overall figures on youth disengagement from the labour market mask significant internal differences, being labour market opportunities not evenly distributed among young people. Well-documented and marked it is indeed the segmentation of labour market opportunities across a set of important circumstances such as ethnicity, gender (see for example Crepaldi et al., 2014), social background (Checchi and Dardanoni, 2002; Checchi and Flabbi, 2013; Checchi and Peragine, 2010; Peragine and Serlenga, 2008), and territory (Spreafico et al., 2014). Those circumstances not only affect the level of opportunities and achievements in education (Ferreira and Gignoux, 2014; Woessmann, 2004), which is, according to the Human Capital theory (Becker, 1964; Mincer, 1974), the main determinant of labour market outcomes, but they have been found also to still play a substantial role in labour market outcomes, even after controlling for the level of education achieved, thus suggesting that inequality of opportunity persists beyond human capital (see for instance Bjorklund and J¨antti, 2009; Blanden, 2013; Guell et al., 2015; Pellizzari, 2010). Notwithstanding the varied disadvantages that the youth face in accessing labour market highlight the complex interrelation between personal, familiar and contextual circumstances, most of the existing literature investigating young people’s disengagement from the labour mar2

Economic growth has been sluggish in Italy for the past 15 years with rate surpassing 2% only once since 2001, primarily because of low productivity growth (European Commission, 2014)

3

ket, even when controlling for multiple factors, focuses on the role of specific disadvantages. However, as growingly recognized, it is not a single factor or a set of separate factors, but the combination between different and multilayered circumstances that affect individuals’ positions and may determine individual differences in terms of opportunities (Barnes et al., 2011). This analysis moves from recognizing the lack of a deep understanding of the extent to which different disadvantages interact and exacerbate each other and aims at examining how the combination and interplays between personal characteristics and a plurality of familiar and contextual factors affect young people’s opportunities in entering the labour market. The paper makes several contributions to the empirical literature. First, since our focus is on young individuals, we broaden the measure of labour market exclusion beyond wages, by utilizing other characteristics of job quality mainly related to the job contract and the reason of non-employment. Wage is indeed not the best indicator to assess youth’s labour market conditions for a twofold consideration: on the one hand, at the early stage of careers, incomes are typically low and their distribution is squeezed; on the other hand, there is a higher share of young individuals with no income, in particular because of unemployment – and this is especially true for Italy – or other non-employment conditions. Second, by concentrating our search only on individuals at the beginning of their working career, our methodology allows us to take full account of the role of background circumstances in affecting youth labour market opportunities. Indeed, later in individual’s working life the relationship between initial disadvantages and labour market outcomes tends to be moderated by previous job career and therefore less affected by circumstances beyond individual’s control. Third, different from the previous literature, we estimate the effects of several different circumstances potentially affecting the outcome at the same time. In addition, we also investigate the interrelations between these circumstances, allowing for different potential ‘joint effects’ of disadvantages. Finally, by using data covering a long time span, we are able to examine whether the recent labour market reforms and macroeconomic crises have amplified or attenuate the sensitivity of young people’s labour market opportunities to both specific disadvantages and to their interactions. The paper is organized as follows: Section 2 discusses the methodological assumptions lying behind the theoretical framework adopted. Section 3 describes the dataset and the main variables operationalised. The empirical analysis and the results are discussed in Section 4. The last section summarises the main findings and concludes.

4

2

Theoretical background

The theoretical background we employ builds on the classic literature on equality of opportunity. The standard framework assumes that the individual outcome, y, depends on the individual effort exerted, e, and a set of circumstances, c, according to a generic function: y = Y (c, e)

(1)

where Y (·) is the cumulative distribution function. In order to estimate the effects of the circumstances, we need to make some assumptions, whose meaning and implications are very effectively rationalised in two papers by Lefranc et al. (2009), Ramos and Van de gaer (2015a), and Ramos and Van de gaer (2015b), while Pignataro (2012) provides a wider discussion on this topic. First, we need to assume that, once we control for circumstances, the distribution of effort is the same across individuals, that is: e = G (e | c). This implies that part of the effort might be actually correlated to circumstances, for instance, a ‘bad example’ from the parents may affect effort, but the part of effort unrelated to circumstances has the same distribution across types. Second, we need to assume that the outcome is strictly increasing in effort, that is: ∂Y /∂e > 0. Last, we should rule out any other element affecting the outcome, other than circumstances and effort. These assumptions are somehow less restrictive than it seems at a first glance: the last assumption is innocuous, as far as we can set freely the elements of the vector of circumstance, c, that is every source of inequality that we want to compensate for.3 The second assumption is theoretically acceptable, implying that any increase of effort translates in a greater outcome. Unfortunately, this is not always true in the real world, since the outcomes are typically discrete (e.g. income does not increase smoothly for every – unobservable – change of effort). Finally, the first assumption states that the distribution of effort is the same across groups of individuals characterised by the same circumstances, ruling out any systematic innate differences across groups.4 If these assumptions hold, then equality of opportunity is satisfied if and only if Y (· | c) = Y (· | c0 ) , ∀ (c, c0 ). In this framework, inequality of oppor3

According to equality of opportunity theories, one should compensate for any source of inequality outside the individual choice set, for instance gender, place of birth, or parental background. Whether or not we should compensate for luck is more debatable, and there is no general agreement on this. 4 This assumption excludes the presence of ‘genetic’ differences in the effort exerted across groups of individuals, since any institutional and social environment is considered as a circumstance.

5

tunity can be defined as the difference of outcome due to the differences in circumstances, that is: D (· | c, c0 ) = Y (· | c) − Y (· | c0 ) , ∀ (c, c0 )

(2)

where D is a measure of inequality of opportunities and D = 0 if Y (· | c) = Y (· | c0 ). In order to study the effect of some circumstances on the inequality of opportunities, in principle we would need to estimate how every element of the vector c affects the whole distribution of the outcome, but such analysis is possible only graphically. For instance, graph 1 shows the distribution of the outcome for two different types of individuals: according to the ‘traditional’ literature (see for instance Lefranc et al., 2009, p.1199, for an analogous graphical representation), it clearly emerges that one group is more disadvantaged than the other. However, in this contribution, since we consider several circumstances at the same time, we are not able to compare different types graphically: first, the comparison should involve hundreds of types (that is, individuals with the same observable circumstances), that is hundreds of distributions; second, in order for the distributions to be compared, the number of individuals within every cell must be sufficiently high: even if we have a very large sample, the size of some cells could be too small to efficiently estimate the shape of the distribution. Because of these reasons, instead of a graphical analysis, we employ regression techniques. In particular, we estimate the effects of several circumstances on individuals’ outcome. More formally, in a model where the outcome y is regressed on the circumstances c according to a model y = cb + ε, every coefficient bi represents the average effect of a circumstance ci on the outcome. If b = 0, then it means that on average y is unaffected by the circumstances, or equality of opportunity. Opposite, if b 6= 0, there is a statistical difference between the outcomes of individuals characterised by different circumstances. Consequently, it is possible to estimate the effect of a specific circumstance on equality of opportunity by taking the partial derivatives of y with respect to the corresponding element of c using econometric techniques. Moreover, it is also possible to investigate the joint effects and interactions of different circumstances. Overall, with respect to the standard empirical literature on equality of opportunity, we are not able to determine how circumstances affect the complete distribution of the outcome, but we can determine the average shift of the distribution considering not only several circumstances simultaneously, but also the interacted effects of such circumstances. In the following of the paper we follow exactly this approach, by regressing the exclusion from the labour market on a set of individual predetermined 6

