The low skills trap: the failure of education and social policies in preventing low-literate young people from being long-term NEET

ABSTRACT This paper investigates to what extent the likelihood of young people being long-term NEET can be explained by low literacy skills, how this varies across advanced countries, and how this cross-national variation can be explained by education and social policies. We use PIAAC data and include macro-level indicators on education and social policies. We analyze the likelihood of being long-term NEET versus being in employment or in education/training among some 34,000 young people aged 20–30 from 25 countries. We find that low-literate young people are more likely to be long-term NEET. While NEET risks are associated with countries’ institutional characteristics, this does not mean that these characteristics and policies always work in favour of low-literate young people. Although high levels of (enabling) ALMP generally reduce the risk of being NEET, they do so less for low-literate young people. Additionally, young people living in social-democratic welfare states are less likely to be NEET, but low-literate young people seem to profit less from this.


Introduction
In 2019, 13% of young people aged 15-29 in OECD countries were Not in Employment, Education or Training ('NEETs') (OECD 2020a).This risk of being NEET is likely to be associated with a lack of basic skills, such as literacy skills.Literacy is a fundamental, 'key information-processing skill' (OECD 2013a), and while such skills are crucial to learn and obtain new knowledge, not everybody acquires them to a sufficient degree.People with low literacy skills are generally less integrated in society.They are more likely to drop out of school and to be unemployed, and they are also less politically and socially involved (Finnie and Meng 2001;Leuven, Oosterbeek, and Van Ophem 2004;Rudd, Kirsch, and Yamamoto 2004;Hernandez 2011;OECD 2013aOECD , 2015)).Phrased differently, low-literate young people lack the skills that are important for success in school and on the labour market, and are therefore disproportionally more likely to to become NEET (OECD 2015).This paper studies the extent to which young people with low literacy skills are more likely to be long-term NEET compared to young people with higher literacy skills, and to what extent relevant institutions and policies serve as a safety-net for these low-literate people.
The concept of NEET originates in the wish of policymakers to identify young people who are not well integrated in society, mainly because they 'disappear from the radar': They are not registered in education, but often also not officially registered as 'unemployed' because they are not entitled to unemployment benefits yet (Eurofound 2017). 1 NEETs are regularly framed as a societal problem (Eurofound 2012), and policymakers try to decrease the number of NEETs and design policies to stimulate the integration of these young people into the education/training system or the labour market (Eurofound 2012(Eurofound , 2017)).These institutional arrangements and social policies contribute to both education and labour market participation, and some institutions specifically support young people in their school-to-work transition (Breen 2005;Gebel and Giesecke 2011;Levels, Van der Velden, and Allen 2014).However, most of these institutional arrangements and social policies are general in nature and not specifically targeted towards the most vulnerable group of low-literate young people.It is not clear to what extent the educational and social safety-nets work equally well for the lower and the higher literate young people.
Therefore, we explore to what extent some institutional characteristics not only reduce the NEET risk in general, but also specifically prevent vulnerable low-literate young people to become NEET.Our research questions are: (1) To what extent are low levels of literacy skills associated with a high risk of being longterm NEET? (2) How does this vary across advanced countries?(3) To what extent, can this cross-national variation be explained by education and social policies?
To answer these research questions, we use data from the Program for the International Assessment of Adult Competencies [PIAAC] (OECD 2013b).PIAAC is a large cross-national adult literacy survey of adults in 33 advanced countries.The data contain information on many of the individual-level variables commonly used to explain NEET risks, such as educational attainment, work experience, socioeconomic background, migration background, family situation, health, etc. (Coles et al. 2002;Strelitz and Darton 2003;Cassen and Kingdon 2007;Social Exclusion Task Force 2008;Eurofound 2012).We link these data to a country database with characteristics on education and social policies, allowing us to investigate how individuals with low literacy skills fare in different institutional settings.
Our analyses contribute to literature in two main ways.First, we contribute to the literature on low literacy.When looking at school and labour market success, the literature mainly focuses on the level of education as a proxy for human capital, rather than on an individual's actual skills level (OECD 2018).However, while education is likely to enhance literacy skills and is indeed strongly positively correlated with it, skills and educational attainment cannot and should not be equated.There exists considerable skills heterogeneity within educational attainment groups and the skills distribution of educational attainment groups overlap considerably, with the most strongly literate lower educated individual being more literate than a considerable group of middle educated individuals.These underlying skill differences are likely drivers of relevant outcomes (Levels, Van der Velden, and Allen 2014;OECD 2015).
Second, we advance the literature on institutions and NEETs.Individual's risk to become NEET plausibly arise from interactions between individual (life-course) characteristics and countries' education and social policies (Hodkinson 1996;Hodkinson and Sparkes 1997;Kogan and Müller 2003;Müller 2005).As such, the effectiveness of institutions and policies to combat NEET varies between different social groups.Completely disillusioned and disengaged young people with low literacy skills are likely to respond very differently to institutional incentives than, for instance, high-literate young people who might also have difficulties making the school-to-work transition.These interactions are rarely studied and ill-understood.For example, we found that much of the literature that takes the relevance of institutions for young peoples' participation in education and the labour market into account, focus on young people in general and not specifically on NEETs ( Van der Velden and Wolbers 2003;Breen 2005;Bol and Van de Werfhorst 2011;Gebel and Giesecke 2016).

