How does the place of residence affect young people’s disengagement?

Young people are one of the most vulnerable segments of the population in the labour market. This evidence is due to many personal and macro-economic factors and applies to all countries worldwide. However, the young people's disadvantage strongly differs across countries and, in some cases, also within the countries. In this paper, we aim at analyzing the influence that the place of residence exerts on the propensity to the NEET (young individuals not in employment, education, or training) status in Italy, the country that, in Europe, shows by many years the highest share of NEETs, with an evident internal variability. We applied a multilevel model using macro-data with deep territorial detail: the regional, the provincial, and even the municipality level. Results highlight the significant effect on the NEET phenomenon of factors linked to the territory, including the share of enterprises, the innovative content of the economy, and the quality of the education system. Investing in education, even in adults’ continuing education, can significantly reduce the share of NEETs, especially in the most depressed areas of the South of Italy.


Introduction
Young people are among the most vulnerable segments of the population in the labour market for several reasons, including mainly their lack of job experience and their lack of experience in job finding (Pastore et al. 2021).In times of crisis, they are usually the most hit category of workers because they often receive fixed-term or temporary contracts that cannot be renewed, but even for those working with an indefinite contract, because, in the case of dismissals, employers typically tend to use the last-in-first-out practice (the last hired are the first to be fired).
These facts concern all countries worldwide, but the level of young people's disadvantage in comparison to their adult peers strongly differs across countries.The ratio between the youth (15-24 years) and the adult (25-64 years) unemployment rates is the indicator usually used to measure the youth disadvantage.Across European countries, in 2014, it ranged from 1.6 in Germany to 4.7 in Luxembourg.Italy showed the highest value just after Luxembourg and Romania, with 3.9, indicating therefore a stronger penalization, in particular, for Italian young people (from ad hoc elaborations from Eurostat data).However, a more exhaustive indicator for measuring the loss of economic potential due to the exclusion of young people from the labour market is the NEET rate, which refers to the share of young people not in employment, education, or training.The age class usually considered for this indicator is 15-29 years, but in the last years, even more frequently, a lot of socio-economic literature refers to a wider age class, involving people until 34 years, because of the increased number of years of education and of the longer transitions to work.The NEET rate is the only indicator for the youth condition included in the Sustainable Development Goals of the UN Agenda 2030 − in pillar 8 'Decent work and economic growth' − and it has the advantage of viewing the loss of economic potential due to inactivity besides unemployment.The NEET rate has been usually studied by examining individual personal characteristics (Eurofound 2012(Eurofound , 2016;;Alfieri et al. 2015;Bell and Blanchflower 2010) or by analysing the macro-economic factors measured at the national level in comparative analyses across countries (Caroleo et al. 2020;Ryan 2001;Robson 2008).With reference to personal characteristics, a low level of education, an immigration background, the female gender, and a low household income are all factors predisposing to the NEET condition (Quintano et al. 2018).At the macro-economic level, the causes are ascribable to the education system characteristics, the labour market functioning, and the degree of efficacy of the Institutions regulating the school-to-work transition.Across European countries, Italy shows the highest share of NEETs.The causes of this negative primate are manifold.Scholars identified the high levels of unemployment, an education system unable to provide the skills required by the labour market, and the rigidity and inefficiencies of the institutions regulating the school-to-work transition.However, these studies have been unable to explain why young people living within the same country, sharing the same institutions, policies, and laws, and having identical personal characteristics, manifest so many different propensities regarding the NEET condition.A recent stream of literature has highlighted relevant intra-national disparities in the youth condition in the labour market, despite the existence of a homogeneous economic framework at the country level (Raffe 2014).In this context, the mesolevel framework of analysis assumes a crucial importance, because even in the presence of a common national context, the interaction of individuals with the local institutions may be different in different contexts.When we look at the meso-level of analysis, we consider all those services and structures which affect the interaction of young people and the various actors operating in the labour market.They could be identified with the education system, whose scope is to transmit to young people those skills and competencies required by employees, the Public Employment Services (PES), which should act favouring the match between the supply and demand of work in the labour market, should assist young people in CV preparation, inform on training opportunities, giving advice on how to face a competition or a job interview, etc.These Institutions are usually regulated and organized at the national level.In the case of Italy, however, significant differences in their functioning and sometimes even in their regulation could arise in a sub-national context.Very high in Italy is even the level of internal regional inequalities.Indeed, we can distinguish among the more developed and industrialized Northern and Central regions and the Southern ones, showing, instead, significant structural challenges and economic stagnation, despite the European plan of cohesion and social policy which tries to boost with various initiatives the economic growth of the European most depressed regions.Consequently, the NEET rates in the South of Italy are very high, but the differences in the share of NEETs detectable at a sub-national level are not simply explainable in terms of the well-known existent socio-economic divide between the North and the South of Italy and require a more analytical study.
Recently, the Italian Statistical Institute (Istat), in line with Eurostat, has experimented with new information sources and innovative methods in data production.In particular, new experimental social and economic indicators for even smaller territorial areas, such as municipalities and provinces, have been disseminated via the "A misura di comune" system.The availability of data at such a small territorial level opens up many new analysis prospects, particularly for the NEET phenomenon.
This paper aims to analyze the Italian NEET determinants linked to the place of residence, considering a deep territorial detail reaching the municipal level.The availability of information for different territorial details suggests using a multilevel model for the statistical analysis.Indeed, besides the municipal level, we considered broader territorial domains, such as provincial, regional, and macro-regional (corresponding to the NUTS − nomenclature of territorial units for statistics − 3, 2, and 1 levels).
The study of the NEET phenomenon at a sub-national level addresses many relevant research questions.First of all, the substantial uniformity of labour-market policies and institutions, which in Italy follow a national strategy, allows identifying the role exerted by the local social and economic context where an individual has grown up; second, the analysis of the phenomenon at the municipal level allows verifying the role of the different degrees of urbanisation on the NEET propensity (rural vs urban areas).Results highlight a strong relationship between the NEET phenomenon and the place where individuals live.The presence of many enterprises, the propensity to innovation, and the quality of the education system strongly contribute to the explanation of the territorial differences in the share of NEETs.Another interesting outcome is the identification of the different effects of these characteristics at the local level.The most relevant outcome from this study is that an increase in the efficacy of the education system can produce a significant reduction in the share of NEETs, especially in the most depressed area of the South of Italy.
The outline of the paper is as follows.Sect."The framework of analysis" details the framework of analysis, while Sect."The Italian territorial specificities" focuses on the effect of the territory on the NEET phenomenon.Sect."Data and methodology" details the data and methodology, while Sect."Results" presents the results.Conclusions are provided in Sect."Conclusions".

