James Browne and Rodrigo Fernández - OECD · FACES OF JOBLESSNESS IN SPAIN: ANATOMY OF EMPLOYMENT BARRIERS James Browne and Rodrigo Fernández

FACES OF JOBLESSNESS IN SPAIN: ANATOMY OF EMPLOYMENT BARRIERS

James Browne and Rodrigo Fernández

http://www.oecd.org/social/faces-of-joblessness.htm

mailto:[email protected]



mailto:[email protected].



Policy Analysis Note (PAN) for Spain © OECD 2016 5

TABLE OF CONTENTS

1. INTRODUCTION ....................................................................................................................................... 7

2. LABOUR MARKET AND SOCIAL CONTEXT .................................................................................... 10

3. EMPLOYMENT BARRIERS IN SPAIN ................................................................................................. 17

4. FACES OF JOBLESSNESS IN SPAIN .................................................................................................... 21

5. CONCLUSIONS ....................................................................................................................................... 29

ANNEX A. LATENT CLASS RESULTS ...................................................................................................... 33

ANNEX B. LATENT CLASS ANALYSIS AND MODEL SELECTION .................................................... 35

Tables

Table 2.1. Risk of poverty or social exclusion, 2014 ................................................................................. 13 Table 3.1. Employment-barrier indicators ................................................................................................. 19 Table A.1. Latent class estimates ............................................................................................................... 33 Table A.2. Characterisation of the latent groups ....................................................................................... 33

Figures

Figure 2.1. Employment rates: strong recovery from the crisis ................................................................. 10 Figure 2.2. Population groups with potential labour market difficulties ................................................... 15 Figure 2.3. Composition of the Spanish population with labour market difficulties ................................. 16 Figure 3.1. Employment barrier: conceptual framework ........................................................................... 17 Figure 3.2. Number of simultaneous barriers ............................................................................................ 20 Figure 4.1. Share of individuals facing multiple employment barriers in each group ............................... 28 Figure B.1. Selection of the optimal number of latent classes ................................................................... 36

Boxes

Box 2.1. Individuals with potential labour market difficulties (target population) .................................... 14 Box 4.1. Group 1: “Labour-market inactive women with low education

and weak financial incentives” .................................................................................................................. 21 Box 4.2. Group 2: “Unemployed prime age adults with low work experience” ........................................ 22 Box 4.3. Group 3: “Experienced but low-skilled unemployed men” ........................................................ 22 Box 4.4. Group 4: “Well-educated prime age adults with weak labour market attachment” .................... 23 Box 4.5. Group 5: “Early retirees with weak financial incentives” ........................................................... 23 Box 4.6. Group 6: “Unemployed women with low work experience” ...................................................... 24 Box 4.7. Group 7: “Low-skilled women in unstable jobs” ........................................................................ 25 Box 4.8. Group 8: “Labour-market inactive mothers with low work experience” .................................... 25 Box 4.9. Group 9: “Low-skilled individuals with health problems and high levels

of earnings-replacement benefits” ............................................................................................................. 26 Box 4.10. Group 10: “Educated parents in short-term unemployment or working part time” .................. 26 Box 4.11. Group 11: “Unemployed youth without any past work experience facing

scarce job opportunities” ............................................................................................................................ 27 Box 4.12. Group 12: “Unemployed mothers actively looking for work but facing

scarce job opportunities” ............................................................................................................................ 27 Box 4.13. Group 13: “Short-term unemployed men with high earnings-replacement benefits

facing scarce job opportunities” ................................................................................................................. 28


ACKNOWLEDGEMENTS

This document was produced with the financial assistance of the European Union Programme for

Employment and Social Innovation “EaSI” (2014-2020, EC-OECD grant agreement VS/2016/0005,

DI150038). It is part of a joint project between EC and OECD (VS/2016/0005 (DI150038), Cooperation

with the OECD on Assessing Activating and Enabling Benefits and Services in the EU) covering six

countries: Estonia, Ireland, Italy, Lithuania, Portugal and Spain.

The report incorporates feedback received during the project kick-off seminar, held at OECD in Paris

on 3 March 2016 with the participation of the EC and representatives from all participating countries. Lead

authors gratefully acknowledge contributions from colleagues at the OECD (Herwig Immervoll, Daniele

Pacifico, Céline Thévenot), as well as comments from colleagues at the World Bank (Aylin Isik-Dikmelik,

Sandor Karacsony, Natalia Millan, Mirey Ovadiya, Frieda Vandeninden, Michele Davide Zini), who are

undertaking a parallel and closely related joint project with the EC covering six further EU Member States.

All views and any errors in this report are the responsibility of the authors. In particular, the report should

not be reported as representing the official views of the OECD, of the European Union, or of their member

countries.

This project is co-funded by the European Union


FACES OF JOBLESSNESS IN SPAIN

ANATOMY OF EMPLOYMENT BARRIERS

1. INTRODUCTION

This Policy Analysis Note (PAN) for Spain assesses the characteristics and employment barriers of

working-age individuals with no or weak labour-market attachment. It is one of six such country notes in a

joint EC-OECD project covering Estonia, Ireland, Italy, Lithuania, Portugal and Spain. The objective

of this project is to provide a novel perspective on employment difficulties, and to aid in the identification

of policy approaches to overcome them. The project website at http://www.oecd.org/social/faces-of-

joblessness.htm provides further information.

Each PAN develops profiles of key employment barriers and quantifies their incidence and intensity

among jobless individuals and among those who work or earn very little or intermittently. The underlying

conceptual framework and statistical approach is described in an associated methodological background

paper (Fernandez et al., 2016; Immervoll and Isik-Dikmelik, 2016) and is consistent with that employed in

a related EC-World Bank activity covering six further EU countries. The empirical results from each PAN

will be used to inform a dialogue on policy approaches and options that could address the most prevalent

employment barriers in selected population groups and strengthen their labour-market attachment. This

dialogue will take place in a second part of the EC-OECD project. Its results and an associated policy

inventory will be presented in a series of six Country Policy Papers (CPP).

A key motivation behind this project is the finding from the literature on activation and employment-

support policies (AESPs), and on social protection systems more generally, that careful targeting and

tailoring to individual circumstances are crucial factors for policy success.1 However, policy discussions do

not necessarily reflect this. They often refer to broader labour-market groups such as “young people”,

“older workers”, “people with disabilities” or “lone parents”. Similarities of employment barriers among

members of such broader groups are implicitly assumed but not well documented (for instance, being

“young” is not an employment barrier). As a result, policy interventions targeted on the basis of

characteristics such as age, health status or family situation alone may be ill-adapted to the needs of jobless

individuals and those with precarious employment patterns. An in-depth inventory of people’s employment

barriers, and an identification of groups who share similar combinations of labour-market obstacles, can

contribute to a better match between individual needs and available support, and make associated policy

interventions more effective and less costly.

Countries frequently seek to account for individual circumstances and labour-market difficulties by

means of powerful statistical tools that “profile” individual benefit claimants using administrative data.

Such tools are useful for tailoring the employment programmes that each registered individual is offered.

They often rely on administrative data, which have distinct advantages, but tend to cover only a subset of

the out-of-work population, such as the registered unemployed. As a result, the profiling tools built around

these data typically cannot be used to provide a broader perspective on the employment barriers facing the

entire population of those with no or weak labour market attachment. This note complements existing

profiling instruments by adopting more of a “birds-eye” approach that considers the employment barriers

of all those with no or weak labour market attachment. This sizeable and heterogeneous group constitutes

1. See for example OECD (2013a, 2013b, 2014a, 2015a); Immervoll and Scarpetta (2012); Arias et al.

(2014); World Bank (2013); European Commission (EC) (2015); Eurofound (2012).




the potential client group for AESPs. Understanding their employment barriers is not only important for

linking up services provided by different institutions, but it is also essential for identifying groups who

would benefit from employment-related programmes or incentives, and who are not currently clients of

any of the institutions providing such measures.