Theory and hypotheses
To provide a theoretical explanation for the relationship between having lower literacy skills and being long-term NEET, we start by identifying various mechanisms through which individual and institutional characteristics generate NEET periods during a young individual's life course.Our theoretical frame explains why some young people do not have a job and are not in education or training either.
Literacy skills are part of young people's human capital and are important predictors of economic and social participation.These skills are important for productivity, i.e. for success on the labour market, but also necessary to develop other skills, i.e. to be successful in education.One of the key factors that explain why some young people become NEET may be differences in the extent to which they have mastered relevant skills (OECD 2015;Gladwell, Popli, and Tsuchiya 2016).Young people with low general skills are at greater risk of dropping out of school (Cairns, Cairns, and Neckerman 1989;Audas and Willms 2001;Traag and van der Velden 2011).This could be because young people with low skills have more difficulty in acquiring and developing further skills.In addition, low-skilled youth are also less likely to engage in training or invest in further schooling, for example, because their experience has taught them that the low returns to education do not justify an additional investment (Mincer 1974), or because they do not expect to be able to complete the (additional) education programme (cf.Fan and Wolters 2014).
This makes them more likely to leave education and try to find a job on the labour market.However, they may be hampered in that effort, as having lower literacy skills also decreases the chances of success on the labour market (Barrett 2012).According to various matching theories (Kalleberg and Sorensen 1979;Logan 1996;Müller 2005) and the queuing theory (Thurow 1975;Sørensen and Kalleberg 1981), employers strive for the best possible match, given their preferences, opportunities and constraints.Employers rank all available applicants in an imaginary queue on the basis of expected training costs: the applicant who is the first in line, will be hired (Thurow 1975).Consequently, young people with low skills are more likely to end up at the back of the job queue and are therefore less likely to find and keep a job.For this reason, we expect that: Low-literate young people are generally more likely to be long-term NEET (Hypothesis 1).
In theory, some institutional characteristics should 'protect' low-skilled people and thus weaken the relationship between low skills and being NEET, while other institutional characteristics might do the opposite and strengthen this individual-level relationship.In the next paragraphs, we will argue that the vocational orientation of an educational system may affect the extent to which low-literate young people are likely to be longterm NEET.Furthermore, we will examine the extent to which activating labour market policies (ALMP) and welfare state regimes may affect the relationship between low literacy and the likelihood of being long-term NEET.

Vocational orientation
In countries that mainly offer vocationally oriented education, young people with low literacy skills have the opportunity to continue studying in vocational tracks in upper secondary education.This may offer an alternative for low-literate young people and give them a protected position on the labour market (Shavit and Müller 2000;Ryan 2001).Moreover, in vocationally oriented educational systems, vocational tracks have more prestige than in more academically oriented systems (Shavit and Müller 2000;Solga 2002;Iannelli and Raffe 2007).
Previous literature strongly suggests that vocational education systems indeed smoothen the transition from school to work (Van der Velden and Wolbers 2003;Breen 2005).From the employers' perspective, we expect that in vocationally oriented systems the level of general skillsincluding literacy skillsis less important, because employers want to hire people who need less training and are directly productive on the job.Therefore, they will place less emphasis on general skills and more emphasis on occupation-specific skills.In these labour markets, young people with a vocational degree have fairly good labour market opportunities, since their skills are directly applicable in the workplace.
On the other hand, in countries where the education system is less vocationally oriented, general skills become more important in the hiring process and tertiary education is more often required on the labour market (Shavit and Müller 2000).Here, higher literacy skills indicate higher levels of trainability.In addition, in these countries vocational education is more stigmatized and regarded as a safety net that serves as a remedial option for young people that are not able to study at higher educational levels (Shavit and Müller 2000;Solga 2002;Iannelli and Raffe 2007).
Based on these arguments, we expect that for vocationally educated the effect of having low literacy skills on the chance to become NEET is lower in countries that have a strong vocational orientation compared to countries that mainly offer academic education.This leads to the following hypothesis: Vocationally educated young people with low literacy skills are generally more likely to be long-term NEET than vocationally educated young people with higher literacy skills; however, this individual-level relationship is less strong in countries where the education system is more vocationally oriented (Hypothesis 2).