Defining NEETs
The term NEET was introduced in the UK for the first time in the late 90 s.Since then, it has been widely studied by scientists from psychology, sociology and economy.Most of these studies searched the NEET determinants looking at the personal characteristics of young people, such as gender (females), the country of origin (immigrants), the level of education (compulsory school or below), and the familiar context [a deprived familiar (Spielhofer et al. 2009) and social background (Quintano et al. 2018;Brunello and De Paola 2014;Bertola et al. 2007;Billari 2004)].However, the condition of NEET is connected even to factors acting at the macro, meso and local level (Berigel et al. 2023).
Many economists studied the effect of the macro-economic context, looking, in particular, at the education system and the labour market characteristics (Bratti et al. 2008;Caroleo and Pastore 2012;Hadjivassiliou et al. 2018;Piopiunik and Ryan 2012).
The term NEET includes both unemployed and inactive.While unemployment is connected to the low demand for work and to occupational mismatches, inactivity may be due to discouragement, and therefore indirectly ascribable to unemployment, but it may also be due to disability, the need to assist children or the elderly, social exclusion, an attitude of limited social and political participation (Walsh 2010; Caroleo et al. 2020;De Luca et al. 2019;Balan 2015), or the lack of resilience capacities and skills (De Luca et al. 2020).However, the NEET condition may also be apparent, hiding the involvement in the informal economy.Indeed, despite the reform that in 2017 introduced the new labour inspectorate agency, undeclared work remains widespread in Italy, especially in the South.

The main macro-factors affecting NEETs
The Italian education system Focusing on Italy and its education system, its organization is mainly State-founded, organized in a rigid and sequential way, with a vocational education path not very efficacious.Consequently, young people need to acquire the work-related skills in the period after education (Pastore et al. 2021;Quintini et al. 2007).The recent reforms regulating the apprenticeship system have not produced relevant effects on the NEETs reduction (Luca et al. 2019), and vocational-oriented higher education remains limited in scope.As a result, in Italy the school-to-work transition (i.e. the period from the completion of studies to the attainment of a stable job) is one of the longest ones (Pastore et al. 2020(Pastore et al. , 2021;;Caroleo et al. 2020;Pastore 2019).Moreover, the Italian education system does not appear sufficiently able to retain many young people in education and stimulate the attainment of high levels of education (De Luca et al. 2020;SALTO-YOUTH Inclusion Research Centre 2011;Sweet 2012).Indeed, compared to the other EU countries, Italy shows one of the lowest shares of tertiary graduates and one of the highest shares of early school leavers; for instance, in 2017, only Romania performed worse among EU countries (Alexa and Baciu 2021).Considering the international context of increased globalization and digitalization, the low Italian enrolment in high education negatively affects the ability of Italian young people to confront the new challenges in entering the labour market, which requires an increasingly specialized and skilled workforce.
The connection between early leavers and NEETs is strong and widely investigated in the literature: those who prematurely leave school have significantly higher probabilities of remaining in the NEET conditions for a longer period or even of not leave anymore this condition (De Luca et al. 2020).Indeed, being low-educated is the effect of young people's disengagement with school, and exposes them to a greater risk of unemployment, social exclusion, and poverty, because they did not acquire the skills and qualification required for employment (Alexa and Baciu 2021).At societal level, high rates of early school leavers affect the quality of human resources, have repercussion on the local level of productivity and competitiveness, depressing the economic growth.However, even if this phenomenon is often perceived almost exclusively as an individual problem, it should be considered as a complex process, affected by three overlapping factors: besides the individual one, the institutional and the structural/systematic ones (Marchlik and Tomaszewska-Pękala 2016).According to Coleman (1990), at the macro-level important factors in determining the share of ESLs are the education system characteristics while at the meso-level we find the school characteristics, where school is considered as the institution and the space in which the interactions between students, parents and teachers take place.
Moreover, considering the school characteristics with a specific focus on the levels where school drop out takes place, secondary-education teachers are currently not sufficiently valued and stimulated (European Commission 2019) and are − on average − older than their European peers (OECD 2014 database).These facts require the need to totally overhaul the education system.