A comprehensive assessment of potential employment barriers requires detailed information on

people’s skills, work history, health status, household circumstances and incomes. The European Union

Survey on Income and Living Conditions (EU-SILC) contains rich information for identifying and

assessing potential barriers to employment and is the primary source of data for this note. EU-SILC offers

cross-country comparability, an extended reference period2 over which one can assess the respondents’

main activity status, and detailed information on individual and family circumstances including people’s

work-related skills end education, work history, health status, income sources, tax liabilities and benefit

amounts. However, there is a relatively long time-lag between data collection and availability (EU-SILC

2014 was made available in February 2016). EU-SILC also contains less detailed information on labour-

force status than standard labour-force surveys.

In the Spanish SILC data for 2014, 45% of the working-age population3 can be considered to face

potential labour-market difficulties. The remainder of this note refers to this group as the “target

population”. 4 Of this 45%, 30% did not work at all throughout the reference period and a further 15% had

“weak labour market attachment” with either unstable jobs, limited working hours or zero or near-zero

earnings. Potential employment barriers that are particularly common among the target population include

no recent work activity (67% of the target population), low education or professional skills (52%), scarce

job opportunities (46%) and limited total past work experience (44%). Other potential barriers, such as

health limitations, care responsibilities, high levels of earnings-replacement benefits and high levels of

non-labour income are frequent among some sub-groups, but less prevalent overall.

The results of the statistical clustering analysis suggest that the target population can be separated into

13 distinct groups with similar employment-barrier profiles within each group. Focusing on the prevailing

characteristics in each group, the emerging clusters may be summarised as follows:

1. “Labour-market inactive women with low education and weak financial incentives” (14% of those

with no or weak labour market attachment)

2. “Unemployed prime-age adults with low work experience” (13%)

3. “Experienced but low-skilled unemployed men” (11%)

4. “Well-educated prime-age adults with weak labour market attachment” (9%)

5. “Early retirees with weak financial incentives” (9%)

6. “Unemployed women with low work experience” (7%)

7. “Low-skilled women in unstable jobs” (6%)

8. “Labour-market inactive mothers with low work experience” (6%)

2. EU-SILCdata provide information on individuals’ labour-market status at different points in time during

the reference year (each of the twelve months) and at the time of the interview. This note uses all 13 data

points to characterise people’s employment status.

3. Ages 18 to 64, excluding individuals in full-time education or in compulsory military service.

4. This figure is expected to be higher than the average proportion of working-age people in Spain who were

not in paid work during 2013 (the reference year for the 2014 SILC) according to EU Labour Force Survey

statistics, which was 41%. The difference can be explained by the high share of people with a weak labour

market attachment in Spain. These people were in work only a small proportion of the year (and then, on

average, considered as “unemployed” or “inactive” by Labour Force Surveys). So the figures are not

necessarily inconsistent.


9. “Low-skilled individuals with health problems and high levels of earnings-replacement benefits”

(5%)

10. “Educated parents in short-term unemployment or working part time” (5%)

11. “Unemployed youth without any past work experience facing scarce job opportunities” (5%)

12. “Unemployed mothers actively looking for work but facing scarce job opportunities” (4%)

13. “Short-term unemployed men with high earnings-replacement benefits facing scarce job

opportunities” (4%)

These group labels indicate that proxy groupings, which are commonly referred to in the policy

debate, such as “women”, “disabled”, “youth”, include distinct sub-groups with very different

employment-barrier profiles. For instance, the following combinations of employment barriers are

common for women with children: low work experience combined with care responsibilities (Group 8),

care responsibilities and a lack of job opportunities (Group 10) and care responsibilities, scarce job

opportunities and low education or skills (Group 12). Employment barriers are also highly heterogeneous

for youth (Groups 4 and 11), workers with intermittent employment patterns (Groups 4, 7 and 10) or older

individuals (Groups 1 and 6). As shown in Section 4, these groups also differ markedly with respect to

their poverty risks, material deprivation levels and other family or individual circumstances.

Most individuals in the target population face more than one potential employment barrier

simultaneously. Four in five face at least two such barriers, and about one in four show three or more. For

instance, a vast majority of the “Unemployed youth without any past work experience facing scarce job

opportunities” (Group 11) combine scarce job opportunities, low education or skills with a complete lack

of previous work experience. Similarly, many “Labour-market inactive mothers with low work experience”

(Group 8) lack work experience and also have care responsibilities that may limit their availability for paid

work. For the majority of the target population, therefore, addressing one type of employment obstacle

may not be enough to boost employment levels. From a policy perspective, these results point to a need to

carefully combine or sequence different activation and employment support measures, and to co-ordinate them

across policy domains and institutions.

The rest of this note proceeds as follows. Section 2 provides some background information on the

evolution of social and labour market conditions in Spain and how this compares with other EU countries.

Section 3 uses the most recent EU-SILC data to provide quantitative indicators of the intensity and

incidence of different types of employment barriers. Section 4 applies a statistical clustering technique to

organise the population of individuals with no or weak labour-market attachment into groups with

homogeneous combinations of employment barriers. It also presents key demographic and socio-economic

characteristics that are relevant for deciding policy priorities and approaches for each group. A short

concluding section highlights selected possible directions for extending the approach further.


2. LABOUR MARKET AND SOCIAL CONTEXT

Trends in employment, unemployment and labour-market inactivity

In Spain as well as in the five other countries covered by this project, the economic crisis has

significantly impacted labour markets, in turn causing increased poverty and material deprivation. The

impact of the crisis in Spain was especially long lasting, with employment rates continuously declining for

six consecutive years after the start of the crisis and a recovery starting only in 2013/14.

Figure 2.1 shows employment-to-population ratios between 2007 and 2015 and compares these with

the EU average. During the crisis, the employment rate in Spain fell by 11 ppts, from 70% in 2007 to less

than 59% in 2013. This is the largest peak to trough fall among the six countries analysed. Since 2014

employment rates have recovered. However, by 2015, the rate in Spain was only slightly above that of

Italy and remained well below the four other countries included in this project, and also well below the

EU average.

Despite its size, the fall in employment rates provides only a partial picture of the extent of labour-

market slack during and after the recession:

A decline of the working-age population by around one million over this period was driven by

historically low fertility rates and substantial drop in net immigration: migrant inflows fell from

15.7 per thousand inhabitants in 2005 to 6.7 in 2013, while outflows increased from 1.1 to 8.3 per

thousand. In 2013, the outflow of foreign nationals exceeded inflows by 211,000. There was also

a net outflow of 41,000 Spanish nationals. Among those Spanish nationals emigrating, about

three quarters were of working-age (OECD, 2015d).

The share of involuntary part-time work rose from 3.9% of employees in 2007 to 10% in 2015,

indicating a substantial degree of underemployment.

Figure 2.1. Employment rates: strong recovery from the crisis

In % of working-age population

Note: The EU average is weighted.

Source: Eurostat Labour Force Statistics.

50

55

60

65

70

75

80

2007 2008 2009 2010 2011 2012 2013 2014 2015

Estonia Ireland Italy Portugal Spain Lithuania European Union 28


Trends in employment rates were mirrored by substantial movements in unemployment, which

peaked at 26.1% in 2013. It has since fallen to 20.4% (Q1 2016) but remains much higher than the

EU average (8.8%). Economic activity rates have increased slightly during this period as a result of higher

female labour force participation.

Much of the adjustment from lower labour demand in Spain during the crisis took the form of lower

employment rather than lower wages. Inflexible sectoral collective bargaining arrangements with

automatic wage indexation meant that even before the crisis the most common response for firms facing an

adverse demand shock was to reduce employment (OECD, 2014b). A recent reform in 2012 has sought to

address this by increasing firm-level flexibility in sector-wide collective bargaining agreements and

increasing the possibilities for firms to opt out of sector-wide agreements even without the consent of

social partners.

Recent estimates indicate a sharp increase in structural unemployment since 2007 (European

Commission, 2016). Possible key reasons for this structural change, suggested by a shift of the Beveridge

curve since 2008 (European Commission, 2016, Graph 2.4.10) include the following:

A greater mismatch between current labour supply and labour demand in terms of skills. In

2014, about one third of those in employment were low-skilled against 42% before the crisis,

which suggests a shift of labour demand towards higher skills requirements and a continuing

significant need for upskilling. In 2014, the share of 25-54 years old individuals with low

educational attainment (ISCED 0-2) was 39%, (compared to an EU average of 21%) According

to PIAAC survey data, Spain has comparatively few top performers in any of the measured

assessment areas. In mathematics 8% performed at level 5 and 6 (the highest levels),

considerably below the OECD average of 12.6% (OECD, 2013c). Despite the variety of options

available to adult learners, adults with low levels of skill remain less likely to participate in job-

related education or training than their counterparts in other countries. Only 19% of low-skilled

adults in Spain participated in some form of formal or non-formal adult education or training in

Spain in 2012, compared to the OECD average of 31% (OECD, 2015e).