Activating labour market policies (ALMP)
Active labour market policies are aimed to improve the educational and labour market prospects of young people.We expect that the higher the amount of public spending on ALMPs in a country, the more likely it is that the country is improving the educational and labour market prospects of young peoplefor instance, by training or re-training young people in skills that are required on the labour market.Therefore, we expect the effect of having low literacy skills on the chance to become NEET is lower in countries with higher levels of public spending on ALMPs compared to countries with lower levels of such public spending: Low-literate young people are generally more likely to be long-term NEET, but this individuallevel relationship is less strong in countries with higher levels of public spending on ALMPs (Hypothesis 3).
ALMPs can be characterized as 'enabling' and 'enforcing' policies (Dingeldey 2007;Knotz 2012).Enabling policies assume that certain jobseekers lack the necessary skills to enter or re-enter the labour market.These policies facilitate training, for example, but also make it easier to combine work and family care.We expect that in countries that mainly focus on enabling policies, low-literate young people are less likely to be long-term NEET because these policies focus on training or re-training young people in skills that are demanded by the labour market.In addition, these policies enable young people to find a job that fits their skills.Therefore, we expect the effect of having low literacy skills on the chance to become NEET is lower in countries with strong enabling ALMPs compared to countries with weaker enabling ALMPs.We assume that: Low-literate young people are generally more likely to be long-term NEET, but this individuallevel relationship is less strong in countries with stronger enabling ALMPs (Hypothesis 4).
In contrast, enforcing policies assume that some people lack the motivation or incentives to find a job.These policies aim to tighten the readiness-to-work requirements and suitability criteria by, for instance, reducing the level and duration of financial benefits (Knotz 2012).Rather than stimulating young people in training, re-training or further education, these policies force young people into a job or education, independently of their level of skills.Therefore, we expect the effect of having low literacy skills on the chance to become NEET is lower in countries with strong enforcing ALMPs, compared to countries with weaker enforcing ALMPs.Hence, we hypothesize: Low-literate young people are generally more likely to be long-term NEET, but this individuallevel relationship is less strong in countries with stronger enforcing ALMPs (Hypothesis 5).

Welfare state regimes
One of the aims of welfare state regimes is to help vulnerable people by offering social security.However, the way welfare states are organized varies and we expect that this will also create differences among the risks for low-literate young people to become NEET.Welfare state regime types are indeed associated with NEET rates and patterns (Jongbloed and Giret 2021; Pastore 2018).The most widely used classification of welfare states is the one developed by Esping-Andersen (Esping-Andersen 1990).He distinguishes social democratic, liberal and the conservative welfare states.This classification has limitations as it seems to fit less well for the Mediterranean, and transitional (post-socialist) countries (Arts and Gelissen 2002).Various changes to and expansions on Esping-Anderson's classical categorization have been proposed (for example: Gallie and Paugam 2000;Ferrera and Rhodes 2000), and the conceptualization policy and categorisation of countries into ideal-typical regimes has strongly influenced the literature on transition regimes (Walther 2006;Pohl and Walther 2007;Hadjivassiliou et al. 2016).Building on this literature, in addition to the three clusters of welfare states distinguished by Esping-Anderson, we distinguish a Mediterranean cluster (Southern-European countries) and a transitional cluster (East-European countries) (Hadjivassiliou et al. 2016).
The social-democratic welfare states are characterized by high levels of decommodification and strong universalism.It stimulates work-engagement and individual independency; however, it also includes highly distributive benefits not dependent on individual contributions (Esping-Andersen 1990).One would expect therefore that groups at risk, such as people with low literacy levels, might receive additional support in such welfare states to help them participate fully in society.In liberal welfare states there is low decommodification and strong individualistic self-reliance.Here, the rights to social benefits are limited to the bare minimum which is meant to stimulate work rather than using benefits (Esping-Andersen 1990).At the same time, solidarity is low, and this implies that participation in society is seen as an individual responsibility with few supports for groups at risk.The conservative welfare states follow a moderate level of decommodification and benefits are based on former contributions and status (Esping-Andersen 1990).They are more corporatist in nature and the role of state is seen as complementary to the role of families.The Mediterranean welfare states are more strongly family-oriented, with little state intervention, a more fragmented social safety net and more particularism and clientelism (Ferrera, 1996).The transitional welfare states are the former communist countries with the previously associated strong state intervention and solidarity, that have undergone a dramatic change towards a liberal market-driven economy, but lacking the basic social safety nets that characterize liberal countries (Kogan and Müller 2003).
Based on this typology of welfare states we expect that the effect of literacy on young people's chance to become NEET is lower in social-democratic welfare states (because these welfare regimes are more likely to help the most vulnerable people) and higher in Mediterranean and transitional welfare states (due to the lack of state interventions and the relatively under-developed ALMPs; Hadjivassiliou et al. 2016).Therefore, we hypothesize: Low-literate young people are generally more likely to be long-term NEET, but this relationship is less strong in social-democratic welfare states than in Mediterranean and transitional welfare states (Hypothesis 6).