0

20

Exclusion (0-100 scale) 40 60 80

100

Figure 1: Distribution of exclusion from the labour market for two types

0

.2

.4

.6

Less disadvantaged

.8

1

Most disadvantaged

Source: lfs, Italian survey, 1992-2013. ‘Less disadvantaged’ refers to males living in the North before 1999 with parent not unemployed; ‘Most disadvantaged’ refers to females living in the South or Islands after 2007 with parent unemployed.

characteristics, and their interactions, that turn out to be the most disadvantageous for young people.

3

Definitions, data and measures

In order to analyse the labour market outcomes of young individuals5 we need to exploit the variability both in terms of labour market itself and in terms of educational achievement, that represent the main source of selection outside the labour market for young individuals.6 The European Union Labour Force Survey (lfs from now on) is a large sample survey providing detailed information on labour status, job characteristics, and education at individual level. The sample unit of lfs is the household, so that it is possible to retrieve some parental background information for the young still living with their parents. For these reasons, we believe it is the most appropriate source of data to investigate the labour market outcomes of the young. Lfs is released by Eurostat at yearly basis 5

Eurostat defines as ‘young’ all individuals aged 15-24 and all the official statistics referred to young take individuals in this age bracket as the reference group. 6 On the issue of selection in the labour market, see Eurostat (2015b).

7

since 1983. However, because of data consistency, we limit our analysis to the period 1992-2013.7 Such a long time-span allows us splitting the sample in sub-periods, according to major socio-economic events with a significant effect on labour market outcomes for the young, such as the financial crisis in 1992 or the great depression starting in 2008. Each wave of lfs is made up by four quarterly surveys and each unit may be interviewed for several quarters. However, unfortunately, it is not possible to exploit the panel dimension of the survey because of the lack of a consistent household id over years. We select individuals according to several criteria: first, we include only individuals aged 15-24; second, we exclude household heads and all young not living in their parental household. Indeed, being the household the sample unit, it is not possible to retrieve information on parental background for those young who do not live in the same household as their parents. Moreover, young household heads may have different decision processes with respect to education and labour market, that might bias the estimates. Finally, they only represent a very low share of young in Italy: only 46,160 over 674,750 young (about 6.8%) do not live in their parental household. Third, we retrieve information on parents and households characteristics. Table 1 summarises the overall sample and the working sample sizes over time. Our estimates refer to a large sample of 628,590 young, belonging to 460,704 households, consisting in turn of 1,783,990 individuals. The empirical analysis relies on three sets of variables described hereafter: i) the exclusion from the labour market, that is the outcome; ii) a set of circumstances, that are pre-determined individual conditions possibly affecting the outcome; iii) a set of disadvantages, that are the subset of circumstances that result to influence negatively the outcome. In details, our measure of exclusion from the labour market is derived by information on the labour market status, such as whether the individual is employed, unemployed, or not looking for a job; if employed, whether she has a full-time or a part-time job, and whether she has a permanent or a temporary job ; in the latter case, the duration of the temporary job contract; if unemployed, whether she is long-term (more than 12 months) or short-term unemployed. Table 2 shows the composition of labour market positions of youngsters in Italy between 1992 and 2013. As far as there are several labour market statuses, there are also several possible ways to order these statuses to get a reliable individual measure of exclusion from the labour market. Preliminarily, we exclude all the young 7

2013 is the last available year as on July 2015. This category includes individuals who are classified as ‘inactive’ according to the main variable ilostat, but whose status cannot be retrieved by the variable mainstat because of inconsistency. Therefore, we classify them as inactive, without better specifications. 8

8

Table 1: Sample size by year Year 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Total

Entire sample Working sample Individuals Households Individuals Households 201,007 72,909 74,555 18,834 200,550 73,118 72,779 18,544 198,935 73,170 71,266 18,160 203,434 75,374 71,241 18,150 202,432 75,321 70,056 17,807 201,541 75,294 68,314 17,416 201,835 75,346 67,217 17,184 200,625 75,173 65,404 16,677 199,367 75,275 63,829 16,273 196,236 75,216 60,687 15,471 194,041 75,202 57,750 14,785 192,359 75,294 56,333 14,508 172,264 66,441 49,016 12,529 422,157 164,597 120,031 30,740 413,946 164,822 114,836 29,594 408,695 165,758 111,246 28,785 404,811 166,219 107,115 27,889 399,828 167,634 103,371 27,101 395,085 167,811 99,779 26,255 391,770 168,383 97,004 25,634 363,173 155,426 90,485 23,992 364,751 155,840 91,676 24,376 6,128,842 2,439,623 1,783,990 460,704 Source: lfs, Italian survey, 1992-2013.

Young 27,175 26,515 25,911 25,698 25,050 24,276 23,768 22,974 22,394 21,245 20,046 19,505 16,936 41,747 40,054 38,736 37,455 36,190 34,909 34,002 31,773 32,231 628,590

who are student, disabled, pensioners, or in compulsory military service, since these individuals do not participate in the labour market by definition. Then, we order the different statuses and assign a score ranging from 0 to 100 to every condition. Of course, this methodology is intrinsically arbitrary and subject to the criticism that the results depend on this classification. However, even if we show in the paper only one possible classification, our results are robust not only to different, arbitrary classifications, but also to a fuzzy data methodology where scores are ‘endogenously’ assigned on the basis of relative shares (see Chiappero-Martinetti, 2006, for more details). Moreover, it is straightforward to order the employment statuses according to the main dimensions considered: the first is the labour market condition, namely whether individuals are employed, unemployed or inactive, the former being the less excluded from the labour market and the latter the most excluded. Second, among employed, we order workers according to two additional criteria: full-time versus part-time and permanent versus temporary jobs, where full-time are less excluded than part-time workers and permanent are less excluded than temporary. Third, among temporary workers we intuitively assume that the longer is the job contract, the less excluded is the worker. The arbitrariness issue arises according to three decisions: first, whether

9

Table 2: Labour market outcomes (individuals, 15-24) Variable Employed of which Employee of which Part-time Full-time and of which Temporary job of which Less than 3 months 3-12 months 12-24 months More than 24 months Missing Permanent job Self-employed Family worker Unemployed of which Less than 6 months 6-12 months More than 12 months Missing Inactive of which Student Disabled Pensioners Not looking for a job Other inactive8 Military Total Source: lfs, Italian

Obs. 138,486

Share 22.03

116,637

84.22

15,839 100,798

13.58 86.42

39,618

33.97

6,003 12,522 5,310 9,535 6,248 77,019 12,693 9,156 59,112

9.40

18,117 10,898 28,927 1,170 424,372

30.65 18.44 48.94 1.98 67.51

365,683 3,182 6 13,639 41,862 6,620 628,590 survey, 1992-2013.