Labour-market segmentation hindering an efficient job re-allocation process. The Spanish

labour market is characterised by a high level of duality between those on permanent contracts

who benefit from very strong employment protection legislation and those on temporary

contracts. The effect of these protections is that it is much more expensive for firms to make

individuals on permanent contracts redundant than those on temporary contracts and therefore

many of those on temporary contracts were made redundant during the recession: the proportion

of workers on temporary contracts fell between 2007 and 2012 despite around nine in every ten

new contracts during this period being fixed term. Recent reforms have sought to bring the costs

of terminating fixed-term and permanent contracts closer in to line, and appear to have been

successful in increasing the number of new hires on permanent contracts (OECD, 2014b, 2016b).

However, an OECD report examining the impact of these reforms suggested that more could be

done in this area. For example, the report recommended reducing severance costs for large

employers to levels closer to the EU average (OECD, 2014b).

Difficult-to-access re-employment and job-search support. Investments in Active Labour

Market Policies (ALMPs) in Spain remain limited. The participation rates in ALMPs among

active jobseekers was the 7th lowest in the EU in 2013 at 7.2%, and despite the marked increase

in unemployment, spending was below the EU average. Moreover, spending does not appear to

be well targeted. Spending on vocational training activities accounted for only about 22% of the

ALMP budget in 2015, whereas more than half was spent on employment incentives, subsidised

employment measures and rehabilitation. Around 20% of the ALMP budget was spent in 2015 on

measures of direct job creation (public works), which have been shown to be relatively

ineffective at improving employability, including in Spain (Card et al., 2010; ESTEP, 2014). A

recent OECD study recommended reducing reliance on employment subsidies and public works

programmes and increasing training opportunities for the long-term unemployed. It also argued


for tighter targeting of employment subsidies and direct employment creation measures on those

furthest from the labour market (OECD, 2015b).

Youth unemployment has historically been high in Spain, and has strongly increased since 2007,

reaching a peak of 55.4% in 2013. In 2014 and 2015, the rate started falling but at 46% (Q1 2016), it still

more than twice the rate at the beginning of the crisis (18% on average in 2007), and the second-highest in

the EU after Greece. Very high unemployment rates among youth in Spain are mainly the result of scarce

job opportunities. Similarly, the number of young people aged 15-24 not in employment, education or

training (NEET) has increased between 2007 (12%) and 2015 (16%) and is now significantly above the EU

average. Moreover, more than a third of those who are NEET do not live with anyone in paid work and are

hence at risk of poverty, and half have low skills (OECD, 2016b). The Spanish government has introduced

the National Youth Guarantee scheme which gives those enrolled an offer of employment or training, but

this has not yet delivered the expected results (European Commission, 2016). One million youth are

expected to benefit from this programme, however, up to February 2015, only 211 000 youth were

registered and of those only about 60 000 found a job.

Unemployment in Spain also shows a distinct geographical pattern with much higher rates in the

South (30% in 2015) and relatively low rates in the North-East (15%).5

As unemployment has remained very high for an extended period of time, rates of long-term have also

increased, peaking in 2013 at 13% of the economically active population. It has fallen to 11% in 2015, but

remains at more than twice the EU average. Very-long-term unemployment (more than two years) only

started to fall later, and has fallen more slowly. The very-long term unemployment rate was 7.6% in 2015,

close to three times the EU average.

Income support for the unemployed in Spain mainly consists of a general national contributory

scheme and a non-contributory benefit for those who are not eligible for the contributory one. (There are

also some minor unemployment benefit schemes for specific categories of worker). Entitlement to

unemployment benefits was tightened, especially for the long-term unemployed, as part of the structural

labour-market reforms introduced in 2012. As a result of persistent long-term unemployment, a large share

of jobseekers are no longer entitled to unemployment benefits. In 2007, about 77% of the unemployed in

Spain received income support (either contributory or non-contributory), but by 2014 only 37% were

covered.6 In addition, minimum-income support schemes remain relatively fragmented and poorly

co-ordinated across geographic areas, institutions and levels of government. They are administrated at local

level, with large regional disparities and low coverage (European Commission 2016). Partly as a result,

poverty risks for those who lose their jobs are high. A lack of accessible income support also encourages

the unemployed to take precarious jobs or jobs for which they are overqualified rather than waiting to find

a job that matches their abilities and qualifications.

Incidence of economic hardship

The combination of high levels of unemployment and high rates of poverty among the unemployed

produces a very significant challenge in Spain. 23% of working-age individuals are at-risk of poverty, the

highest rate among the six countries studied in this project, and well above the EU average (17%). Poverty

rates are high among those in work too, and particularly among households with children. As well as

having low employment rates at the individual level, Spain also has a high proportion of working-age

adults living in households with very low work intensity: many of those who are not in employment live in

workless households. Although rates of severe material deprivation are relatively low in Spain (below the

EU average), the proportion of working-age adults at risk of poverty or social exclusion (AROPE) is,

5. Source: Eurostat Labour Force Survey, regional statistics.

6. Source: OECD Benefit Recipients database SOCR (http://www.oecd.org/social/recipients.htm).

http://www.oecd.org/social/recipients.htm


again, the highest of the six countries studied (see Table 2.1 below). Spain is also one of the most unequal

countries in the EU with the fourth-highest Gini coefficient for disposable income (34.7 in 2014),

following a very sharp increase since 2012.

Table 2.1. Risk of poverty or social exclusion, 2014,

In % of people aged 16-64

1. Individuals aged 18-64. 2. Individuals aged 18-59.

Source: Eurostat (EU-SILC 2014).

Target groups for activation and employment-support policies

Individuals with labour market difficulties frequently move between non-employment and different

states of “precarious” employment. As a result, limiting attention to “snapshots” of non-employed (or

underemployed) individuals at a specific point in time, such as those based on labour force surveys, may

not capture the true extent of labour-market difficulties or the need for policy intervention. To cover the

potential scope of AESPs, the target population of the analysis in this note therefore includes working-

age individuals who are “persistently” out of work (either unemployed or labour-market inactive for more

than 12 consecutive months) as well as individuals whose labour-market attachment is “weak”.7 “Weak”

labour-market attachment can include individuals with unstable jobs working only sporadically, those on

restricted working hours, and those with very low earnings (due to, for example, working informally or in

very low productivity self-employment). Box 2.1 defines the sub-groups of this population and explains

how they are identified using the EU-SILC data. The target population is a sub-set of the reference

population of working-age adults relevant for AESPs. The reference population, in turn, is defined as all

working-age adults except for full-time students and those in compulsory military service as these groups

are typically outside the scope of AESPs. For simplicity, the rest of this note also refers to this reference

group as the “working-age population”.

7. This note does not attempt to distinguish between voluntary and involuntary joblessness or reduced work

intensity. Individuals can of course choose to be out of work, or in part-time or part-year employment,

voluntarily, and some surveys ask respondents whether they “want to work”. However, those saying they

do not want employment, or prefer to work part-time or part-year, may do so as a result of employment

barriers they face, such as care obligations or weak financial incentives, which policy might potentially

address. If extended voluntary labour-market inactivity or underemployment creates or exacerbate certain

types of employment barriers, it may subsequently give rise to involuntary labour-market detachment or

partial employment in later periods.