The Italian labour market
With reference to the labour-market characteristics, Italy, Greece, and Spain, share the highest levels of unemployment and labour-market rigidity among the EU countries.The recent two-tier reforms aimed at introducing flexibility increased the recourse to temporary contracts, but the labour market remained rigid while job precariousness, especially for young people, increased.The more recent reforms aimed at creating hiring incentives to reduce the recourse to temporary contracts (the socalled "Decreto Dignità") remained ineffective, with the active labour-market policies weak and poorly integrated with social policies (Bruno et al. 2013;OECD 2018).Second-level (decentralized) bargaining continues to play only a limited role, partly due to the high share of small firms.In contrast, the high tax-wedge burden prevents employers from hiring, especially young people.Among the government initiatives taken to reduce the level of NEETs, the Youth Guarantee Fund has a relevant role.It has been planned at the EU level, based on the experience of Scandinavian countries, and aims to affect the frictional and mismatched components of unemployment.It was addressed to all the NUTS2 European regions with a youth unemployment rate higher than 25%.It is founded on a commitment by the Member States to ensure that all young people receive an offer of apprenticeship, training, continuing education, or employment suited to their abilities and experience within four months of becoming unemployed or leaving education (Escudero and Lopez 2015).However, its implementation within each member state has led to different results, primarily because it requires an in-depth transformation of the entire schoolto-work transition system.In Italy, this remains challenging because implementation suffers from the high complexity of the system and difficulties in managing a large number of projects at a local and regional level (European Commission 2020).

The regional inequalities
In the last years, the empirical evidence shows at the EU level a progressive process of gap reduction among the member States.However, at least some European countries like Italy, sharp regional divides tend to persist and, in some cases, even to increase (OECD 2016, p.19;Shankar and Shah 2003;Widuto 2019;Iammarino et al. 2018).For this reason, in recent years, both the socio-economic literature and the Institutions managing the official statistics provision are paying even more attention to the regional and even to the local level of analysis.A recent study from the European Commission (2017) identified two types of lagging regions in the EU, whose level of development is significantly lower than the EU average.The first group is formed by the low-growth regions, characterized by a persistent lack of growth, a per capita GDP of up to 90% of the EU average, and an absence of converging trend from 2000 to 2013.The second group is composed of the low-income regions, the ones remaining far below the EU average GDP per capita.They include all regions with a GDP per capita below 50% of the EU average in 2013.All the Southern Italian regions belong to the first group of lagging areaas.A recent paper from Diemer et al. (2022) identified the same regions as affected by a development trap.The authors define a regional development trap as the state of a region unable to retain its economic dynamism in terms of income, productivity, and employment.The reasons they identify for this situation are the absence of consistent policy attention, because at the European level the focus is mainly on the least advanced regions, while national policies are most focused on reinforcing the winners, to boost the national economy.Even according to a recent study by Iammarino et al. (2018), who identified 4 groups of European regions, from those with the highest GDP (120-149% of EU or national averages) until those with the lowest income (less than 75% of EU or national averages), all the Southern Italian regions resulted included in this last group.The authors show how the demographic and labour market characteristics of these four groups markedly differ, with strong synergies that of course involve all the richest ones.Conversely, the most depressed regions show even the stronger discontent towards globalisation and the European Union as a whole.as showed by Dijkstra et al. (2020), which analyzed the anti-EU vote and the influence on this behaviour, besides of personal characteristics, of the place of residence and in particular of its level of economic decline.
Looking at the Eurostat regional statistics, indeed, Italy, the UK and Spain have the most pronounced economic and social disparities across regions, despite the plans related to national economic and social policies, while Finland, Ireland, Sweden, and Germany show minimal internal differences.In particular, in Italy, the Northern Italian areas show high levels of economic development, the same achieved by Switzerland and most German and Norwegian regions.Conversely, the Southern Italian regions, together with most Eastern European areas and some Portuguese and Spanish regions, are the most in need of cohesion funds.Southern Italy is, therefore, in some respects, more similar to the most depressed areas of Portugal or Bulgaria than to the Northern Italian regions (Castellano and Rocca 2019;Faina and Lopez-Rodriguez 2004).In the South of Italy, productivity is particularly weak due to structural factors, such as insufficient and ineffective investment.Moreover, lower levels of education and innovation limit the positive impact of measures adopted to support private investment.Further, the South lags in terms of intangible asset and transport infrastructure, waste management, water infrastructure, and the prevention of risks from natural disasters.

The Italian administrative territorial organization
Italy is administratively organized into regions, provinces, and communes.Seven regions are allocated in the North-Centre of Italy (Aosta Valley, Liguria, Lombardy, Piedmont, Emilia-Romagna, Friuli-Venezia Giulia, Trentino-South Tyrol), four in the Centre (Lazio, Marche, Umbria, Tuscany) and the other seven in the South (Abruzzo, Apulia, Calabria, Campania, Basilicata, Molise, Sicily, and Sardinia; the last two ones are the major islands).The regions are not federal States, but they are the first-administrative divisions of the Italian Republic: they have political autonomy in some specific fields: for example, in the management of the local healthcare system and of the environment (in particular, in the waste collection and disposal).The second-administrative divisions of the Italian Republic are one hundred and seven districts, commonly referred to in Italian legislation as provinces.Fourteen of these provinces are metropolitan cities. Provinces are on the intermediate level between a municipality (commune) and a region.The Italian municipalities are the basic-administrative divisions and are seven thousand nine hundred one, allocated all over Italian regions and provinces.The population size of each municipality is different and is related to the resident inhabitants in its territory.The place of residence is the Italian municipality where an inhabitant is enrolled in the municipal demographic register.