Data
We use data from the Programme for the International Assessment of Adult Competencies (PIAAC), collected in 24 participating countries between August 2011 and March 2012 and in 9 additional countries between April 2014 and March 2015.The survey is designed to provide valid and reliable estimates of adults' competences in key information-processing skills, to identify proficiency differences between sub-groups of the population, to understand development, maintenance and use of skills, as well as to determine the impact of proficiency levels on life chances.Country samples contain some 5,000 adults between the ages of 16 and 65.Respondents were interviewed in computer-assisted personal interviews and took a computer-based test (although pencil-andpaper data collection strategies were also used).Respondents were given assessment tests designed to directly measure their general skills.More specifically, these tests measure numeracy and literacy skills, and respondents' capacity to solve problems in technology-rich environments.The survey is cross-culturally and cross-nationally valid.
To prepare the data for analyses, we selected people between the ages of 20 and 30. 2 Data from Russia were not analyzed for technical-administrative reasons; data from Cyprus, Lithuania, Singapore and Turkey were not analyzed because data were missing for some or all of the macro-indicators; Chile was not analyzed because of the particularly high proportion of low-literate young people.We also deleted the specific oversample of PISA 2000 survey respondents in Denmark because these targeted respondents were not part of the international target population definition (OECD 2013a).Finally, we weighted data from Canada, since the original Canadian sample was much larger than the samples in other countries.To weight back, we drew a random sample of 35%.These selections resulted in a total sample of N = 34,347.Lastly, we deleted cases with missing values on the NEET variable (N = 403) and the other individual-level variables (N = 221).These selections resulted in a total working sample of N = 33,742 respondents aged 20-30 from 25 countries.

Measurements
Descriptive statistics of all variables are presented in Table 1.Below is a description of how the variables were measured.

Dependent variable
. Employed, education/training or long-term NEET: We construct a variable indicating whether young people have been long-term NEET using three separate questions whether the respondent (a) has had paid work, (b) participated in formal education or (c) participated in non-formal education. 3Respondents who did not participate in any of these activities in the period of 12 months preceding the survey are labeled as long-term NEET.The reason to take a period of 12 months or more is to exclude respondents whose NEET status is largely transitory, such as short-term unemployed, and absence of work due to illness or pregnancy.The time spent as NEET is a useful way to distinguish between NEETs who are vulnerable and those who are not (see for example: Levels et al. 2022) Respondents who participated in paid work were labeled as employed, and respondents who participated in formal or non-formal education were labeled as in education/training.If young people were both employed and in education/training, we labeled them as being employed.