Spain Estonia Ireland Italy Lithuania Portugal EU28

People at risk of poverty or social exclusion 32 25 29 29 26 28 25

People at risk of poverty

All 23 20 17 20 18 19 17

Not working 36 36 31 31 35 32 31

Working 13 12 6 11 8 11 10

full-time 10 11 3 10 7 9 8

part-time 23 20 11 17 24 31 16

Households without children 16 25 15 16 18 16 15

Households with children 28 18 16 24 20 23 19

People living in households with severe material deprivation (1)

All 8 6 9 12 12 10 9

Households without children 6 7 6 10 16 10 8

Households with children 9 5 10 13 12 11 10

People living in households with very low work intensity (2)18 8 21 13 9 13 12


Clearly, not everybody experiencing potential labour market difficulties may be an intended target for

AESPs.8 The broad definition of labour market difficulties adopted in this note is not intended to be

prescriptive about the appropriate scope of AESPs; instead, it seeks to inform policy decisions by

documenting the employment barriers and circumstances of individuals with no or weak labour market

attachment. The approach is thus descriptive and takes no position on whether policy intervention is

justified for specific groups. The resulting profiles of employment barriers are intended to facilitate discussions

of the strengths and limitations of different policy interventions for concrete groups of individuals. They can

also be used to help inform decisions on whether to channel additional policy efforts towards specific priority

groups.

Box 2.1. Individuals with potential labour market difficulties (target population)

The target population in this note includes those who are persistently out-of-work, as well as those with weak labour-market attachment.

The persistently out-of-work population (long-term unemployed or inactive) includes individuals reporting no employment activity throughout the reference period. The reference period corresponds to 12 consecutive monthly observations in the income reference year (January-December of year T-1) plus one additional observation at the moment of the interview (in year T).

The group with weak labour market attachment refers to individuals reporting employment activity during the reference period matching any of the following three situations:

i) Unstable jobs: individuals working only a limited number of months throughout the reference period. The

threshold is equivalent to Eurostat’s low-work-intensity measure: Above zero but no more than 45% of potential working time in the income reference year. To reconcile information reported for the income reference period and at the moment of the interview the following individuals are also considered in this group: 1) Workers who report no work activity during the income reference period but who are working at the moment of the interview and, 2) workers with between 45% and 50% of work activity during the income reference period who do not report any work activity in either the last month of the income reference period or at the moment of the interview.

ii) Restricted hours: workers who spent most or all of the reference period working 20 hours or less a week.

Error! Reference source not found. However, individuals working 20 hours or less who are not likely to have

dditional work capacity, e.g. due to ongoing education or training, are excluded.

iii) Near-zero earnings: individuals reporting some work activity during the income reference period but

negative, zero or near-zero monthly earnings.2 In addition to possible classification error, situations included

in this group could signal potential labour market difficulties, such as underpayment and/or informal activities.

1. The 20-hours threshold is approximately in-line with the 45% “part-year” threshold that identifies the group with unstable jobs. For a 40-hours working week in a full-time job, 45% of full-time would correspond to 18 hours a week. However, in SILC, the distribution of working hours in the main job shows a high degree of bunching at 10, 15, 20 and 25 hours a week. As the closest multiple of 5, a value of 20 hours was therefore chosen.

2. The near-zero earnings threshold is set in Spain at EUR 111/month. This value corresponds broadly to the 1st percentile of the

SILC earnings distribution.

Figure 2.2 shows the evolution of the target population in Spain between SILC survey years 2008 and

2014 (since the reference period is the year prior to the interview, these data refer to the period 2007 to

2013). Despite the major definitional differences, the resulting patterns are consistent with the trends based

on LFS data shown earlier in Figure 1. As was shown previously, the recovery from the crisis has been

particularly slow in Spain: labour market indicators started to improve only in 2014. Since the last

8. It is worth noting that, with a definition of working-age as 18-64, some individuals whom policy makers

may wish to include in the scope of AESPs are not included in the target group in this note. Although the

18-64 age cut-offs are common in comparative empirical work, they are becoming less suitable as

populations age, especially in countries that are actively seeking to increase retirement ages beyond 65.


available SILC year is 2014 (which mostly reflects individuals’ labour status in 2013), this mild recovery is

not yet reflected in the results. Consequently, Figure 2.2 shows an uninterrupted rise in the number of

unemployed, inactive and workers weakly linked to the labour market between 2007 and 2013 (SILC years

2008 and 2014). In view of the timing of Spain’s labour-market recovery, it is important to keep in mind

that the cut-off for all SILC-results in this note is 2013–14.

Figure 2.3 shows the size and composition of the target population in SILC 2014. At 45%, the group

with potential labour-market difficulties is larger in Spain than in the other countries considered in this

project with the exception of Ireland. Two thirds of the target population were out of work throughout the

reference period, the most common status was unemployment (36% of the target population). 17%

reported that they were engaged in domestic tasks and 7% that they were unfit to work. The majority of

individuals with “weak labour market attachment” (underemployment) spent part of the year out of the

labour force (unstable jobs), although there are also sizeable groups of individuals who worked part-time

throughout the year (8% of the target population) or who report working throughout the year but have very

little earnings (5% of the target population).

Figure 2.2. Population groups with potential labour market difficulties

In % of reference population

Source: Calculations based on EU-SILC 2008-2014. See Box 2.1 for the definitions of the three groups.

17 17 16 17 16 14 14

4 6 1012 14 15 16

10 1113

11 12 14 15

0

10

20

30

40

50

2008 2009 2010 2011 2012 2013 2014

% of the reference population

Inactive Unemployed Underemployed

Persistently out of work


Figure 2.3. Composition of the Spanish population with labour market difficulties

Note: The six-country average is unweighted.

Source: Calculations based on EU-SILC 2014. See Box 2.1 for the definitions of the three groups.

Persistently out of work (67% of the target population)

Weak labour market attachment (33% of target population)

Unemplo-yed (36%)

Retired (6%)

Unfit to work (7%)

Domestic tasks (17%)

Other inactive

(2%)

61%55%

12%15%

27% 30%

average of sixcountries ESP

Persistently out of work

Weak labour market attachment

No major difficulties

Restricted hours (8%)

Near-zeroearnings (5%)

Unstable jobs

(25%)

"Target" population

(45%)

Working age population

(100%)


3. EMPLOYMENT BARRIERS IN SPAIN

Working-age individuals with no or weak labour-market attachment may face a number of

employment barriers that prevent them from fully engaging in employment activities. A thorough

understanding of these barriers is a pre-requisite for designing and implementing policy interventions in a

way that is well-targeted and suitably adapted to the circumstances of different policy clients. Following

Immervoll and Scarpetta (2012), this note examines three types of employment barrier, namely (see

Figure 3.1):

Insufficient work-related capabilities , e.g. a lack of skills, work experience, care responsibilities and

health-related limitations;

Lack of financial work incentive to look for a “good” job, e.g., because of low potential pay,

relatively generous out-of-work benefits, or access to high levels of income independent of their own

work effort (such as capital income or earnings of other family members);

Scarce job opportunities, e.g., a shortage of vacancies in the relevant labour-market segment due to

shocks or cyclical factors, or because of skills mismatch, discrimination, dual labour markets or other

frictions in the labour market.

Figure 3.1. Employment barriers: conceptual framework

Source: Fernandez et al. (2016).

The employment barriers outlined above cannot all be measured directly. To operationalise the

concepts, this note implements a set of workable indicators in each of the three main categories. Fernandez

et al. (2016) provides a fuller discussion of the indicators and their rationale, including descriptive statistics

for selected countries. The indicators used for Spain are as follows:

Employment barrierlack of job opportunities

Employment barrierlack of work-related capabilities

Better-qualityemployment

Policy interventions

Individuals with potential labour market

difficulties

Employment barrierlack of financial incentives

Jobless Unstable jobs

Restricted working hours


Capability, item 1. “Low” skills: if an individual has low professional skills (their most recent

job was in the lowest two categories of the ISCO-08 classification system).9 Those who

demonstrate high skills by having a tertiary degree are assumed not to face this employment

barrier even if their most recent job was low-skilled. If an individual has no work experience at

all, they are also included in the “low skills” group.

Capability, item 2. Two measures of work experience:

No recent work experience: if an individual did no paid work during the reference period

(i.e. they were without employment for at least 12 months).

“Low” relative total work experience: the indicator takes one of three values: 1 for those

who have no past work experience at all, 2 for those who have some work experience but

have worked less than 60% of the time since they left full-time education, and 3 otherwise

(i.e., if their total work experience is not “low”).