10

15.15 31.61 13.40 24.07 15.77 66.03 9.17 6.61

86.17 0.75 0.00 3.21 9.86 1.05 100

the duration of job contract is more or less important than the part-time condition; second, where to place self-employed and family workers in the scale; third, what is the distance between different conditions. As for the first, we tried both options and results are virtually identical. For these reason, we decide to show only the case in which job duration is more relevant. Indeed, in Italy job duration is associated to other characteristics, such as unemployment and illness benefits, parental leaves, paid holidays and so on, all related to the concept of exclusion. Regarding the second issue, we classify self-employed and family workers just one step below full-time, permanent workers. Indeed, even if these jobs have an intrinsic uncertainty with respect to the less excluded category, self-employed and family workers can be considered as better off with respect to most excluded workers. Also in this case, as a robustness check we replicate the analysis excluding these two categories of workers from the analysis and results are virtually unchanged (see figure 2 discussed hereafter). Finally, with respect to the relative distance in the scale, we decide to assign about one third (0-30) to permanent workers, self-employed and family workers; the central part of the scale (40-75) to temporary workers; and the remaining part (80-100) to unemployed. With respect to this issue, arbitrariness is unsolvable. However, as a robustness check, we also run a multinomial logit model, that allows ignoring the relative distance among individuals, and results confirm our scale.9 Tables 3 and 4 show the exclusion scores assigned to every category and the share of young for each category over time and the overall mean exclusion, while figure 2 represents the average time trend and figure 3 its geographical distribution. In table 3 we observe a clear decreasing trend of permanent job. As discussed in section 1, several job market deregulations have successively increased the number of atypical contracts. Since these reforms applied only to new job contracts, they have affected mainly young workers, whose share of permanent jobs halved from about 45% to slightly more than 20%, including also part-time workers. Symmetrically, the share of temporary contracts increased dramatically: more than 15% of the sample have a job contract lasting less than 12 months, and 5% lasting less than 3 months, making them very vulnerable and ‘excluded’ from the labour market. Finally, there is not a clear trend in the share of unemployed and discouraged young, whose share is constantly very high, ranging from about 25% in 2007 to more than 40% in 1997. Unsurprisingly, the labour market exclusion is much larger in the Southern regions, on average twice as large as in the North-Eastern regions. 9

Considering the differences in the estimation methodologies, results are comparable and coherent to those reported in section 4 and available upon request.

11

Table 3: Labour market outcomes and exclusion (individuals, 15-24) Labour market status

Exclusion Shares of score 1992 1997 2002 Full-time, permanent 0 42.53 36.91 37.10 Self-employed 15 4.69 5.33 5.25 15 7.22 5.13 6.55 Family worker Part-time, permanent 30 1.48 1.97 2.20 Full-time, temporary, > 24 months 40 0.42 0.57 2.63 45 0.09 0.07 0.33 Part-time, temporary, > 24 months Full time, temporary, 12-24 months 50 1.49 1.91 3.36 55 0.20 0.22 0.47 Part time, temporary, 12-24 months 55 2.47 2.65 2.89 Full time, temporary, unknown dur. Part time, temporary, unknown dur. 60 0.94 0.78 0.93 Full time, temporary, 3-12 months 60 1.64 2.32 2.74 65 0.61 0.61 0.97 Part time, temporary, 3-12 months Full time, temporary, < 3 months 70 0.83 0.61 0.99 75 0.25 0.26 0.31 Part time, temporary, < 3 months Short term unemployed 85 14.80 14.38 11.92 Unknown-term unemployed 90 0.42 0.51 0.30 95 11.75 17.14 14.42 Long-term unemployed Discouraged 100 8.15 8.62 6.63 Source: lfs, Italian survey, 1992-2013.

workers 2007 31.94 7.90 2.62 4.97 7.36 0.88 2.21 0.35 1.52 0.73 7.26 2.18 3.75 1.31 12.24 0.46 8.01 4.32

2012 16.94 6.85 2.08 5.59 7.34 1.57 1.26 0.35 1.16 1.01 7.26 3.28 3.34 1.66 17.32 0.80 16.42 5.77

Total 33.46 6.06 4.68 3.26 3.69 0.58 1.97 0.31 2.10 0.87 4.30 1.52 1.97 0.77 14.15 0.50 13.13 6.67

Notes: For missing values on the duration of the job contract or of unemployment, average values have been assigned.

40

Labour market exclusion 45 50 55

60

Figure 2: Exclusion from the labour market over time

1992

1995

1998

2001

Year

All individuals

2004

2007

2010

2013

No self-employed and family workers

Source: lfs, Italian survey, 1992-2013.

Table 4: Summary statistics on exclusion from the labour market Variable Obs Mean Std. Dev. Min Max Exclusion 211,237 44.65 39.76 0 100 Source: computations on lfs, Italian survey, 1992-2013.

12

Figure 3: Exclusion from the labour market by region

54.5 45.6 36.8 27.9

-

63.4 54.5 45.6 36.8

Source: lfs, Italian survey, 1992-2013.

The second set of variables exploited in the empirical analysis includes information on individual circumstances and parental background. The former are individual characteristics outside the choice set of the young worker, such as gender, age, region of residence, migration status, degree of urbanization of the living area.10 The latter includes parental age, education, and labour market status, household size and composition, work intensity.11 In order to include also single-parent households in the sample, we define parental education as the level achieved by the most educated parent; parental labour market status as the status of the father, if present, or the mother, if the father is not in the household; parental age, as the age of the eldest parent. Tables 5 and 6 show the descriptive statistics of our sample. Third and last, starting from the set of individual circumstances and parental background, we identify a subset of conditions that turn out to be positively correlated to exclusion from the labour market. These conditions can be defined as disadvantages, as far as they have a negative impact on 10

We assume that the households’ choice about the region of residence and the degree of urbanization does not depend on the job outcomes of the young individuals still living in the household. 11 Eurostat defines ‘the work intensity of a household [as] the ratio of the total number of months that all working-age household members have worked during the income reference year and the total number of months the same household members theoretically could have worked in the same period’ (Eurostat, 2015a).

13

Table 5: Summary statistics on individual circumstances Variable Age 15-19 20-24 Total Gender Male Female Total Migration status At least one parent migrant Native Missing (pre-2004) Total Education Lower secondary or less Upper secondary Tertiary or above Missing Total Degree of urbanization Densely populated area Intermediate area Thinly populated area Total

Obs.