Country-level variables
. Vocational orientation: Bol andVan de Werfhorst (2014, 2016) combined two data sources (from the OECD and UNESCO) measuring the percentage of students enrolled in upper secondary vocational programmes.The variable was standardized by the original authors.The index we use to measure vocational orientation has a mean of 0.28, a standard deviation of 0.91 and a range from −1.84-1.74; the higher the score, the more strongly vocationally oriented the system is. .ALMP: We use public expenditure on active labour market policies as a percentage of GDP provided by the OECD (OECD 2020b).We use data from the year before the year of the interview and standardized this variable (mean 0, standard deviation 1, range from −1.17-3.57).The higher the score, the higher the expenditure on ALMPs. .Enablement: This indicator is based on the indicator constructed by Knotz (2012).The indicator is based on several underlying dimensions of enablement, such as wage subsidies, job counselling, training, family services, etc. (see Knotz 2012).We standardize this variable (mean 0, standard deviation 1, range from −1.60-1.96).The higher the score, the greater the effort in enabling ALMPs.A1; for an overview of the values on country-level variables per country see Appendix, Table A2.
. Enforcement: We use the enforcement indicator from Knotz (2012) The scores are computed as averages of the effort scores on the underlying dimensions of enforcement such as low benefit level, obligation to job-search, etc. (see Knotz 2012).We standardize this variable (mean 0, standard deviation 1, range from −1.95-1.99).The higher the score, the greater the effort in enforcement ALMPs. .Welfare states: We cluster countries into the three types of welfare states as indicated by Esping-Andersen (1990), and add two additional clusters, Mediterranean and transitional welfare states, based on Hadjivassiliou et al. (2016).Based on Gal (2010), we include Israel in the Mediterranean welfare states.South-Korea is left out of the analyses (For an overview of the countries per cluster, see Appendix Table A2). .Output gap: To control for the business cycle we include the output gap (i.e. the difference between the actual and potential GDP expressed as a percentage of GDP) of the year before the interview (OECD 2019a).We standardize this variable (mean 0, standard deviation 1, range from −3.96-1.26).A positive value indicates a higher demand for workers.This indicator is commonly viewed as a better indicator for demand-supply imbalances than unemployment statistics.However, we do run a robustness analysis using unemployment statistics as well (see Appendix, Table A6). 4 Individual-level variables . Low literacy skills: PIAAC measures three types of general skills: literacy skills, numeracy skills, and skills related to problem solving in technology-rich environments (OECD 2013a(OECD , 2016)).These skills measures are constructed using adaptive testing methods.We focus on the domain of literacy skills. 5Literacy is defined as 'understanding, evaluating, using and engaging with written texts to participate in society, to achieve one's goals, and to develop one's knowledge and potential' (OECD 2012b).It does not include the ability to write or produce text.After the test, ten plausible values are generated that represent the respondent's literacy skills.
In line with the literacy skills levels constructed by the OECD (2016), we define young people as low-literate when they score at or below level 1 (score below 226) on the literacy test.This is the lowest level of literacy.People who score at level one can 'complete simple forms, understand basic vocabulary, determine the meaning of sentences, and read continuous texts with a degree of fluency' (OECD 2013a, 67) and we compare them with young people with a score above level 1 (score 226 or above).To measure this, we estimate for each individual the level of literacy skills (at or below vs. above level 1) based on the ten plausible values provided by PIAAC.We account for both the sampling error component and the variance due to imputation of the literacy scores (for detailed information about PIAAC and technical issues, see: OECD 2016).Figure 1 shows the percentage of young people with low literacy skills per country, based on the ten plausible values. 6Japan (2.1%), Korea (3.9%) and Finland (4.3%) have the lowest percentages of young people with low literacy skills, while Israel (20.7%),Italy (22.1%) and Greece (25.7%) have the highest.
. Male: A dummy signifying whether the respondent is male or female.
. Age: Included as a continuous variable.Additionally, we include a squared term for age to account for non-linearity of the relation between age and being long-term NEET.
. Migration status: Measured using dummies to distinguish non-immigrants, first-generation immigrants (both parents and respondent foreign born) and second-generation immigrants (both parents foreign born, respondent born in test country).Non-immigrants are the reference category. .Parental education level: A categorical variable indicating the educational attainment of the most highly educated parent.We distinguish between two lower-educated parents, at least one medium-educated parent and at least one higher-educated parent.We also include a dummy for missing values.Respondents with two lower-educated parents are the reference category. .Educational attainment: Is based on a collapsed version of the International Standard Classification of Education (ISCED) 2011.We distinguish three levels: low education (ISCED 0/1/2/3C short), medium education (ISCED 3A/3B/3C long/4A/4B/4C) and high education (ISCED 5A/5B/6).A low education level forms the reference category. .Vocational track: A dummy signifying an academic or vocational track.

Methods of analysis
Our hypotheses focus on explaining the individual-level relation between having low literacy skills and being long-term NEET.We assess whether this relation differs crossnationally and is moderated by institutional characteristics.To test our hypotheses, we perform multinomial regression analyses to account for the different possible outcomes: in employment, in education or training or long-term NEET.We add cluster standard errors to correct for the hierarchical clustering of respondents within countries.In addition, we apply the Final Weight 'spfwt0' to the model to make the data representative of the target population.We estimate these models using the mlogit package in Stata 15.
Since PIAAC is a cross-sectional data source, results should be interpreted as associations rather than causal effects.Therefore, our analyses do not account for reverse causality between literacy skills and the risks of being long-term NEET.Also, we cannot control for unobserved heterogeneity (e.g.factors such as motivation and abilities) that may obscure the relationships.These caveats should be taken into consideration when evaluating our results.Thus, our results should not be interpreted as an attempt to formally identify the causal impact of education or social policies.However, we may evaluate whether or not our coefficients are in line with what we would expect to see if we would assume a causal relationship.