Capability, item 3. Health limitations: if an individual reports some or severe long-standing

physical or mental limitations in daily activities.

Capability, item 4. Care responsibilities: if an individual has a (minor or adult) family member

who requires care10

and is either the only potential care giver in the household, or the only person

in the household who is economically inactive or working part-time because of care

responsibilities.

Incentives, item 1. “High” non-labour income: if household income other than that relating to

the work efforts of the individual in question,11

, is more than 1.4 times the median value in the

working-age population (EUR 12 156/year, adjusted for household size).

Incentives, item 2. “High” earnings-replacement benefits: if an individual’s earnings-

replacement benefits received during the reference year exceed 60% of their estimated potential

earnings in work.12

Opportunity (one item only). “Scarce” job opportunities: if an individual has a “high” risk of

not finding a job despite active job-search during at least seven months, and willingness to take

up employment (as stated at the moment of the SILC interview). The risk is estimated in a

regression including region, age group, gender, level of professional skills and education as

independent variables and being long term unemployed or involuntarily working part time as the

dependent variable (see Fernandez et al., 2016 for more details). Individuals with an estimated

risk of more than 1.6 times the median value in the working-age population are considered to

face “scarce” job opportunities. Scarce job opportunities not only present a barrier to employment

in the short term, but if jobseekers become discouraged and stop active job search, it could lead

to further problems in the longer run.

9. This indicator is different from that in Fernandez et al. (2016), which classifies individuals who have

achieved less than upper secondary education as facing an employment barrier. Given the extent of skills

mismatches in the Spanish labour market discussed in Section 2, it was felt that low levels of professional

skills were more likely to present a barrier to employment than low education.

10. Family members assumed to require care are children under the age of 12 receiving less than 30 hours of

non-parental childcare a week and adults reporting severe limitations in daily activities due to their health

and being economically inactive throughout the reference period (and in the case of those of working age,

that permanent disability is the reason for their inactivity).

11. This includes earnings, individual-level earnings replacement benefits, and the individual’s share of

household-level earnings replacement benefits.

12. Potential earnings are estimated in SILC with a regression model corrected for sample selection. See

Fernandez et al. (2016) for details.


Table 3.1 shows the shares of individuals in the target and the broader reference populations facing

each employment barrier. As expected, the incidence of each barrier is significantly higher in the target

population. In most cases, barriers are also more prevalent among those who were out of work throughout

the entire reference period than for those with weak labour-market attachment. Common barriers in Spain

include low skills, scarce job opportunities and low relative total work experience. These barriers are each

faced by nearly half of the target population. The patterns are broadly consistent with the Spanish labour-

market context discussed in Section 2 (high unemployment and long-term unemployment are high, scarce

job opportunities and barriers to job reallocation, and low skills among a significant part of the population).

A special case is the “no recent work experience” barrier, which not only acts as a potential employment

obstacle but also is a direct result of the way the target population is defined: by definition, those who were

persistently out of work did not work at all during the reference period. As a result, 100% of this group are

shown as facing “no recent work activity” as a potential barrier.

The other employment barriers, in particular care responsibilities and health limitations, are somewhat

less prevalent overall, but may still be very important for some sub-groups. For instance, 12% of the target

population receive high levels of earnings replacement benefits, and a similar percentage has no work

experience at all.

By construction, the underemployed and the “persistently out of work” have different levels of the

work experience barriers. But for most other barriers, the incidence is not very different between the out-

of-work and the underemployed groups. This suggests that both sub-groups face quite similar employment

difficulties that AESPs could address. The one exception is health limitations, which are more common

among the out-of-work group, perhaps because health limitations are in many cases severe enough that

individuals are unable to undertake any paid work, or because benefit entitlement conditions do not

encourage employment for this group.

In practice people’s individual and family circumstances are complex and often lead to situations

where they face multiple barriers to employment. Figure 3.2 shows the number of (simultaneous) barriers

faced by individuals in the target population. Only 17% of the target population face only a single barrier

to employment. One third face two simultaneous barriers, another third face three, and 13% face at least

four. Only 4% face no major employment barrier. For this group, the employment-barrier indicator may be

slightly below the respective thresholds used in this note, or they are not working or underemployed for

reasons unrelated to the barriers discussed here. They may face other barriers, or they may simply have a

strong preference for leisure. The next section uses a statistical clustering technique to examine which

combinations of barriers are most common.

Table 3.1. Employment-barrier indicators

% of population facing different types of barrier

Note: See text for definitions and thresholds.

Source: Calculations based on EU-SILC 2014.

AllPersistently out

of work

Weak labour

market attachment

Insufficient work-related capabilities

"low" education or professional skil ls 37 52 57 43

No professional skil ls (no past work experience) 6 12 18 0

Positive but low relative work experience 27 44 44 42

No recent work activity 30 67 100 0

"Some" but "low" recent work activity 13 29 0 89

Health limitations 17 25 30 14

Care responsabilities 7 14 15 12

Lack of financial work incentives

"High" non-labour income 32 30 31 28

"High" earnings replacements 7 12 13 10

Scarce job opportunities

Scarce job opportunities 21 46 46 46

"Target" populationWorking age

population


Figure 3.2. Number of simultaneous barriers

% of target population

Note: The six-country average is unweighted.


13

32

33

17

4

4 or more barriers 3 barriers 2 barriers single barrier No major barrier

13

28

32

20

6

Spain average of six countries


4. FACES OF JOBLESSNESS IN SPAIN

This section applies the method described in Fernandez et al. (2016) to segment the target population

into groups of individuals with similar combinations of employment barriers. Using the 2014 SILC data

for Spain, the segmentation process leads to the identification of 13 groups of individuals with no or weak

labour market attachment (the “target population”).13

The following paragraphs describe each group in detail. At the end of each paragraph a box reports a

Venn diagram showing extent and degree of overlap of the main barriers characterising the group, as well

as a list of selected individual and household characteristics with a “high” probability of occurring in the

group. Together, this information can help attach suitable labels (“faces”) to group members, although the

labels are necessarily arbitrary to some extent and cannot substitute for careful examination of the

comprehensive list of employment barriers and socio-economic characteristics, as reported in Annex

Tables A.1 and A.2.

Group 1 (14% of the target population): “Labour-market inactive women with low education and

weak financial incentives”. This group consists of older (average age 56) women (100%) who were

economically inactive throughout the reference period (80%). 66% of them have worked before (on

average for 14 years) but for 59% of them this work experience is low relative to their potential. Another

common employment barrier characterising this group is low skills (72%). The group has the lowest

average level of education of all the groups (8.6 years) which is also often associated with low professional

skills: almost all of this group’s previous employment was at the skill level of clerks and sales people or

lower. Individuals in this group often live in households with one or more working adult (56%) and thus

can draw from significant income that does not depend on their own work effort (52%). On average, the

group face 2.7 simultaneous employment barriers, with the most common being low skills, low work

experience and weak work incentives resulting from high non-labour incomes in the household.

Box 4.1. Group 1: “Labour-market inactive women with low education and weak financial incentives”

Main employment barriers(1)

Selected characteristics(2)

% of the

target pop.

- 56 years old (average) - Women - Inactive - 14 years of paid work experience - 8.6 years of schooling (average) - Average equivalent disposable income: EUR 15 022 (most likely

to be in third and fourth income quintiles)

- 2.7 simultaneous employment barriers

1. Surface areas of shapes in the diagram are proportional to the number of group members facing the related barrier (“Proportional Venn Diagrams”). The outer square represents the group size (100%). The diagram shows the three most prevalent barriers in the group and is based on the indicators discussed in Section 3. An exception is the recent work experience indicator. Although this indicator is included in the numerical results in Annex Table A.1, it is not shown in the diagrams as its high prevalence (due to the strong two way causal link with the other barriers) would dominate all other barriers in the graphical representation in all but two groups.

2).Characteristics that distinguish this group from other groups, i.e., categories that have a high probability of occurring in the group. Table A.2 reports individual and household characteristics in more detail.

3. Income quintiles are calculated for the entire national population.

Source: Calculations based on EU-SILC 2014, see Annex Tables A.1 and A.2 for full results.