Share

321,174 307,416 628,590

51.09 48.91 100

326,852 301,738 628,590

52.00 48.00 100

34,985 309,048 284,557 628,590

5.57 49.17 45.27 100

358,736 256,971 12,203 680 628,590

57.07 40.88 1.94 0.11 100

219,494 253,385 155,711 628,590

34.92 40.31 24.77 100

Variable Region of residence North Piemonte Valle d’Aosta Lombardia Trentino - Alto Adige Veneto Friuli - Venezia Giulia Liguria Emilia Romagna Centre Toscana Umbria Marche Lazio South and Islands Abruzzo Molise Campania Puglia Basilicata Calabria Sicilia Sardegna Total

Obs.

Share

242,062 40,747 8,219 68,506 32,560 34,687 13,470 13,755 30,118 99,561 30,898 11,928 14,429 42,306 286,967 14,807 16,468 59,766 46,452 21,084 39,298 65,526 23,566 628,590

38.51 6.48 1.31 10.90 5.18 5.52 2.14 2.19 4.79 15.84 4.92 1.90 2.30 6.73 45.65 2.36 2.62 9.51 7.39 3.35 6.25 10.42 3.75 100

Source: lfs, Italian survey, 1992-2013.

Table 6: Summary statistics for households Variable Household composition Parental age Household size Members aged > 65 Members aged < 14 Variable Parent’s education Lower secondary or less Upper secondary Tertiary or above Missing Total

Variable Single parent household Non single parent household 628,590 51.19 Total 628,590 4.07 628,590 6.80% Parent’s labour status 628,590 34.37% Employed Unemployed Obs. Share Retired Other out of labour force 324,157 51.57 Missing 230,102 36.61 Total 73,003 11.61 Variable 1,328 0.21 628,590 100 Work intensity Source: lfs, Italian survey, 1992-2013. Obs.

Mean

14

Obs. 75,120 553,470 628,590

Share 11.95 88.05 100

484,394 22,514 74,923 46,757 2 628,590 Obs. 626,369

77.06 3.58 11.92 7.04 0.00 100 Mean 56.27%

the labour market outcome and they are outside individual control. Anticipating some of the results, the disadvantages depend on gender, age, parental unemployment, household composition – in particular in terms of young members and single parenthood – and population density. Moreover, we consider as disadvantages living in Southern regions and Islands. Finally, we decide to split the whole period in four sub-periods, on the basis of the major economic and institutional changes affecting the labour market conditions for the young discussed above, namely 1992-1997, before the riforma Treu; 1998-2002, before the riforma Biagi; 2003-2008, before the economic crisis; 2009-2013, during the economic crisis.

4

Methods and Results

The empirical analysis consists of two different steps. The first is devoted to retrieve the disadvantage dimensions starting from a wider set of predetermined characteristics, as described in the previous section; the second investigates whether and how the disadvantages – individually and interacted – have some effects on the dependent variable. As for the first, the labour market outcomes of young depend at first on their educational choices. Therefore, we need to deal with the issue of selection, that may bias the estimates on the labour market outcomes. The most straightforward methodology is the so-called Heckman selection model (see Heckman, 1979, for the original model), that consists of a system of equations, as follows: ( y ∗ if h∗ > 0 y= (3) − if h∗ ≤ 0 ( 1 if h∗ > 0 h= (4) 0 if h∗ ≤ 0 where y is the exclusion from the labour market, that is observed only if the individual is in the labour market or similar conditions,12 while it is not observed if she is in education or disabled. This selection variable is identified by h, that is a binary variable taking value 1 if individual is in the labour market and 0 otherwise. In turn, the two latent variables y ∗ and h∗ are 12

Because of easy of exposition, from now on we consider ‘in the labour market’ all the types in table 3, even if it is not the proper definition.

15

identified as follows: y ∗ = cb1 + ε1 h∗ = xb2 + ε2

(5) (6)

where the vector x includes all the circumstances c affecting the exclusion from the labour market, plus at least one variable (the so-called exclusion restriction) that does not influence the outcome, but only the selection variable. Once controlling for parental labour market status, parental education and age are the best exclusion restrictions, since they are highly correlated to young education decisions (that are very much dependent on parental education and age), but not to exclusion from the labour market (once the young decides to enter the labour market, its outcome is hardly affected by these two variables). Depending on the assumptions on the error terms ε1 and ε2 , it is possible to compute the b’s by maximum likelihood or by a two-step procedure (see for instance Cameron and Trivedi, 2008, pp.541-557). In the paper we show results from the former, that allows to cluster standard errors at household level, but two-step estimations provide results not significantly different.13 Table 7 reports the first stage of the Heckman selection model of three different specification. In the first, we include information on individuals (age and gender), parents (education, labour status, age), households (work intensity, population density in the area of residence, size and composition, region of residence), and we control for the year. In the second specification, we run exactly the same model, the only difference being the presence of migration status, and then a shorter period (2004-2013). Finally, in order to check whether the (slight) differences between the two models are due to the presence of migration status as a regressor or to the time span, we run the same specification as in model 1, but only on the period 2004-2013. Since estimations in the second and third specifications are virtually identical, we decide to drop migration status and to exploit the longer time span.14 Looking at the marginal probabilities (columns 2, 4, 6), results confirm prior expectations: young aged 20-24 and males are more likely to be in the labour market than in education, and the same is true for young with non employed parents (the effect being larger for unemployed than for retired) and with single parents, likely because of the need for additional income in 13

It is worthy highlighting that, by focusing only on predetermined conditions, we solve the potential reverse causality issue: by definition, being predetermined, the circumstances in c cannot be influenced by the outcomes. 14 The results in the following of the paper are unaffected by the presence of migration status unless stated otherwise.

16

the household. Moreover, the probability of being in the labour market is higher in non-densely populated areas and for larger households, while it is decreasing with parent’s education and age, and household work intensity. In this case, it is likely that elder and more educated employed parents have a higher income, increasing the probability of achieving a tertiary education degree for the offspring. Moreover, beside the effect of income, there is also a cultural effect that might lead the offspring of more educated parents to acquire more education. The second stage of the Heckman selection model is reported in table 8, where the three specifications correspond to those in table 7. Together with the Heckman model, we report the estimations of an analogous single stage linear regression model. The possible differences between coefficients are due to the selection process, that may bias the results. Each coefficient represents the average variation of exclusion from the labour market of individual in each condition, with respect to the reference group. In the following, we mainly refer to the results in model 1, with the complete time span. Results from different periods are discussed in the following. The gender gap is dramatically high. Over the 0-100 scale, being a woman increases the exclusion from the labour market by about 8 p.p., once controlling from the selection of women in education, and by 13 p.p. otherwise. Another very relevant negative circumstance is the labour status of the parents. Having the father (mother) unemployed increases the exclusion from the labour market by about 11 p.p. with respect to an employed parent, while the effect of other statuses is smaller and estimated on relatively few observations. A similar effect, even if smaller, is that of work intensity: since it is not a categorical variable, the coefficient is the difference in labour market exclusion between a household where all members in working age are not employed at all (work intensity equals to 0) and a household where all members work full-time (work intensity 1). Living in densely populated areas, that is mainly cities, increases exclusion by 5 p.p., an effect twice as large as the one of a single parental household. The effect of household size is negligible and not always significant. Finally, once controlling for selection, the effect of age is not as large as expected: it is 14 to 16 p.p. in the ols model and 3.5 to 7 in the Heckman selection model, the difference being due to the fact that younger individuals are much more likely to select in education. Overall, younger workers are more disadvantaged, and this disadvantage is much larger in the 2004-2013 period (models 2 and 3). The analysis of the circumstances affecting the exclusion from the labour market allows us identifying the main individual disadvantages, that is – as discussed in the previous section – exactly the set of the most relevant circumstances at individual level. In order to choose which circumstances 17