Descriptive analyses
Figure 2 presents percentages of the 20-30-year-old population that are long-term NEET in the various countries.A large cross-national variation can be observed.The countries with the lowest proportions of long-term NEETs are the Netherlands (2.1%), Norway (2.9%) and Austria (3.2%).Countries with high proportions of long-term NEETs include Italy (15.6%), the Slovak Republic (16.7%) and Greece (19.0%).The proportion of longterm NEETs in the PIAAC data is generally lower than what is reported in the literature (Eurofound 2012).However, the pattern in the observed rates across countries closely resembles what we know from the literature (Eurofound 2012).
Figure 3 (Appendix, Table A3) shows the relationship between low literacy skills and being long-term NEET.While the strength of the relationship varies across countries, it shows that in all countries the relation is positive and for most countries it is also significant.This indicates that having low literacy skills generally increases the risk of being long-term NEET, which is in line with Hypothesis 1.However, the figure also indicates that there are considerable crossnational differences in the strength of the relation, and that in some countries (e.g.Japan, Italy, the United States, Korea, and Finland) the relation is not statistically significant.

Multinomial regression analyses
We perform a multinomial regression analysis to account for young people's likelihood of being in employment or in education/training rather than being long-term NEET.For the margin plots see Appendix Figure A1.In Table 2, Model 1, we expected that the relationship between vocationally educated young people with low literacy skills and being long-term NEET, would be less strong in countries where the education system is more vocationally oriented.However, the interaction term in Model 1b is not statistically significant, so we refute Hypothesis 2.
Model 2 shows the results of ALMPs in general.We find that in countries with high levels of ALMPs, young people are less likely to be long-term NEET, mainly because they are in education or training.However, the cross-level interaction shows that low-literate young people do not profit from this: In countries with high levels of ALMPs, lowliterate young people are more likely to become NEET and less likely to be either employed or in education/training.This is the opposite of what we expected in Hypothesis 3.
Model 3 shows similar results for enabling ALMPs.We find evidence that young people in countries with high levels of enabling ALMPs are in general more likely to be in education/training than to be long-term NEET (Model 3a), but this does not hold for young people with low-literacy skills.The interaction term in Model 3b shows that low-literate young people profit less from enabling ALMPs than high-literate young people.This is the opposite of what we expected and refutes Hypothesis 4.
When we look at enforcing ALMPs, we observe in Model 4a that the main effect of enforcing ALMPs is negative: The higher the level of enforcing ALMP in a country, the less likely young people are to be in education/training and the more likely they are to be NEET.However, this holds stronger for high-literate than for low-literate young people.Low-literate young people are in fact more likely to be in education/training than to become NEET.This partially confirms Hypothesis 5.
Lastly, Model 5 shows the results on the differences between types of welfare states.We find that young people living in social-democratic welfare states are more likely to be employed or in education/training than in any of the other types of welfare states.However, if we look at the interaction terms, none of the coefficients are significant in the education/training versus NEET specification.This indicates that in the different welfare states (conservative, liberal, transitional, Mediterranean), having low literacy is   not related to the likelihood to be NEET instead of being in education/training.The same holds for the social-democratic welfare states, as the main effect of having low literacy skills is also not significant.In the specification of being employed versus being NEET, we find the opposite of what we expected.The main effect for having low literacy skills in Model 5b (i.e. the association for low-literate young people in social-democratic welfare states) is significant, indicating that low-literate young people in these welfare states are less likely to be employed than to be NEET.On the other hand, many of the interaction terms for the other types of welfare states are borderline positive significant, indicating that in these other welfare states, low-literate young people are relatively less affected by the overall higher likelihood in these countries to become NEET instead being employed.We therefore refute Hypothesis 6.

Additional analyses
We have conducted several additional analyses to check whether our results are robust.First, we keep the number of countries equal across all analyses by leaving out the countries with at least one macro-indicator missing (see Appendix, Table A5), and find that our conclusions did not substantially change.
Second, to include a different measure of labour market shortages, we controlled for youth unemployment rate rather than output gap (see Appendix, Table A6).We found that this did not change our conclusions.
Third, we checked to what extent the results change when we define young people as low-literate when they score at or below level 2 (score below 276) (see Appendix, Table A7).Like in the original analyses, we find that enforcing ALMPs weaken the relation between low literacy and long-term NEET versus being in education/training.In contrast with the original analyses, the interaction term is also significant for the specification of long-term NEET versus being employed.
Lastly, we re-estimated the analyses by leaving out one single country at the time.This does not change our main conclusions (results available on request).