13. Annex A outlines the segmentation method and the process that lead to the identification of the 13 groups.

Fernandez et al. (2016) describes in detail the econometric model and the related methodological framework.


Group 2 (13% of the target population): “Unemployed prime age adults with low work experience”.

This group is made up of prime age (average age 35) unemployed individuals. 99% were unemployed

during the reference period, and 83% were still unemployed at the time of interview. The majority are

actively looking for work (80% were actively seeking a job at the time of the interview) but they struggled

to find work due to a scarcity of job opportunities. Although everyone in this group has some past

employment record, for 79% of them this is low relative to their age and education. Another common

obstacle to finding work is their level of skills, which are low for 59% of the group: most do not have an

upper secondary education and almost all previously worked in jobs at the skill level of clerks and sales

people or lower. Most individuals in this group face at least two of these three employment barriers (the

group average is 2.7 simultaneous barriers, as in Group 1).

Box 4.2. Group 2: “Unemployed prime age adults with low work experience”

Main employment barriers Selected characteristics % of the

target pop.

- 35 years old (average) - Unemployed (average spell – 11.8 months) - 9 years of paid work experience - 10.2 years of schooling (average) - At risk of poverty - Average equivalent disposable income: EUR 8 709 (mostly in the

bottom income quintile)


Group 3 (11% of the target population): “Experienced but low-skilled unemployed men”. The

majority of this group are prime age (47 years on average) men (76%) with significant past work

experience (25 years on average), who were unemployed for the majority of the reference period (94%). At

the time of the interview 86% remained unemployed despite in most cases (82%) actively seeking work.

The most common barrier to employment this group faces therefore is scarce job opportunities (90%). The

only other common barrier for individuals in this group is low skills which is an obstacle for 61% of the

group. 69% received unemployment benefits (EUR 5 300/year, on average), but nevertheless individuals in

this group have the lowest average equivalent disposable income (EUR 8 630/year) and 55% are at risk of

poverty. The characteristics of this group are similar to those of Group 2, the main differences being that

this group is older and has greater work experience.

Box 4.3. Group 3: “Experienced but low-skilled unemployed men”


target pop.

- 47 years old (average)

- Men

- Unemployed (average spell – 12.3 months)

- 25 years of paid work experience

- 10.2 years of schooling (average)

- At risk of poverty

- Average equivalent disposable income: EUR 8 630 (bottom income quintile)


Group 4 (11% of the target population): “Well-educated prime age adults with weak labour market

attachment”. Individuals in this group are of prime working age (average age 33) with a recent

employment record (96%). However, their attachment to the labour market is weak: 57% only worked for

part of the reference period and 33% worked less than 20 hours a week for most of the reference period.

This suggests that they are subject to marginal employment and are unable to secure jobs with permanent

Education/Skills(59%)

Low work experience

(79%)

Opportunities(100%)


Opportunities(90%)


contracts that are subject to much stricter employment protection. Individuals in this group typically

worked at clerk and sales skill level (55%) or higher (25%), and have the highest education levels of all

groups. This group has on average 1.2 simultaneous employment obstacles, the lowest of all 13 groups

(Figure 4.1). This suggests that either employment barriers each affect employment possibilities

independently, or that members of this group were work ready and unemployed for only a fairly short

period of time during the reference period: indeed, 81% of this group were in employment by the time of

interview. The main two barriers characterising this group are low work experience relative to potential

experience (37%) and high levels of household income that do not directly depend on their own work

effort (33%).

Box 4.4. Group 4: “Well-educated prime age adults with weak labour market attachment”


target pop.


- Employed


- 13 years of schooling (average)

- Average equivalent disposable income: EUR 11 483 (bottom two income quintiles)


Group 5 (9% of the target population): “Early retirees with weak financial incentives”. This group

is relatively old (average age 61 years) and the majority are men (73%). They have considerable paid work

experience (37 years on average) and have the highest equivalent disposable income of the 13 groups

(EUR 18 477/year on average). Most previously worked at the skill level of craft and machine operators or

clerks and sales people, and their level of education is no different to the average among the target

population of 10.4 years. The proportion facing the low skills barrier is thus 41%. Most members of this

group are labour-market inactive (77%), with 48% describing themselves as retired, and 17% as being unfit

for work. 40% can draw on income sources that are independent to their own work effort, which in most

cases are old-age benefits (37% receive an average of EUR 21 934/year). Moreover, 31% receive sickness

and disability benefits (EUR 15 594/year on average) and 26% receive unemployment benefits

(EUR 8 678/year on average). For 41% of the group these earnings replacements benefits are high relative

to their potential earnings in work, which further weakens financial incentive to undertake paid work. This

group are less likely to face multiple simultaneous employment barriers (see Figure 4.1) than other groups,

but 56% face more than one barrier, the most common combination being low education and skills and

weak financial work incentives.

Box 4.5. Group 5 “Early retirees with weak financial incentives”

Non-labour incomes

(33%)

Low work experience

(37%)


target pop.


- Men

- Retired/Inactive



- Average equivalent disposable income: EUR 18 477 (top two income quintiles)


Non-labour incomes

(40%)

Education / Skills(41%)

Earningsreplacement

(41%)


Group 6 (7% of the target population): “Unemployed women with low work experience”. This

group consists mostly of women (67%) in prime working age (average age 49), who were unemployed for

the majority of the reference period (70%) and still unemployed at the time of the interview (68%).

However, only 56% of the group reported that they were actively looking for work at the time of the

interview, indicating that they had become discouraged from looking for work after a long spell of

unemployment: these unemployed individuals have the longest average unemployment spell of any group

(12.9 months). Although 96% have worked in the past, for 87% of them work experience is low given their

age with average work experience among this group being 12 years. Other barriers often faced by this

group are scarcity of job opportunities (55%) and low skills (41%): most members of this group do not

have an upper secondary education, and most previous employment was at the skill level of clerks and

sales people or lower. Most individuals in this group face the low work experience barrier and one of these

other barriers, and some face all three (see Box 4.6 below). The average number of barriers faced by this

group is 2.5. Despite being mostly unemployed, only 35% of this group receive unemployment benefits,

and the amount received is relatively low at EUR 3 896 per year. As a result, 54% of this group are at risk

of poverty.

Box 4.6. Group 6 “Unemployed women with low work experience”

Group 7 (6% of the target population): “Low-skilled women in unstable jobs”. Individuals in this

group are women (80%) of prime working age (average age 44) with a recent employment record (93%).

However, this group’s labour market attachment is weak: 67% worked for only part of the reference period

and 35% worked less than 20 hours a week for most of the reference period. This suggests that this group

are subject to marginal employment and are unable to secure jobs with a permanent contract that are

subject to much stricter employment protection legislation. Moreover, many of this group have an

incomplete employment record: 46% have low levels of work experience relative to their potential. The

largest barrier to employment in the group is a combination of low levels of education (third lowest

average level of education among the 13 groups at 9.6 years) and low professional skills (83% previously

worked at the craft and machine operator skill level or lower). This group has on average 2 simultaneous

employment obstacles, with low work experience relative to potential experience (46%) frequently

overlapping with low skills (81%).


target pop.


- Majority Women

- Unemployed (average spell – 12.9 months)




- Average equivalent disposable income: EUR 8 927 (bottom income quintile)


Education / Skills(41%)

Low work experience

(87%)

Opportunities(55%)


Box 4.7. Group 7 “Low-skilled women in unstable jobs”

Group 8 (6% of the target population): “Labour-market inactive mothers with low work

experience”. Individuals in this group are prime age (average age 40) women living in families with a

partner who is in paid work (80%) and their young children (99%). These persons have on average

1.5 young children with the youngest being five years old. The majority report being economically inactive

(62%) at the time of interview. 16% report being unemployed (16%), but only 9% were actively seeking a

job at the time of the interview. The most common barrier to employment is that most (79%) have care

responsibilities for their young children. The other main barrier is that although most of these individuals

have past work experience (81%), for many (63%) this is low relative to their age and education level. The

average number of simultaneous employment barriers is 2.6: in addition to these two main barriers, some

members of this group face scarce job opportunities (35%), have high levels of non-labour income (16%)

or health limitations (12%).