Table 7: Effects of circumstances on exclusion from the labour market - First step (selection) Reported beta Age 15-19 Age 20-24 Female Male Par.educ.: up to Lower sec. Par.educ.: Upper sec. Par.educ.: Tertiary or above Par.status: Employed Par.status: Unemployed Par.status: Student Par.status: Retired Par.status: Other Elder parent’s age Work intensity Densely populated area Intermediate area Thinly populated area Household size Household members 14 or younger Single parent household

Model 1 (1) (2) Coeff. Margin b/se b/se ref. ref.

Region fixed effect Year fixed effect χ2 Obs.

Model 3 (5) (6) Coeff. Margin b/se b/se ref. ref.

1.171*** 0.004 ref.

0.351*** 0.001 ref.

1.287*** 0.006 ref.

0.361*** 0.001 ref.

1.286*** 0.006 ref.

0.361*** 0.001 ref.

0.285*** 0.004 ref.

0.079*** 0.001 ref.

0.361*** 0.005 ref.

0.092*** 0.001 ref.

0.362*** 0.005 ref.

0.093*** 0.001 ref.

-0.492*** 0.004 -1.099*** 0.008 ref.

-0.143*** 0.001 -0.275*** 0.002 ref.

-0.475*** 0.006 -1.083*** 0.011 ref.

-0.130*** 0.002 -0.254*** 0.002 ref.

-0.473*** 0.006 -1.079*** 0.011 ref.

-0.130*** 0.002 -0.253*** 0.002 ref.

0.322*** 0.011 -0.180 0.114 0.035*** 0.008 -0.036*** 0.009 -0.001*** 0.000 -0.108*** 0.008 ref.

0.092*** 0.003 -0.047 0.029 0.010*** 0.002 -0.010*** 0.002 -0.000*** 0.000 -0.030*** 0.002 ref.

0.295*** 0.015 0.023 0.148 0.027** 0.011 -0.074*** 0.012 -0.003*** 0.001 -0.084*** 0.012 ref.

0.079*** 0.004 0.006 0.038 0.007** 0.003 -0.019*** 0.003 -0.001*** 0.000 -0.022*** 0.003 ref.

0.298*** 0.015 0.019 0.149 0.024** 0.011 -0.076*** 0.012 -0.003*** 0.001 -0.091*** 0.012 ref.

0.080*** 0.004 0.005 0.038 0.006** 0.003 -0.019*** 0.003 -0.001*** 0.000 -0.023*** 0.003 ref.

0.052*** 0.005 0.089*** 0.006 0.080*** 0.003 -0.061*** 0.004 0.153*** 0.007

0.014*** 0.001 0.024*** 0.002 0.022*** 0.001 -0.017*** 0.001 0.042*** 0.002

0.047*** 0.007 0.077*** 0.008 0.063*** 0.004 -0.079*** 0.006 0.144*** 0.009 0.097*** 0.009 -0.952*** 0.035 Yes Yes 70411.7*** 343294

0.012*** 0.002 0.020*** 0.002 0.016*** 0.001 -0.020*** 0.002 0.037*** 0.002 0.025*** 0.002

0.046*** 0.007 0.076*** 0.008 0.065*** 0.004 -0.078*** 0.006 0.146*** 0.009

0.012*** 0.002 0.019*** 0.002 0.017*** 0.001 -0.020*** 0.002 0.037*** 0.002

-0.924*** 0.035 Yes Yes 70338.9*** 343294

Yes Yes

At least one parent migrant Constant

Model 2 (3) (4) Coeff. Margin b/se b/se ref. ref.

-0.980*** 0.025 Yes Yes 128230.3*** 625107

Yes Yes 625107

Yes Yes 343294

343294

Source: lfs, Italian survey, 1992-2013. Significance of coefficients: *** p ≤ 1%, ** p ≤ 5%, * p ≤ 10%. Dependent variable: being in education (0) or in the labour market (1). Heckman models estimation technique: Maximum likelihood. Standard errors clustered at household level to account for possible correlation between brothers.

18

Table 8: Effects of circumstances on exclusion from the labour market Second step Estim. method Age 15-19 Age 20-24 Female Male Par.status: Employed Par.status: Unemployed Par.status: Student Par.status: Retired Par.status: Other Work intensity Densely populated area Intermediate area Thinly populated area Household size Household members 14 or younger Single parent household

Model 1 (1) (2) Heckman OLS b/se b/se ref. ref.

Model 2 (3) (4) Heckman OLS b/se b/se ref. ref.

Model 3 (5) (6) Heckman OLS b/se b/se ref. ref.

-3.554*** 0.389 ref.

-13.355*** 0.195 ref.

-6.988*** 0.541 ref.

-16.500*** 0.274 ref.

-6.956*** 0.540 ref.

-16.625*** 0.274 ref.

-7.913*** 0.183 ref.

-12.882*** 0.162 ref.

-6.628*** 0.259 ref.

-10.887*** 0.224 ref.

-6.561*** 0.259 ref.

-10.842*** 0.224 ref.

11.410*** 0.475 -11.552** 4.802 -1.679*** 0.281 -2.546*** 0.392 -4.453*** 0.361 ref.

7.619*** 0.441 -7.936 4.880 -1.851*** 0.277 -1.930*** 0.375 -2.879*** 0.348 ref.

10.773*** 0.622 -4.634 6.295 -0.485 0.409 -1.565*** 0.519 -3.717*** 0.492 ref.

7.901*** 0.595 -5.083 6.162 -0.211 0.406 -0.081 0.502 -2.370*** 0.478 ref.

10.903*** 0.622 -4.878 6.287 -0.676* 0.408 -1.617*** 0.519 -3.873*** 0.492 ref.

8.043*** 0.595 -5.472 6.150 -0.496 0.405 -0.143 0.502 -2.558*** 0.478 ref.

-4.444*** 0.210 -4.760*** 0.254 -0.160 0.108 0.642*** 0.193 2.255*** 0.290

-4.769*** 0.204 -5.469*** 0.245 -0.361*** 0.102 1.725*** 0.182 1.893*** 0.276

-3.116*** 0.283 -4.002*** 0.332 -0.459*** 0.148 2.730*** 0.254 1.400*** 0.370

57.696*** 0.748

-3.091*** 0.283 -3.929*** 0.332 -0.569*** 0.148 2.648*** 0.254 1.244*** 0.370 3.492*** 0.384 60.528*** 0.944

-2.750*** 0.289 -3.353*** 0.343 -0.557*** 0.155 1.529*** 0.266 2.093*** 0.385

39.755*** 1.023 0.253*** 0.012 3.584*** 0.002 Yes Yes

-2.736*** 0.289 -3.316*** 0.343 -0.635*** 0.155 1.474*** 0.266 1.971*** 0.385 2.434*** 0.391 44.050*** 1.301 0.256*** 0.016 3.542*** 0.003 Yes Yes

43.885*** 1.299 0.259*** 0.015 3.543*** 0.003 Yes Yes

60.585*** 0.945

At least one parent migrant Constant athrho lnsigma Region Year f.e. R-squared F test χ2 Obs.