The provision of official statistics at the regional and local level
The interest towards the regional level of analysis is constantly increasing in the socio-economic literature.The reasons for such growing interest are connected to the incapacity of the national level of study to explain the existence of relevant subnational disparities when studying different types of phenomena.In response to this increased need of statistical indicators at a subnational level, many suppliers of official statistics, such as Eurostat and the OECD, have begun to pay attention to the regional domains (Eurostat 2013).In particular, at the EU level, the regional domains identified by the NUTS 1, 2, and 3 levels have become the benchmark for EU policy-makers (Capello and Nijkamp 2009).Many EU-specific actions to remove economic and social disparities have been undertaken exclusively for the NUTS areas with the greatest needs in this context (European Parliament Committee of Regional Development 2013).This interest in regional detail has also been reflected at the national level by the National Statistical Offices.In Italy, ISTAT started to produce many socio-demographic indicators at the sub-national level corresponding to the macro-regional (North-West, North-East, Central, South, and Insular Italy), regional, and provincial domains.The corresponding territorial code is NUTS 1, 2, and 3, respectively.The regional and provincial domains also represent administrative districts with partial autonomy in the political sphere.However, in recent years, ISTAT has gone beyond the provincial detail.Following the program of experimental statistics, it proposed a set of indicators measured at the municipal level, where each municipality identifies the smallest local government unit.The focus on a similar level of territorial detail allows controlling for the effect of the degree of urbanization.

Empirical evidence on the regional and local dimension of analysis
From a macro-economic perspective, the NEET phenomenon has usually been studied at the national level, assuming the invariance of the phenomenon within each state.The analysis of the regional statistics showed that even the NEET rates significantly vary at the regional and local level.To explain the reasons of this high variability within the same national framework, we have to consider the influence exerted by the place where an individual has grown up, particularly the place where the individual started the transition from school to work.According to several sociological theories, the place where an individual lives strongly influences their attitudes and behaviours (Brofenbrenner 1979).Economic theory, meanwhile, shows that the local area influences labour-market outcomes through the specific opportunities available and the local network affecting people's job-search capacity (Dalziel 2015).Indeed, with the specific focus on the local context, we can even identify what Hodgson and Spours (2013) identify as the "local learning ecology".Similarly, the neighbouring territorial regions exert a strong influence in terms of economic interdependence, with possible gravitational effects (Tandoh-Offin 2011).
Recent literature on the NEET phenomenon has highlighted that in the majority of EU countries, the metropolitan areas act as a magnet, especially for the young, highly qualified, and mobile generations (Eurostat 2017).Conversely, young people living in rural areas show a disadvantage due to the costs of mobility if they wish to continue their schooling and due to the lack of opportunities of training and jobs, usually located in the metropolitan areas (Caroleo et al. 2021).However, Eurostat data (Eurostat online database by degree of urbanization) show that according to the degree of urbanization, in Italy the levels of NEETs are not significantly lower in the cities.This outcome could be due to the fact that where the institutions regulating the transition from education to work are not well developed, there is the risk that in the urban areas young people may find a greater disadvantage due to the difficulties to count on a network of friends and relatives which can help them to find a job is high.Indeed, it is well recognized that in Italy the most used channel for job search is the informal network of family and friends (see among the others Caroleo et al. 2021;Mussida and Zanin 2020).The main reason for this is that the labour-market infrastructure is underdeveloped (public and private employment agencies, schools, and universities) or declining (public competition) and the economy is mainly based on micro, small, and medium enterprises (OECD 2017).
Despite this evidence, few scientific works have chosen sub-national detail as the territorial unit of analysis to investigate NEET levels.One notable exception is the work of Bacher et al. ( 2017), which studied the NEET phenomenon in the Austrian regional context in relation to the different policies and education systems characterizing the albeit small territory of Austria, divided into federal states.Some years before, Bruno et al. (2014) analyzed the impact of the economic crisis on the NEET rate and on the youth employment rate in the EU regions.They found that NEET rates were persistent compared to the youth unemployment rate and that their persistence increased during the economic crisis.
Figure 1 reports the NEET rates for the Italian municipalities.The figure highlights the strong variability in the NEET rates with pronounced variations even internal to the Southern and the Northern areas, with high level of NEETs even in the coastal areas, where, besides the presence of big cities such as Naples, Bari or Palermo, there are even important touristic localities that, eprobably, are unable to generate a persistent economic growth.
However, previous studies have highlighted that, while there has been a progressive process of gap reduction among member states at the EU level, at least some European countries show sharp regional divides (OECD 2016, p.19;Shankar and Shah 2003).In recent years, in response to this evidence, many suppliers of official statistics, such as Eurostat and the OECD, have begun to pay attention to the regional domains (Eurostat 2013).In particular, at the EU level, the regional domains identified by the NUTS 1, 2, and 3 levels have become the benchmark for EU policy-makers (Capello and Nijkamp 2009).Many EU-specific actions to remove economic and social disparities have been undertaken exclusively for the NUTS areas with the greatest needs in this context (European Parliament Committee of Regional Development 2013).This interest in regional detail has also been reflected at the national level by the National Statistical Offices.In Italy, ISTAT started to produce many socio-demographic indicators at the sub-national level (i.e.NUTS 1, 2, and 3 areas), corresponding to the macro-regional (North-West, North-East, Central, South, and Insular Italy), regional, and provincial domains.These latter two domains also represent administrative districts with partial autonomy in the political sphere.However, in recent years, ISTAT has gone beyond the provincial detail.Following the program of experimental statistics, it has proposed a set of indicators measured at the municipal level.The Eurostat regional statistics show that Italy, the UK and Spain have the most pronounced economic and social disparities across regions, despite the plans related to national economic and social policies, while Finland, Ireland, Sweden, and Germany show minimal internal differences.In particular, in Italy, the Northern Italian areas show high levels of economic development, the same achieved by Switzerland and most German and Norwegian regions.Conversely, the Southern Italian regions, together with most Eastern European areas and some Portuguese and Spanish regions, are the most in need of cohesion funds.Southern Italy is, therefore, in some respects, more similar to the most depressed areas of Portugal or Bulgaria than to the Northern Italian regions (Castellano and Rocca 2019;Faina and Lopez-Rodriguez 2004).In the South of Italy, productivity is particularly weak due to structural factors, such as insufficient and ineffective investment.Moreover, lower levels of education and innovation limit the positive impact of measures adopted to support private investment.Further, the South lags in terms of intangible asset and transport infrastructure, waste management, water infrastructure, and the prevention of risks from natural disasters.
The NEET rates also reflect this disparity, as we can see from Fig. 1.Finally, the use of the municipal level allows controlling for the effect of the degree of urbanization on the NEET phenomenon.Recent literature has highlighted that in the majority of EU countries, the metropolitan areas act as a magnet, especially for the young, highly qualified, and mobile generations (Eurostat 2017).Conversely, young people living in rural areas show a disadvantage due to the costs of mobility if they wish to continue their schooling and due to the lack of opportunities of training and job (Caroleo et al. 2021).It is interesting to verify if this evidence can be found in Italy too or not, because in countries like Italy, where the institutions regulating the transition from education to work are not well developed, there is the risk that in the urban areas young people may find a greater disadvantage due to the difficulties to count on a network of friends and relatives which can help them to find a job is high.Indeed, it is well recognized that in Italy the most used channel for job search is the informal network of family and friends (see among the others Caroleo et al. 2021;Mussida and Zanin 2020).The main reason for this is that the labour-market infrastructure is underdeveloped (public and private employment agencies, schools, and universities) or declining (public competition) and the economy is mainly based on micro, small, and medium enterprises (OECD 2017).
Despite this evidence, few scientific works have chosen sub-national detail as the territorial unit of analysis to investigate NEET levels.One notable exception is the work of Bacher et al. (2017), which studied the NEET phenomenon in the Austrian regional context in relation to the different policies and education systems characterizing the albeit small territory of Austria, divided into federal states.Some years before, Bruno et al. (2014) analyzed the impact of the economic crisis on the NEET rate and on the youth employment rate in the EU regions.They found that NEET rates were persistent compared to the youth unemployment rate and that their persistence increased during the economic crisis.