Limitations
Before discussing our findings and drawing conclusions, it is good to note some shortcomings of our analyses.First, PIAAC is a cross-sectional survey, with all the limitations that are associated with that.One problem, for example, is that literacy skills are measured at the time of the interview, while skill levels may change during periods of inactivity.When young people are neither in employment nor in education/training, their skills may erode through lack of practice (OECD 2012a).This depreciation of human capital increases over time (OECD 2012a) and may lead to reverse causality (NEET affecting the skills levels instead of the other way around).We see this as a main limitation to our analyses.
Second, an ideal data set would have allowed to compare changes within a country on the relation between literacy skills and NEET.This would require not only historical data regarding changes in the institutional characteristics in a country, but also individual level data on an individual's propensity to become NEET before and after such changes.Such a combination of time series data within and across countries is a very powerful tool to analyze the effect of policy interventions.Unfortunately, these data do not exist yet.
Third, our study was subject to some noteworthy data limitations.We were unable to distinguish the months of NEET as a linear dependent variable, as the PIAAC survey did not ask for the duration of NEET.We, therefore, decided to focus on young people who were NEET for at least one year, since the long-term scarring effect will be the most prevalent in that group.However, a more dynamic approach to being NEET would have been better (e.g.Brzinsky-Fay 2007;Contini, Filandri, and Pacelli 2019;Dicks, Levels, and Van der Velden 2020;Levels et al. 2022).Furthermore, the NEET concept is rather criticized for being inherently vague (Levels et al. 2022) and for lumping together groups that are quite different, e.g.young people that have disillusioned withdrawn from the labour market to young adults taking a gap year after successfully finishing their studie.Unfortunately, the PIAAC data do not allow for further distinguishing the group of NEETs.By focusing on the long-term NEETs (i.e.those in NEET for 12 months or more), we hope to at least cover the 'hard-core' of the group.
Fourth, while young people entered the labour market at different times, the education system measure for vocational orientation and the measures on (enabling and enforcing) ALMPs are not time-varying.The issue here is that valid and reliable time-varying measures do not exist.We assume that this is not a problem because macro-variation in the countries under study changes slowly and the ranking is unlikely to have changed.Follow-up research, for example on single countries, should focus on institutional reforms to confirm our results.