20897.4*** 625107

Yes Yes 0.177 965.0*** 209991

6909.5*** 343294

Yes Yes 0.138 441.6*** 100166

6839.5*** 343294

Yes Yes 0.138 450.0*** 100166

Source: lfs, Italian survey, 1992-2013. Significance of coefficients: *** p ≤ 1%, ** p ≤ 5%, * p ≤ 10%. Dependent variable: exclusion from the labour market. Heckman models estimation technique: Maximum likelihood. Standard errors clustered at household level to account for possible correlation between brothers.

19

including in the set of disadvantages, we follow two criteria: first of all, the magnitude of the effect and the relevance from a policy perspective; second, the parsimony in terms of types, since a minimum number of individuals within the same type – that is, with the same set of disadvantages – is necessary to precisely estimate the effects. Combining these two criteria, we select as disadvantages the following circumstances: being in the age 1519, being a woman, having the parent unemployed,15 living in a city (that is, where the population density is defined as ‘high’), in a household with a single parent, or with at least one member younger than 15. Moreover, relying on the literature and on the timing of the labour market reforms, we split the sample in two geographical areas (North and Centre, South and Islands) and in four periods (1992-1997, 1998-2003, 2004-2007, 2008-2013), in order to study whether the relative importance of disadvantages changed over time and across regions. Table 9 shows the effects of the subset of disadvantages on the exclusion from the labour market in the whole period (column 1) and in the four subperiods. Several interesting results emerge by looking at the time trends: first of all, the effect of age became increasingly important over the period. Second, the importance of the two main sources of disadvantage, gender and region, decreased significantly over time. Unfortunately, this trend is due to the fact that exclusion from the labour market worsened for men and for young living in Northern and Central regions (the overall trend can be inferred by figure 2 and by the value of the constant terms in the table of estimations). However, the reduction of the gender and regional gap still represents an improvement from an equality of opportunity perspective. Finally, also the role of population density declined, while the presence of young members in the household increased its relevance, likely due to the decline of childcare support from public and private institutions. The analysis of the interactions between disadvantages may help to understand how the average effects distribute across individuals. Table 10 reports all the pairwise interactions between all variables, where each 2x2 cell originates from a regression analogous to column (1) of table 9, the only difference being that two sources of disadvantages are interacted in every regression. Being all dummy variables, the meaning of the interaction is not the usual one: taking as an example the top-left cell, individuals without the two relevant disadvantages (that is, men aged 20-24) are the reference group. Men in the 15-19 age bracket have a disadvantage score 3.0 p.p. higher than the reference group, while women in the 20-24 age bracket experience an exclu15

It is worthy to recall that the variable refers to the father, if present in the household, and to the mother otherwise (see section 3).

20

Table 9: Effects of disadvantages on exclusion from the labour market - By period Period Estim. method Disadv.: 15-19 years Disadv.: woman Disadv.: parent unemployed Disadv.: density of population Disadv.: single parent Disadv.: young members Disadv.: South and Islands Constant athrho lnsigma Year f.e. χ2 Obs.

All sample (1) Heckman b/se 2.532*** 0.416 7.646*** 0.188 15.018*** 0.445 5.376*** 0.188 2.204*** 0.252 0.479** 0.218 18.112*** 0.209 20.790*** 0.485 0.271*** 0.013 3.599*** 0.002 Yes 16976.7*** 627260

1992-1997 (2) Heckman b/se -1.211 0.866 9.308*** 0.361 16.480*** 0.968 7.605*** 0.367 1.814*** 0.509 -0.655 0.442 24.363*** 0.443 12.768*** 0.643 0.284*** 0.030 3.647*** 0.005 No 6041.4*** 154623

1998-2003 (3) Heckman b/se 1.859** 0.815 9.274*** 0.381 15.850*** 0.951 7.385*** 0.384 0.889 0.563 -0.161 0.468 26.670*** 0.431 14.943*** 0.576 0.185*** 0.024 3.600*** 0.003 No 6458.6*** 128604

2004-2007 (4) Heckman b/se 4.112*** 0.805 6.893*** 0.402 12.397*** 1.055 2.598*** 0.393 2.722*** 0.538 0.415 0.442 12.682*** 0.424 18.904*** 0.516 0.305*** 0.023 3.564*** 0.005 No 1517.6*** 137473

2008-2013 (5) Heckman b/se 7.012*** 0.755 6.014*** 0.346 15.972*** 0.678 2.269*** 0.348 2.984*** 0.424 1.497*** 0.387 9.265*** 0.366 29.269*** 0.523 0.251*** 0.021 3.556*** 0.004 No 1957.2*** 206560

Source: lfs, Italian survey, 1992-2013. Significance of coefficients: *** p ≤ 1%, ** p ≤ 5%, * p ≤ 10%. Dependent variable: exclusion from the labour market. Heckman models estimation technique: Maximum likelihood. Standard errors clustered at household level to account for possible correlation between brothers.

21

sion 7.9 p.p. higher than the reference group. Women in the age 15-19 have an exclusion score of 9.7, that is statistically different at 1% level from the sum of the two single disadvantages,16 meaning that the joint effect of being a young woman is lower than the two individual effects of being a woman and being young. On the contrary, the joint effect of being young and having the father/mother unemployed is higher than the individual disadvantages. The negative effect of gender is attenuated by the presence of any other disadvantages, apart from the geographical area. Indeed, being a woman is the only condition reinforcing the disadvantage of living in the South. Opposite, being young and having younger household members seems to actually reduce the disadvantage of living in the South. Living with a single parent has a strong negative effects on younger individuals (the joint effect is twice as large as the two individual effects) and those living in cities – likely due to weaker family ties and the lack of childcare – while it has virtually no interactions with other disadvantages. A younger household member has a positive effect on young workers and on women, and a small reinforcing effect if a parent is unemployed. Living in a city substantially attenuates the disadvantage of being a woman and being in the 15-19 age bracket. Finally, being young with a parent unemployed reinforces the two single disadvantages, while being young and woman moderates them. From a policy perspective, these information are useful in order to determine not only the most disadvantaged categories, but also the efficacy of policies targeted to some groups, that can be affected by the presence of other, related disadvantages. To this goal, we repeat the analysis focusing only on the last years, after the 2008 crisis. Table 11 shows the same interactions only for the period 2008-2013. While the interacted effects of being aged 15-19 are substantially unchanged, being a woman not only is less detrimental on average, but it is also not reinforced by living in the South, suggesting a – still insufficient – improvement of the female labour market discrimination in those regions. Having a parent unemployed is not moderated by any other disadvantage (apart from gender). The positive effect of younger household members on young individuals is confirmed in the last period, while there are no significant interactions with gender and parent unemployed. As the relative disadvantage of living in the South decreased, its interactions changed: not only having a parent unemployed or single have no statistical interactions with living in the South, but living in a city became less detrimental after the crisis. Finally, what is still highly relevant is the 16

As reported in the table footer, the asterisks report the significance of the χ2 test of equality between the sum of the two single disadvantages and the interaction between them.