Data
To analyze the levels of NEETs in Italy, we based our analysis on the existing international and national statistical databases collecting data at the sub-national.The indicators used in this analysis are extracted from the ISTAT and MIUR (Italian Ministry for Education and Research) databases.They all provide provincial or municipal detail.To select them, we started from the socio-economic literature on the factors affecting the NEET rates (Berigel et al. 2023).In particular, the indicators with the municipal detail are extracted from the ISTAT "A misura di comune" platform, which is a project started in 2016 and included in the ISTAT experimental statistics framework aimed at providing information with highly specific territorial detail, useful for the planning, programming, and management of local authorities.Currently, it contains NEET rates at the municipal level only for 2014.Other indicators with a provincial detail come from the ISTAT "Benessere Equo e Sostenibile" project, which aims to evaluate the progress in terms of welfare of Italian provinces from an economic, social and environmental perspective (Istat 2019).
Finally, the MIUR database provides data on the school system and has a provincial detail.
Specifically, we considered, at least in terms of correlation, the following determinants of the NEET rates (more details are in Table 1):

Economic development
Includes indicators connected to the presence of enterprises (entrepreneurial rate), innovation (patent intensity index and the share of high-tech employees), and territorial attractiveness (attraction index).

Quality of the education system and human capital equipment
Although the education system is regulated at the national level regarding the level of service provision and fundamental guiding principles, local authorities at the regional level have legislative power for education.This power concerns the organization of the school functioning, but it is limited by the power left to the single educational institutions, and it concerns mainly the vocational training and ALMP policies (Castelli 2004).Moreover, each educational institution has a certain level of autonomy in the didactical organization (https:// www.miur.gov.it/).Therefore, differences in the education system's effectiveness may exist at a territorial level.In the present study, we assessed this by considering teachers' personal characteristics (share of young teachers and share of support teachers).At the same time, we used the students' numeracy and literacy skills measured by the OCSE-PISA survey scores, as a proxy of the education system's effectiveness and some indicators of the population level of education achievement (share of people educated at a medium and a high level, and share of adults in continuous education).

Cultural and social participation and attitude towards sustainability
The NEET condition is often associated with a lack of cultural and social involvement, which in many cases determines or is determined by social exclusion.In our analysis, we considered as proxy for this phenomenon the number of not-for-profit organizations (Alfieri et al. 2015;Bell and Blanchflower 2010).

Sustainability
Attitudes towards sustainability is even important and proxied by the share of separate waste collection, because recent studies have found that the presence of more vulnerable NEETs is associated with lower levels of perceived environmental responsibility (Bonanomi and Luppi 2020;Cefalo et al. 2020).

Influence of criminality and the informal economy
High shares of NEETs may effectively hide the involvement of young people in the informal or illegal economy; furthermore, a high percentage of NEETs increases the likelihood of young people being engaged in criminality and the illicit economy (Karyda and Jenkins 2017;Quintano et al. 2018).Therefore, in this analysis we considered the number of reported crimes, the micro-crime rate, and the overcrowding in prisons.