Box 4.8. Group 8 “Labour-market inactive mothers with low work experience”

Group 9 (5% of the target population): “Low-skilled individuals with health problems and high

levels of earnings-replacement benefits”. These individuals are older (average age 52) and 99% report a

long-standing physical or mental limitation, of which 37% are severe limitations. 69% receive sickness and

disability benefits (EUR 14 022/year, on average) and for 45% of this group, these benefits are high

relative to their potential earnings in work, which could weaken their financial incentives to seek

employment. Partly due to their health, individuals in this group are currently largely labour-market

inactive (84%) but most (85%) have some previous work experience. The group has the second lowest

level of education (8.9 years on average) and this creates a barrier to re-employment as 74% have attained

only lower secondary education or below.

Main employment barriers Most frequent characteristics % of the

target pop.


- Women

- Employed


- Lower Secondary education: 9.6 years of schooling (average)


- 2 simultaneous employment barriers


target pop.


- Women

- Inactive

- Couple with children





Low work experience

(46%)

Education / Skills

(81%)

Low work

experience(63%)

Care(79%)


Box 4.9. Group 9 “Low-skilled individuals with health problems and high levels of earnings-replacement benefits”


target pop.


- Inactive



- Average equivalent disposable income: EUR 13 399 (bottom three income quintiles)


Group 10 (5% of the target population): “Educated parents in short-term unemployment or

working part time”. This group is of prime working age (average age 39), has children (95%) and is

relatively well educated (43% at tertiary level, 12.8 years of education on average). The main employment

barriers they face are a lack of job opportunities (59%) and the need to care for children (58%). As such,

many of these individuals are work ready and indeed 58% did some work during the reference period, of

whom 17% worked part time for most of the reference period. Moreover, by the time of the interview 37%

were in employment, though the interrupted work patterns of this group suggest that this could be marginal

employment on temporary contracts. The average number of employment barriers faced by this group

is 2.1: in addition to these two main barriers some members of this group have high incomes from sources

that are not related to their own work effort (35%), low skills (21%) and high levels of earnings

replacement benefits (17%).

Box 4.10. Group 10 “Educated parents in short-term unemployment or working part time”


target pop.


- Unemployed (average spell - 11.2 months)



- Average equivalent disposable income: EUR 12 844 (bottom three income quintiles)


Group 11 (5% of the target population): “Unemployed youth without any past work experience

facing scarce job opportunities” This group consists of young people (average age 24) most of whom were

unemployed during the reference period (82%). Between the reference period and the interview only 9%

had found employment, despite their high levels of motivation (65% of the group were actively seeking a

job at the time of the interview). 71% were unemployed throughout the reference period and at the time of

the interview. This high level of unemployment suggests a lack of job opportunities, which is an

employment obstacle for everyone in the group. Low skills (64%) are another challenge to finding

employment that the group faces. Individuals in this group are likely to face three or more simultaneous

employment obstacles, the second highest among the 13 groups (Figure 4.1). Benefit coverage among this

group is low, with the result that 48% of this group are at risk of poverty and 38% face material

deprivation.

Health(99%)


Earnings replacement

(45%)

Care(58%)

Opportunities(59%)


Box 4.11. Group 11 “Unemployed youth without any past work experience facing scarce job opportunities”


target pop.


- Unemployed (average spell - 12 months)

- No past paid work experience

- Lower secondary education: 10.7 years of schooling (average)



Group 12 (4% of the target population): “Unemployed mothers actively looking for work but facing

scarce job opportunities”. This group consists mostly of women (85%) of prime working age (average

age 35), who were unemployed for the majority of the reference period (82%). These individuals all have

young children (the youngest is four years old on average) and most live with their partner who is

working (77%). Care responsibilities are therefore a barrier for 90% of the group. They have high

motivation (75% were actively seeking a job at the time of the interview) but struggle to find work due to

scarce job opportunities (100%). Although 83% of the group have worked before, for 56% of them their

work experience is low relative to their age and education level. Another obstacle is low professional

skills (63%): most do not have an upper secondary education and in almost all cases previous employment

was at the skill level of clerks and sales people or lower. These obstacles are often faced simultaneously

and thus this group has the highest number of simultaneous barriers of any of the 13 groups (3.6 on

average). This group also contains the highest proportion of migrants of any of the groups (38%).

Box 4.12. Group 12 “Unemployed mothers actively looking for work but facing scarce job opportunities”


target pop.


- Women

- Unemployed

- Couple with children






Group 13 (4% of the target population): “Short-term unemployed men with high earnings-

replacement benefits facing scarce job opportunities”. The majority in this group are prime age (average

41 years old) men (67%) who were unemployed during most of the reference period (94%). However, 69%

did at least some paid work during the reference period, suggesting that they had only recently been made

redundant. The group has the second highest equivalent disposable income of the 13 groups

(EUR 15 792/year on average). The most common employment barrier they face is a lack of job

opportunities (92%). Although members of the group have a long employment record (19 years on

average) their previous jobs have been in low-skill occupations (56% at craft and machine operator skill

level or lower), which is also likely to worsen their re-employment prospects. 81% received unemployment

benefits (EUR 11 306/year, on average) and for 67% of this group, these benefits were high relative to

their potential earnings in work, which could weaken their financial incentives to seek or take up

employment. These barriers are often faced simultaneously: the average number of barriers faced by

members of this group is 2.8.

Opportunities(100%)

Education /Skills

(64%)

No work experience

(71%)


Box 4.13. Group 13 “Short-term unemployed men with high earnings-replacement benefits facing scarce job opportunities”


target pop.


- Men

- Unemployed



- Average equivalent disposable income: EUR 15 792(bottom three income quintiles)


Figure 4.1. Share of individuals facing multiple employment barriers in each group

In descending order of shares facing at least three barriers

Note: Group sizes are reported on the horizontal axis. See Box 2.1 to Box 2.11 for details.


13

32

33

17

4

0

25

50

75

100

4 5 5 4 13 14 6 7 5 6 11 9 9 100

Group 12 Group 11 Group 9 Group 13 Group 2 Group 1 Group 8 Group 6 Group 10 Group 7 Group 3 Group 5 Group 4 Targetpop.

4 or more barriers 3 barriers 2 barriers single barrier no major barrier


5. CONCLUSIONS

This note has used a novel method for identifying, analysing and visualising the most common

employment barrier profiles characterising the Spanish population with potential labour market difficulties.

The underlying premise is that out-of-work individuals (unemployed and inactive) and workers with weak

labour market attachment face a number of possible employment obstacles, and each of them may call for

different policy responses. The success of activation and employment-support policies (AESPs), and of

social protection measures more generally, is expected to hinge on effective strategies to target and tailor

policy interventions to these barriers and to individual circumstances.

The segmentation method uncovers patterns that can provide concrete guidance for policy design and

targeting strategies in Spain. Results show that “short-hand” groupings that are often referred to in the

policy debate, such as “youth”, “women”, “unemployed”, are far from homogeneous, and may distract

attention from the specific employment obstacles that policies seek to address. Indeed, some of these

categories include several distinct sub-groups with very different combinations of employment barriers.

With very high levels of unemployment, it is perhaps unsurprising that the “faces of joblessness” are

quite diverse in Spain. Unemployed differ in their employment histories, skill levels, care responsibilities

and benefit entitlements (Groups 2, 3, 10, 12 and 13). Clearly, support offered to the unemployed should

be carefully tailored to these diverse barriers and needs. Youth unemployment has become a particularly

pressing challenge in Spain in recent years, and the statistical clustering approach identifies two very

distinct sub-groups of young people with labour-market difficulties (Groups 4 and 11):

Individuals in Group 4 are more educated and skilled, but mostly engaged in precarious or

unstable jobs. These youth are weakly attached to the labour market (they have some work

experience) and do not face major employment barriers. Most of them are not poor and live in

households where one or more adults work (probably their parents) so they can rely to some

extent on the support of their families. Although this group would likely be in stable employment

if the labour market in Spain were more dynamic, continued marginal employment may erode

their skills or motivation, which could put them at risk of more severe labour-market problems

later on.

Individuals in Group 11 already face multiple and much more severe employment difficulties.