22

23

Woman Yes

Par. unemp. No Yes

Pop. density No Yes No

Sing. par Yes

Young memb. No Yes

Dis.: Woman No ref. 3.0 x x Yes 7.9 9.7∗∗∗ x x − Dis.: Par. unemp. No ref. 2.4 ref. 7.8 x x Yes 14.4 18.8∗∗ 16.3 20.9∗∗∗ x x + − Dis.: Pop. density No ref. 2.9 ref. 8.7 ref. 15.5 x x Yes 5.6 7.4∗∗ 6.6 12.3∗∗∗ 5.4 19.7 x x − − Dis.: Sing. par. No ref. 2.2 ref. 7.8 ref. 15.1 ref. 5.2 x x Yes 1.7 6.3∗∗∗ 2.6 9.5∗− 2.2 17.0 1.8 8.1∗∗ x x + + Dis.: Young. memb. No ref. 3.7 ref. 7.8 ref. 14.4 ref. 5.4 ref. 2.1 x x Yes 1.7 1.5∗∗∗ .8 7.8∗∗ .4§ 16.8∗∗ .6 5.7 .4 3.1 x x − − + Dis.: South-Islands No ref. 4.3 ref. 6.9 ref. 16.0 ref. 5.1 ref. 2.7 ref. 1.1 Yes 19.1 18.6∗∗∗ 17.3 26.2∗∗∗ 18.2 32.7∗− 17.9 23.7∗+ 18.3 19.6∗∗ 18.4 18.1∗∗∗ − + − − Source: lfs, Italian survey, 1992-2013. Dependent variable: exclusion from the labour market. All regressions include as controls year fixed effects and all the non-interacted disadvantages. Every cell shows the three coefficients related to each of the single disadvantages and the interaction between the two disadvantages. All coefficients are statistically different from 0 at 10% significance level apart from those marked with the § symbol. Significance attached to the interacted terms refers to a χ2 test of equality between the sum of the two singe disadvantages and the interacted term (*** p ≤ 1%, ** p ≤ 5%, * p ≤ 10%) and the sigh in the subscript shows whether the difference is positive or negative. Standard errors clustered at household level to account for possible correlation between brothers.

No

Table 10: Effects of interacted disadvantages - 1992-2013

Age 15-19 No Yes

reinforcing negative effect of living in a city with a single parent.

5

Conclusions

Young individuals are the most affected by characteristics outside their direct control. Indeed, parental features and the state of nature are found in the literature to determine several aspects of individuals’ life. In this paper we investigate how predetermined conditions impact on the labour market status of young individuals in Italy in the last two decades. Building on the previous literature, we analyse not only several predetermined conditions jointly – age, gender, urbanisation, region, household composition, parental education, working status of household members, migration status – but also the pairwise interactions among them. These conditions, or circumstances, affect the labour market outcome in two distinct directions: first, they shape the probability of individuals to leave the formal education at secondary eduction instead of achieving a college degree; second – once individuals enter in the labour market – they influence significantly the outcome of the labour market. The results of the present analysis confirm and extend the previous literature. Expectedly, gender and region of residence turn out to be the most important factors driving the labour market outcomes, together with having a father unemployed, that was never investigated before and seems to play a very relevant role. Other relevant factors are living in densely populated areas, with younger household members, with a single parent, and age. The relative weight of these conditions change both over time and according to other disadvantage characteristics, as determined by the interaction effects analysis. From a policy perspective, there are two possible kinds of intervention: one aims at reducing ‘ex-ante’ the negative effects of the disadvantages, for instance easing the access of women to the labour market or expanding the access to higher education; the other is focused to compensate individuals ‘ex-post’ `a-la Roemer on the basis of their predetermined conditions (see Fleurbaey and Peragine, 2013; Ooghe et al., 2007, for an extensive discussion on their differences). Both interventions should take into careful account the relative importance of the conditions and their interactions, in order to focus primarily on the most relevant, or least acceptable, ones.

24

25

Woman Yes

Par. unemp. No Yes

Pop. density No Yes No

Sing. par Yes

Young memb. No Yes

Dis.: Woman No ref. 9.1 x x Yes 6.4 14.0∗∗ x x − Dis.: Par. unemp. No ref. 8.7 ref. 6.5 x x Yes 13.9 22.1 15.8 17.3∗∗∗ x x − Dis.: Pop. density No ref. 9.1 ref. 7.5 ref. 14.5 x x Yes 3.4 10.9∗∗ 4.9 7.8∗∗∗ 3.2 15.6 x x − − Dis.: Sing. par. No ref. 8.3 ref. 6.4 ref. 13.8 ref. 2.7 x x Yes 2.1 12.2∗+ 3.0 7.9∗∗ 2.4 16.1 1.5 6.8∗∗∗ x x − + Dis.: Young. memb. No ref. 10.2 ref. 6.3 ref. 13.6 ref. 3.3 ref. 2.3 x x Yes 2.9 8.0∗∗∗ 1.7 7.6 1.5 15.9 1.8 4.3 1.5 4.4 x x − Dis.: South-Islands No ref. 10.8 ref. 6.4 ref. 14.1 ref. 4.6 ref. 2.4 ref. 2.8 Yes 10.9 15.3∗∗∗ 10.1 15.9 9.9 23.5 11.0 11.7∗∗∗ 9.9 12.2 10.5 10.3∗∗∗ − − − Source: lfs, Italian survey, 2008-2013. Dependent variable: exclusion from the labour market. All regressions include as controls year fixed effects and all the non-interacted disadvantages. Every cell shows the three coefficients related to each of the single disadvantages and the interaction between the two disadvantages. All coefficients are statistically different from 0 at 1% significance level apart from those marked with the § symbol. Significance attached to the interacted terms refers to a χ2 test of equality between the sum of the two singe disadvantages and the interacted term (*** p ≤ 1%, ** p ≤ 5%, * p ≤ 10%) and the sigh in the subscript shows whether the difference is positive or negative. Standard errors clustered at household level to account for possible correlation between brothers.