Geographical and urban aspects
The degree of urbanization has been considered as it may affect young people's opportunities in the labour market in terms of resources, infrastructure, and services linked to the territory (Simoes et al. 2017).
Data from ISTAT refer to 2014, while data from MIUR on the high secondary school to 2008.Both these statistical sources represent, until now, the most recent data available for this level of territorial detail.However, the temporal gap between the two sources does not represent a relevant issue because those who attended high secondary school in 2008 were 20-24 years in 2014.Therefore, we assume that data from 2008 might be considered a valid proxy for the characteristics of high secondary schools attended by NEETs aged 15-29 in 2014, even in reason of the scarce temporal variation of these data.The different municipality size neither represents a problem, as all the indicators are relative measures.All the indicators presented in the current work are of high interest for the analysis of NEET determinants.However, in the econometric analysis, we introduced only a selection of them because several variables, such as separate waste collection, might have issues of reverse causality.In contrast, other indicators, such as per-capita income and the availability of services in the house, could directly depend on the NEET condition.These variables are analyzed only in their correlation with the NEET indicator, measured at provincial level when we calculate the correlation with the other indicators with the provincial detail (Table 2) and at municipal level for the correlation with the indicators observed at the municipal level (Table 3).
Table 2 shows that the NEET rate is higher where the shares of support teachers are higher (+ 0.432), probably because this latter is an indicator of stronger young people difficulties in learning.The other correlations are instead all inverse and denote that the share of NEETs is lower in the provinces with a higher patent intensity index (− 0.434) and a higher number of not-for-profit organisations (− 0.622).These indicators are indeed connected with the capacity of innovation and a high population's civic sense.Even higher the inverse correlations between the NEET rates and other variables connected with the education system, that is the average score for the tests in literacy (− 0.725) and in numeracy (− 0.754), the share of young teachers (− 0.531) and of adults in continuing education (− 0.552).These results confirm the fundamental role of the education system.A high level of student learning, the possibility to improve or acquire new skills even in the adult age are all factors preventing high rates of NEETs.
Unfortunately, we could not include in the analysis other relevant variables, such as the regional spending in education, the spending in active labour market policies, indicators of public employment services functioning, as they were not available with a similar territorial detail.These indicators are very important to explain the regional and local governance capabilities and the returns of these social investments on the population (Scalise and Hemerijck 2022).
Finally, the relationship between the share of NEETs and the covariates assumed as proxies of criminality (number of reported crimes, overcrowding in prisons, and micro-crime rate) is very weak.In all cases, the coefficients of correlation with NEET rates are negative but near zero; the only exception is the number of reported crimes, showing a slight correlation with NEET rates of − 0.296.Among the correlations observed at municipal level (Table 3), we note a strong and inverse correlation between the NEET rates and the share of waste separate collection.This outcome may indicate that high NEET rates are not due only to labour market failures, but find their roots also in the social sphere.A attention to the environmental  issues may be indeed a clear indicator of social disengagement and scarce municipal services.

Methodology
In reason of the hierarchical nature of the territorial units on which the indicators refer, we estimated different multilevel model specifications.Indeed, when we account for both municipal and provincial/regional information, single-level methods based on ordinary-least-squares (OLS) estimations are liable to suffer estimation problems.Multilevel models have already been successfully applied in similar studies by Caroleo et al. (2020) to estimate the probability of being NEET conditioned to individual and country characteristics, and by Biggeri et al. ( 2001) for the analysis of the school-to-work transition.Specifically, municipal and provincial/regional data show a hierarchical structure, with municipalities nested within provinces and provinces within regions.In other words, we assume that municipalities belonging to the same province have more similar characteristics, as do provinces belonging to the same region.Consequently, young people living in a specific province are likely to be more similar than young people living in different provinces and residuals of the single-level model may be not independent.According to the nested structure of the data (i.e.municipalities nested within provinces/regions), the multilevel model appears to be the most appropriate methodology for this analysis.Another way to correctly investigate similar kinds of data consists in using spatial models (for an overview see Dickson et al. 2019).Multilevel models may explain the variability of the NEET phenomenon at the municipal level, accounting for all the factors linked to the territory that could affect it.Specifically, they recognise the existence of such data hierarchies by controlling for territorial effects.Level-2 factors (i.e.variables at the provincial/regional level) can be used as independent variables to explain not only variability in a Level-1 dependent variable (i.e. the different shares of NEETs), but also variability in random intercepts and random slopes (Bickel 2007).We consider our analysis sufficient to account only for fixed effects, despite the number of groups being high.Using a model with only random intercepts, we assume that the coefficients measuring the relationship between the share of NEETs and each regressor are constant in different provinces/regions (Longford 1993).Only intercepts vary across groups (Heck and Thomas 2000; Rabe-Hesketh and Skrondal 2008).
We also tested two different specifications of three-level models, accounting also for the regions and the macro-regions (NUTS2 and NUTS1 areas, respectively).Multilevel models ensure more efficient estimates and correct the measurement of standard errors, reducing the risk of Type-1 errors (Aiello and Bonanno 2017).They account for residual components at each level in the hierarchy and partition residual variance into a between-provinces/-regions component (i.e.variance of provincial-/ regional-level residuals) and a within-provinces/-regions component (i.e.variance of municipal-level residuals). 1he model with the two-level structure is: where Y ij is the share of NEETs in the i-th municipality in the province/region j, X ij is the vector of municipal-level predictors, Z .j is the vector of provincial-/regionallevel predictors, and β 0 , β 1 , and β 2 are the intercept, the vector of coefficients for municipal characteristics, and the vector of coefficients of provincial/regional characteristics, respectively.In Eq. ( 1), 2 u is the provincial-/regional-level variance (i.e.Level 2), and 2 e is the variance at the municipal level.The expected value of Y ij for given X ij , Z. j , is: The comparison among different specifications can be carried out using the Akaike information criterion (AIC) and the Bayesian information criterion (BIC), which are indices based on likelihood with a penalisation for specifications involving high number of parameters (the difference lies in the way they penalise the richer models).The selected model will present the lower value of AIC (or BIC).Another way to evaluate alternative model specifications involves comparing the models' intra-class correlation coefficients (ICCs), identifying the proportion of variance in the outcome variable that is explained by the grouping structure of the hierarchical model.When the ICC approaches 0, grouping by provinces/regions is useless, and simple regression is sufficient.Otherwise, the higher the ICC, the stronger the capacity of the hierarchical model to contribute to the dependent-variable explanation.In the case of ICC approaching 1, no variance exists to explain the share of NEETs at the municipal level.Finally, the likelihood ratio (LR) test can be used to compare nested models.