Conclusions and discussion
In this paper we aim to explain the extent to which the likelihood that vulnerable young people with low literacy skills become socioeconomically marginalized as long-term NEET, and how this can be explained by countries' education and social policies.We study young people aged 20-30 and use PIAAC data from 25 advanced countries.Multinomial analyses are performed to test whether young people are long-term NEET rather than being employed or being in education or training.
Our results indicate that young people with low literacy skills are caught in a 'low skills trap'.Although the strength of the association differs across countries, we find that young people with low literacy skills are more likely to be long-term NEET.Reasons for this could be related to having more difficulties in acquiring and developing skills, which also makes them less likely to engage in training or further schooling.This is likely to be exacerbated as long-term inactivity can lead to further skills obsolescence (Mincer and Ofek 1982;Bynner and Parsons 1998;Salthouse 2006;Desjardins and Warnke 2012).This makes them more likely to leave education and try to find a job.However, having low literacy skills also decreases the chances on the labour market (Barrett 2012) because employers always strive for the best possible match, given their preferences, opportunities and constraints (Kalleberg and Sorensen 1979;Thurow 1975;Sørensen and Kalleberg 1981;Logan 1996;Müller 2005).Since people learn most from activities in education or at work, when such possibilities are being denied, low-literate NEETs have more difficulty in further developing their skills.Therefore, low-literate NEETs find themselves in a trap from which it is increasingly difficult to escape.
Institutional characteristics might play a role in preventing low-literate young people from being caught in this trap.Their effectiveness in doing so is what this paper has tried to assess.We expected that most of the institutional characteristics under study would weaken the odds for low-literate young people to become NEET.The main finding of this paper is that this is not the case.
Just to be clear, the institutional characteristics that we looked at are systematically related to young people's general risk of becoming long-term NEET.Countries with high levels of (enabling) ALMPs seem more successful in keeping young people in education or training and preventing them from long NEET spells.However, these characteristics do not seem particularly helpful for the most vulnerable group: young people with low literacy skills.Rather the opposite: (enabling) ALMPs work in favour of young people with the highest literacy skills.The same holds when we look at different types of welfare states.Here we find that young people living in social-democratic welfare states are more likely to find themselves in employment or education/training than in a NEET position.However, we did not find evidence that this also holds specifically for low-literate young people.Again, the protective measures do not seem to be targeted towards the most vulnerable group.The only exception to this rule, was when we looked at enforcing ALMPs: These seem to work more in favour of young people with low literacy skills than for those high literacy skills.However, given the overall negative effect of enforcing ALMPs on young people's risk to be pushed out of the education system and become NEET, this is not the kind of social policy a country would probably like to engage in.
So what does this imply for possible policy interventions?Concerns about NEETs and interventions to improve young people's integration in society (be it in education or on the labour market) figure high on national and international policy agendas (Eurofound 2017), but little is known about interventions that specifically focus on the low-literate group (Windisch 2015).When we look at our results, we note that most of the social policies to combat NEET are aimed at the mainstream group: young people with higher literacy skills.This is probably the case because it is easier to provide education, training or jobs to the group of higher-literate young people than to young people with lower literacy skills.If the policy aim is to reduce the overall number of NEETs in a country (and that is often the case (Eurofound 2017)), the higher-skilled NEETs are the low-hanging fruit that will be picked first.It is much more difficult to help young people with low literacy skills, because such interventions can only be effective if they are linked to improving those persons' literacy skills.In addition, low-literate young people are difficult to reach.For example, the German campaign Nur Mutder nächste Schritt lohnt sich.Besser lesen und schreiben lernen ('Take couragethe next step will pay off.Improve your reading and writing skills') used posters, TV and radio commercials to reach out to adults with low literacy skills.While the campaign helped to raise awareness on low-literacy issues, it failed to reach the low-literate people themselves (OECD 2019b).This problem may also have hampered the effectiveness of the education and social policy characteristics we studied in this paper.For example, while ALMPs may seem to focus on young people at risk, the policies are often not designed specifically for groups at risk, particularly if they are low-literate young people (Pawson and Tilley 1997;Vooren et al. 2019).This increases the risk of ALMPs failing to reach low-literate young people, thereby increasing the inequality between low-and high-literate groups.Therefore, policies should first focus on improving literacy skill levels among young people.Skills policies start in primary and secondary education, so that young people will have gathered sufficient skills during the period of mandatory schooling.This is not only important for their educational and labour market success, but also for other daily tasks in life.If they are able to acquire better literacy skills, then education and social policy can further ensure that they can perform better in their economic and social participation in the society.
Nevertheless, a group of young people with low literacy skills will remain and at some point, they will leave education.It is essential for this group that social policy is much more targeted, actively reaching out, and creating awareness about the implications of having low literacy skills.This should not only be focused on the target group itself, but also on (prospective) employers.While providing additional opportunities for this group to follow education and training, much more should be done to foster skills development at work.For example, a Dutch programme entitled Tel mee met Taal offers subsidies encouraging employers to support low-literate employees with extra training (Bersee 2019).Additionally, more attention should be paid to the evaluation of initiatives to promote the effectiveness of such programmes.Randomized control trials could be carried out to evaluate the effects of incentives to support low-literate young people before an approach is widely implemented.
Notes 1.There is some critique on the concept of NEET as well, mainly because it lumps together different subgroups of people who ae inactive, but who are not equally vulnerable.This causes heterogeneity in the NEET category.One way of getting closer to using the NEET concept to measure vulnerability is by focusing on those who remain NEET for a longer time period (see for example: Levels et al. 2022) 2. We limited the sample to young people aged 20-30, because in most countries compulsory education ends after age 18 (OECD 2018).3. The concept of NEET explicitly excludes people that are in education or in training.In the PIAAC data, non-formal education is defined as 'any organised and sustained educational activities' outside the definition of for-mal education that 'may [. ..] take place both within and outside educational institutions, and cater to persons of all ages' (OECD 2011, 31).For example, courses through open and distance education, organised sessions for on-the-job training or training by supervisors or co-workers, seminars or workshops.4. It should be noted that labor market conditions vary within countries, between regions.As our aim is control the association between national institutional characteristics and individual-level outcomes, we control for labor market conditions at the national level.5. Literacy and numeracy are very strongly correlated (r = .9),and problem-solving skills were only measured for two-thirds of the sample.6.All percentages are estimated via the REPEST command by Avvisati and Keslair (2014).

Figure 1 .
Figure 1.Percentage of low-skilled young people across countries.

Figure 2 .
Figure 2. Percentage of long-term NEETs per country.

Figure 3 .
Figure 3. Association between low literacy skills and being long-term NEET.

Table 2 .
Multinomial regression estimates on the relationship between low literacy skills and long-term NEET, employment versus education/training.
education level, educational attainment and output gap; for the full-model, see Appendix, TableA4; Odds ratios.

Table A2 .
Descriptive statistics country-level variables.

Table A3 .
Regression estimates on the relationship between low literacy skills and long-term NEET, per country.

Table A4 .
Multinomial regression estimates on the relationship between low literacy skills and long-term NEET, employment versus education/training.

Table A5 .
Multinomial regression estimates on the relationship between low literacy skills and long-term NEET, employment versus education/training -16 countries.

Table A6 .
Multinomial regression estimates on the relationship between low literacy skills and long-term NEET, employment versus education/trainingcontrolled for youth unemployment rate instead of output gap.

Table A7 .
Multinomial regression estimates on the relationship between low literacy skills (at or below level 2) and long-term NEET, employment versus education/training.< 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001; Standard errors in parentheses; country-level variables are standardized; Models are controlled for male, age, age 2 , migration status, parental education level educational attainment and output gap; Odds ratios.