No

Table 11: Effects of interacted disadvantages - 2008-2013

Age 15-19 No Yes

References Ballarino, G., M. Braga, M. Bratti, D. Checchi, A. Filippin, C. Fiorio, M. Leonardi, E. Meschi, and F. Scervini (2014). “Italy: How labour market policies can foster earnings inequality.” In: Changing Inequalities & Societal Impacts in Rich Countries. Ed. by B. Nolan, W. Salverda, D. Checchi, I. Marx, A. McKnight, I. T¨oth, and H. van de Werfhorst. Oxford: Oxford University Press. Chap. 16. Banerji, A., S. Saksonovs, H. Lin, and R. Blavy (2014). Youth Unemployment in Advanced Economies in Europe: Searching for Solutions. IMF Staff Discussion Note. International Monetary Fund. Barnes, M., R. Green, and A. Ross (2011). Understanding Vulnerable Young People: Analysis from the Longitudinal Study of Young People in England. Research Report. Department for Education. Becker, G. (1964). Human capital: A theoretical and empirical analysis, with special reference to education. New York: Columbia University Press. Berton, F., M. Richiardi, and S. Sacchi (2012). The Political Economy of Work Security and Flexibility. Italy in Comparative Perspective. Bristol: The Policy Press. Bjorklund, A. and M. J¨antti (2009). “Intergenerational income mobility and the role of family background.” In: Oxford handbook of economic inequality. Ed. by W. Salverda, B. Nolan, and T. Smeeding. Oxford: Oxford University Press. Blanden, J. (2013). “Cross-country rankings in intergenerational mobility: a comparison of approaches from economics and sociology.” In: Journal of Economic Surveys 27(1), pp. 38–73. Boeri, T. and P. Garibaldi (2007). “Two-tier Reforms of Employment Protection Legislation: A Honeymoon Effect.” In: Economic Journal 117, pp. 357–385. Bruno, G., M. Choudhry, E. Marelli, and M. Signorelli (2014). “Youth unemployment: key determinants and the impact of crises.” In: Disadvantaged workers. Empirical evidence and labour policies. Ed. by M. Malo and D. Sciulli. Springer International Publishing, pp. 121–148. Cameron, A. C. and P. K. Trivedi (2008). Microeconometrics Using Stata. Stata press. Checchi, D. and V. Dardanoni (2002). “Mobility Comparisons: Does using different measures matter?” In: Research on Economic Inequality 9, pp. 113–145. Checchi, D. and L. Flabbi (2013). “Intergenerational mobility and schooling decisions in Italy and Germany: the impact of secondary school track.” In: Rivista di Politica Economica VII-IX, pp. 7–60. 26

Checchi, D. and V. Peragine (2010). “Inequality of opportunity in Italy.” In: Journal of Economic Inequality 8, pp. 429–450. Chiappero-Martinetti, E. (2006). “Capability approach and fuzzy set theory: description, aggregation and inference issues.” In: Fuzzy set approach to multidimensional poverty measurement. Ed. by A. Lemmi and G. Betti. Springer Verlag. Chung, H., S. Bekker, and H. Houwing (2012). “Joung people and the postrecession labour market in the context of Europe 2020.” In: European Review of Labour and Research 18(3), pp. 301–317. Crepaldi, C., F. Pesce, and M. Lodovici (2014). Social and Employment Situation in Italy. Discussion Note. Directorate General for Intarnal Policies, European Parliament. European Commission (2014). European Economy: Macroeconomic Imbalances Italy 2014. Brussels: Publications Office of the European Commission. Eurostat (2015a). Glossary:Persons living in households with low work intensity. url: http://ec.europa.eu/eurostat/statistics-explained/ index.php/Glossary:Persons_living_in_households_with_low_ work_intensity (visited on 2015-03-11). — (2015b). Glossary:Youth unemployment rate. url: http://ec.europa. eu/eurostat/statistics- explained/index.php/Glossary:Youth_ unemployment_rate (visited on 2015-07-15). Ferreira, F. and J. Gignoux (2014). “The measurement of educational inequality: achievement and opportunity.” In: World Bank Economic Review 28(2), pp. 210–246. Fleurbaey, M. and V. Peragine (2013). “Ex Ante Versus Ex Post Equality of Opportunity.” In: Economica 80(317), pp. 118–130. Garibaldi, P. and F. Taddei (2013). Italy: A dual labour market in transition. Employment Working Paper 144. International Labor Organization. Guell, M., J. Rodriguez-Mora, and C. Telmer (2015). “The Informational Content of Surnames, the Evolution of Intergenerational Mobility and Assortative Mating.” In: The Review of Economic Studies 82(2), pp. 693– 735. Heckman, J. J. (1979). “Sample Selection Bias as a Specification Error.” In: Econometrica 47(1), pp. 153–161. Heyes, J. (2011). “Flexicurity, employment protection and the jobs crisis.” In: Work Employment and Society 25(4), pp. 642–657. Ichino, A., F. Mealli, and T. Nannicini (2008). “From temporary help jobs to permanent employment: what can we learn from matching estimators and their sensitivity?” In: Journal of Applied Econometrics 23, pp. 305–327.

27

International Labour Office (2013). Global employment trends 2013: Recovering from a second jobs dip. Geneva: International Labour Office. Lefranc, A., N. Pistolesi, and A. Trannoy (2009). “Equality of opportunity and luck: Definitions and testable conditions, with an application to income in France.” In: Journal of Public Economics 93(11-12), pp. 1189– 1207. Mincer, J. (1974). Schooling, earnings, and experience. New York: Columbia University Press. OECD (2010). Off to a good start? Jobs for youth. Paris: OECD. — (2015). OECD Economic Surveys: Italy 2015. Paris: OECD. Ooghe, E., E. Schokkaert, and D. Van de gaer (2007). “Equality of opportunity versus equality of opportunity sets.” In: Social Choice and Welfare 28, pp. 209–230. Pellizzari, M. (2010). “Do friends and relatives really help in getting a good job?” In: Industrial and Labor Relation Review 63(3), pp. 494–510. Peragine, V. and L. Serlenga (2008). “Higher Education and Equality of Opportunity in Italy.” In: Inequality and Opportunity: Papers from the Second ECINEQ Society Meeting. Ed. by J. Bishop and B. Zheng. Research on Economic Inequality. Emerald. Chap. 8, pp. 67–97. Pignataro, G. (2012). “Equality of opportunity: Policy and measurement paradigms.” In: Journal of Economic Surveys 26(5), pp. 800–834. Ramos, X. and D. Van de gaer (2015a). “Approaches to Inequality of Opportunity: Principles, Measures, and Evidence.” In: Journal of Economic Surveys in press. — (2015b). Measurement of Inequality of Opportunity Based on Counterfactuals. IZA Discussion Papers 9582. IZA. Scarpetta, S., A. Sonnet, and T. Manfredi (2010). Rising Youth Unemployment During the Crisis: How to Prevent Negative Long-term Consequences on a Generation? Social, Employment and Migration Working Papers 6. OECD. Spreafico, A., A. Peruzzi, and E. Chiappero-Martinetti (2014). “Youth unemployment in the EU15: regional and national inequalities throughout the 2008 economic crisis.” In: Economics and Business Letters 3(3), pp. 177– 190. Woessmann, L. (2004). How Equal Are Educational Opportunities? Family Background and Student Achievement in Europe and the United States. Working Papers 1162. Cesifo.

28