Results
Table 4 shows different model specifications.Model 1 contains only a selection of indicators extracted from the domains of economic development, education, and cultural and social participation in reason of the coefficients' significance.All the covariates included in the model are inversely and significantly related to the share of NEETs, confirming that a consistent presence of enterprises and an entrepreneurial innovative mission can substantially reduce the percentage of NEETs.Attractiveness is measured according to the incoming flows of people as the ratio of total mobility flows.Municipalities attracting people for their higher economic vivacity (1)

Table 4
Model estimates for the determinants of NEETs through different multilevel specifications show lower NEET rates.Contextually, even the aspects connected with the quality of the education system and labour-force involvement in continuing education strongly contribute to reducing the share of NEETs.Where the high secondary school teachers are younger (less than 45 years old) and students' numerical competencies are higher, people are less likely to become NEETs after completing their education and entering the labour market.Thus, these findings highlight the importance of an efficient and effective education system, able to transmit the competencies provided.Further, younger teachers are the signal of an inefficient teacher recruitment mechanism, connected to an efficient labour market, which of course affects even its capacity to train young people.Indeed, OECD data (OECD 2014) show that in Italy the mean age of the teachers, especially of those in the high secondary schools, is very high, even when compared with that of the other EU countries.This is due to the high levels of unemployment, a high number of people, especially females, that want to apply for this type of profession, but it is even due to the recent Italian reforms and mechanisms of recruiting, and the block to the turnover imposed in the years just after the 2007 economic crisis, to favor the recovery.All these factors had, of course, significant repercussions even on the quality of the teaching.Further, younger teachers imply a narrower generational divide, which is extremely important, especially in an increasingly digitalized era.
For Model 1, the ICC is 0.25, showing that the clustering of municipalities within the regions contributes to explaining 25% of the total variability in the share of NEETs.Model 2 adds to the previous covariates two indicators linked to the degree of urbanization.All the other coefficients remain significant and stable.This latter model contains a good mix of indicators, all strongly significant in explaining the share of NEETs.The coefficient related to the attractiveness index appears to be not statistically significant, although its p-value is only 0.11.Regarding the variables connected with the degree of municipal urbanization, living in a city seems to be related to a higher share of NEETs.This finding is in line with the Eurostat data already cited in Sect."Empirical evidence on the regional and local dimension of analysis", but strongly differs from those referred to other European countries (see, among the others, Simoes et at., 2021).The reason for the Italian outcome is probably connected to the high Italian population density, compared to other EU countries (in February 2020, it was 206 P/km2 compared to an EU-average of 117 P/km2, and it was the highest among EU Mediterranean countries; see https:// www.world omete rs.info/ world-popul ation/ italy-popul ation/).In Italy, towns and rural areas are effectively isolated only in the internal mountain areas.Furthermore, because the most frequent way for Italian young people to find a job is by asking friends and relatives (Caroleo et al. 2021;Mussida and Zanin 2020), these informal channels are more developed in small towns than in the cities.However, this outcome needs further investigation.
Models 3 and 4 differ from Model 2 because they include a third hierarchical level.They are, respectively, the NUTS2 (administrative regions) and the NUTS1 (macro-geographical breakdowns in North-West, North-East, Central, South, and Insular Italy) levels.In Model 3, the regional level contributes to an extremely high explanation of the total phenomenon variability, as shown by the ICC coefficients at regional and provincial levels (0.347 and 0.418).However, the entrepreneurial rate and the participation in continuing training2 lose their statistical significance.The structure with three levels can explain the differences in the NEET rates due to the different entrepreneurial rates, strongly unbalanced regarding the concentration of enterprises in the North of Italy.Adults' continuing training is managed and financed mainly at a regional level, leading to significant variations in the level of funding.For example, within the North of Italy, only four Northern regions (Lombardy, Veneto, Emilia-Romagna, and Piedmont) account for 57.7% of the total expenditure (for more details, see ANPAL 2018).
For the comparison of different model specifications, the LR test confirms the advantage in passing from a two-level to a three-level model.This is calculated comparing Models 3 and 4 with Model 2, having the same covariates, but only two (provinces/municipalities) domains.The test is strongly significant for the municipality/province/NUTS2 structure (Model 3), with a chi-square of 40 and a p-value of 0.000.Conversely, for the municipality/province/NUTS1 structure, the chi-square is 3.12, and the p-value is 0.0772.
The very high variability of dynamics connected to the share of NEETs in the national territory, the different mechanisms regulating the dependence relationship between NEETs and the covariates identified in these models (see Table A1 in the Appendix for the descriptive statistics of covariates) suggest further focusing the analysis within each Italian macro-region separately.
Tables 5 and 6 show, respectively, separate model specifications for these NUTS1 areas with two and three hierarchical levels (provinces and municipalities for Table 5; regions, provinces, and municipalities for Table 6).Results show that the characteristics linked to the education system, such as the levels of students' numerical competencies and the share of the young teachers, significantly affect the percentage of NEETs only in Central and in the South of Italy.Focusing on the South, from the descriptive statistics for these indicators reported in Table 1, we can see that it shows the lowest levels of these indicators (only Isles have a lower value), denoting that highest values for these indicators have the capacity of effectively reduce the share of NEETs.Adults' participation in continuing education represents a relevant factor only in the South, probably due to the higher incidence in this area of NEETs and the smaller number of institutions providing this type of service.Indeed, for the South, where the share of people reached by this service is lower: the availability of a variety of formal and informal learning activities and methods tends to increase personal human capital and then to protect from dismissals and facilitates recruitment.This is able to significantly reduce the share of NEETs in the territory with the highest share of NEETs.
In other words, in the South, the availability of continuing education services can create conditions to reduce this phenomenon by providing people with the skills   required by the labour market that school has been unable to offer them.Furthermore, living in a city does not represent a penalization factor only in the North-West of Italy, where we find the wealthiest Italian cities (e.g.Turin and Milan, among others).Indeed, the coefficients of the dummy variables for towns and densely populated areas are negative, even if not both significant at 0.10, indicating that compared to rural areas, living in more densely populated areas is associated with a decrease in the share of NEETs.Conversely, for the other Italian areas, passing from a rural to a densely populated area is associated with a significant increase in the share of NEETs.
These different model specifications show the significantly different composition of the NUTS1 areas in terms of NEET rates.
These findings are difficult to interpret without considering the deep territorial complexity.Unlike other countries, such as Romania and Bulgaria, where the concentration of NEETs in rural areas is strongly significant, for Italy it is important to consider the type of rural municipality because social isolation and economic deprivation connect mainly with the so-called internal areas, identifiable with mountain and rural areas.Conversely, rural areas closer to cities or coastal rural areas show different characteristics.These findings are in line with what the Voluntary national review 2022 showed.This latter was presented at the High-level political forum of the United Nations, finalized to inform about the level of advancement versus the UN 2030 Agenda, and identified a key strategy based on the territorialization of the SDGs.The main message of this initiative was that different regions and types of areas need different measures.
Moreover, the higher rates of NEETs observed in some Italian cities may be even the consequence of the internal/international migration flows, leading to a strong concentration of young people having great difficulties in finding a job in reason of their scarce language knowledge and of their greater difficulties in seeing their educational levels recognized.
Figure 1 shows that the NEET rates in the North-East and the South have a little variability.Conversely, in the other areas, especially in the North-West, there is substantial variability in the NEET phenomenon.The ICC coefficients in Table 5 show that grouping the municipalities within the provinces can explain 33% of the total variance in the North-West.Provinces explain only 10% of total variability for the North-East area and only 4% for the South.The ICCs in Table 6 confirm these findings.For the North-East of Italy, shifting from the provinces/municipality structure model to the model with a region/province/municipality structure does not increase the explained variability in the share of NEET rates.The gain in terms of ICC is practically null, and the model is the same as that in Table 5.Conversely, for the North-West, the ICC for the regional level is the highest, equal to 0.56.The values of the ICCs for Central and Insular Italy are in an intermediate position.

Conclusions
The NEET phenomenon determines a significant loss for economic productivity and, from the social perspective, represents a condition of marginalization that can increase various forms of disengagement.The NEET indicator includes both inactive and unemployed individuals, which represent markedly different conditions but produce similar social exclusion and economic loss outcomes.The causes from which the NEET condition arises may be manifold.This paper has analyzed the NEET phenomenon in Italy, the country with the highest share of NEETs within the EU, focusing on the determinants linked to the place of residence.To achieve this, we used a wide array of indicators made available by ISTAT and MIUR, at provincial and municipal levels.
Results highlight that the variability of the NEET phenomenon can be explained also in terms of characteristics of the place of residence.The most significant factors are some economic aspects, such as the entrepreneurial rate and the share of employees in high-technology sectors, and, more generally, the ability of the territory to attract people.Factors linked to education, such as the share of young teachers and the students' numerical competencies, also play a relevant role in explaining the differences in the percentage of NEETs within the national borders.The analysis highlights that the share of NEETs in Italy is higher in densely populated areas than in small towns and rural areas, except for the North-West area.This means that, in these territories, the cities, instead of exerting a gravitational effect by having more job opportunities and being hubs for innovation, appear as urban areas where social exclusion and unemployment prevail.This finding may be a clear signal of a highly depressed economy, in which even the big cities are unable to create job opportunities for youngsters.
In light of the marked differences existing between the North and the South of Italy, which undoubtedly are also reflected in the municipal variability in the share of NEETs, as a robustness check, we repeated the analysis separately for each macro-area of Italy: North-West; North-East; Central; South; and Insular.Overall, the results obtained at the national level were confirmed within each macro-region, although we can observe some peculiarities.The aspects connected with education, such as the share of young teachers and the mean score for numerical competencies, play a relevant role only in Central and South Italy.Places where the share of adults involved in continuing education is higher are associated to lower percentages of NEETs especially in the South, where the NEET rates are higher, denoting these types of education initiatives and training to create more job opportunities after regular education has been completed.
The major associations emerging for the South of Italy among these variables and the share of NEETs at municipal and provincial level are an interesting result because they highlight that precisely in the territories where the NEET phenomenon is more widespread, a higher than average quality of the education system (measured by the numeracy competencies) and the involvement in continuing education are able to make the difference leading to significant lower share of NEETs.
In conclusion, in this paper, the analysis at the municipal level on the share of NEETs highlighted the importance of factors connected with education and the economic characteristics of the territory, already known in literature.However, an important outcome is the importance of the degree of urbanisation of the place of residence as NEET determinant.Further, the territorial differences in the share of NEETs are not simply explainable in terms of the well-known gap between the North and the South but have more complex dynamics, and the same determinants

Table 1
Variables used in

Table 2
Correlation matrix for provincial indicators(*)