They are less educated, 71% of them have no work experience at all and they also face scarce job

opportunities. Without intensive upskilling and re-employment support, these young people risk

becoming completely disconnected from the labour market.

A focus on care responsibilities provides another example of the diversity of people’s circumstances

and employment barriers. Results from the clustering exercise point to three quite different groups of

parents (Groups 8, 10 and 12) who together represent 16% of the target population. The large majority of

them are prime-age women who were unemployed or inactive at the time of the interview. Care

responsibilities are prevalent in each of these three groups and many group members might respond to

policies that improve childcare provision. However, there are also important differences between the

groups:

Women in Group 8 are mostly labour-market inactive, have relatively little work experience, live

with a working partner and many of them are engaged in domestic housework. This group is


likely to require fairly intensive support to improve work-related skills and find suitable childcare

arrangements before they are able to return to work.

Barriers in Group 10 appear to be less severe. Group members are more educated, and almost

90% were employed or actively looking for a job at the moment of the interview. Their closer

connection to the labour market suggests that they may require less assistance and that many of

them could find stable employment if job opportunities were less scarce.

Individuals in Group 12 are mostly mothers and also appear more motivated to work than those

in Group 8. 61% of them were actively looking for a job at the moment of the interview – but job

opportunities appear very limited for this group and 63% of them are low-skilled and have no or

low work experience. For them, childcare provision alone would likely be ineffective, as

upskilling appears to be an important ingredient to better hiring prospects.

The clustering results are not designed to advocate a focus of AESPs on one group or another. They

do however highlight possible priorities for policy interventions. For instance, very high poverty risks, a

large number of young people or a strong over-representation of women in some groups may signal a need

to review whether existing targeting strategies meet governments’ social cohesion objectives. A high

poverty risk combined with low education or professional skills and scarce job opportunities may call for

caution in applying benefit sanctions (such as for Groups 3 and 11). By contrast, groups with relatively

high incomes and financial disincentives caused by high levels of income replacement benefits (such as

Group 13) may indicate scope for targeted benefit reductions or for tightening benefit eligibility conditions.

Information on the intensity and number of barriers faced by individuals can also provide guidance for

difficult policy decisions, such as those involving trade-offs between helping those in greatest need and

targeting those who are likely to be the most responsive to policy interventions. For example, it is

debatable whether resources should be channelled primarily to those with severe or multiple barriers who

are, in some sense, furthest from obtaining or holding a stable job or to groups with moderate employment

difficulties, for whom policy interventions may have a greater probability of success.

A forthcoming Country Policy Paper to be produced as part of this project will take stock of existing

policy measures for some of the groups identified here. Based on that policy inventory, it will seek to

analyse whether they are well-aligned with the employment barriers that have been identified in the present

note.


REFERENCES

Arias, O.S., C. Sánchez-Páramo, M.E. Dávalos, I. Santos, E.R. Tiongson, C. Grun, N. de Andrade Falcão,

G. Saiovici and C.A. Cancho (2014), Back to Work: Growing with Jobs in Europe and Central Asia,

Wolrd Bank, Washington, DC, Doi: 10.1596/978-0-8213-9910-1.

Card, D., J. Kluve and A.Weber (2010), “Active Labour Market Analysis Policy Evaluations: A Meta-

Analysis”, Economic Journal, No. 120.

Collins, L.M. and S.T. Lanza (2013), Latent Class and Latent Transition Analysis: With Applications in the

Social, Behavioral, and Health Sciences, Vol. 718, John Wiley & Sons.

ESTEP (2014), ES struktūrin s paramos poveikio gyvenimo kokybei, socialin s atskirties ir skurdo

mažinimui Lietuvoje vertinimas, Galutinė vertinimo ataskaita, p. 194.

Eurofound (2012a), “NEETS - Young People Not in Employment, Education or Training: Characteristics,

Costs and Policy Responses in Europe”, Publications Office of the European Union, Luxembourg.

Eurofound (2012b), “Fifth European Working Condition Survey”, Publications Office of the European

Union, Luxembourg.

European Commission (2016a), “Country Report: Lithuania 2016”, Commission Staff Working Document.

European Commission (2016b), “Employment and Social Developments in Europe”.

European Commission (2015), “Upskilling Unemployed Adults (aged 25 to 64): The Organisation,

Profiling and Targeting of Training Provision”, Publications Office of the European Union,

Luxembourg.

Fernandez , R., H. Immervoll, D. Pacifico and C. Thévenot (2016), “Faces of Joblessness. Characterising

Employment Barriers to Inform Policy”, Forthcoming SEM Working Paper, OECD, Paris.

Immervoll, H. and A. Isik-Dikmelik (2016), “Cooperation with the OECD on Assessing Activating and

Enabling Benefits and Services in the EU: OECD-World Bank Joint Methodological Report”,

unpublished report submitted to the European Commission, March.

Immervoll, H. and S. Scarpetta (2012), “Activation and Employment Support Policies in OECD Countries.

An Overview of Current Approaches”, IZA Journal of Labor Policy, Vol. 1(1), pp. 1-20.

OECD (2016a), Getting Skills Right: Assessing and Anticipating Changing Skill Needs, OECD Publishing,

Paris.

OECD (2016b), OECD Employment Outlook 2016, OECD Publishing, Paris.

OECD (2015a), “Activation Policies for More Inclusive Labour Markets”, in OECD Employment Outlook

2015, OECD Publishing, Paris, http://dx.doi.org/10.1787/empl_outlook-2015-7-en.

OECD (2015b), OECD Economic Surveys: Spain 2014, OECD Publishing, Paris.

OECD (2015c), Education at a Glance: OECD Indicators, OECD Publishing, Paris.

OECD (2015d), International Migration Outlook 2015, OECD Publishing, Paris.

OECD (2015e), “OECD Skills Strategy Diagnostic Report – Spain”, OECD Publishing, Paris.

http://dx.doi.org/10.1787/empl_outlook-2015-7-en


OECD (2014a), “The Crisis and its Aftermath: A ‘Stress Test’ for Societies and for Social Policies”,

Society at a Glance: OECD Indicators, OECD Publishing, Paris.

OECD (2014b), The 2012 Labour Market Reform in Spain: A Preliminary Assessment, OECD Publishing,

Paris.

OECD (2013a), “Activation Strategies for Stronger and More Inclusive Labour Markets in G20 Countries:

Key Policy Challenges and Good Practices”, G20 Task Force on Employment, Report prepared for

the G20 Summit in St. Petersburg, July, http://www.oecd.org/g20.

OECD (2013b), “Activating Jobseekers: Lessons from Seven OECD Countries”, in OECD Employment

Outlook 2013, OECD Publishing, Paris, http://dx.doi.org/10.1787/empl_outlook-2013-7-en.

OECD (2013c), OECD Skills Outlook 2013: First Results from the Survey of Adult Skills, OECD

Publishing, Paris.

OECD (2008), “Declaring Work or Staying Underground: Informal Employment in Seven OECD

Countries”, in OECD Employment Outlook 2008, OECD Publishing, Paris,

http://dx.doi.org/10.1787/empl_outlook-2008-4-en.

Schneider, F. (2015), “Size and Development of the Shadow Economy of 31 European and 5 Other OECD

Countries from 2003 to 2015: Different Developments”, mimeo,

http://www.econ.jku.at/members/Schneider/files/publications/2015/ShadEcEurope31.pdf.

Schwarz, G.E. (1978), “Estimating the Dimension of a Model”, Annals of Statistics, Vol. 6(2),

pp. 461-464, Doi:10.1214/aos/1176344136.

Vermunt J. K. and J. Magidson (2016), Technical Guide for Latent GOLD 5.1: Basic, Advanced, and

Syntax, Statistical Innovations Inc., Belmont, MA.

http://www.oecd.org/g20



http://www.econ.jku.at/members/Schneider/files/publications/2015/ShadEcEurope31.pdf


ANNEX A

LATENT CLASS RESULTS

Using the 2014 SILC data for Spain, the segmentation algorithm outlined in Annex B leads to a model

with 13 groups. Table A.1 shows the estimated parameters, i.e. the share of individuals facing the

employment barriers in each latent group and the related group size in the target population (first row).

Groups are ordered by size; colour shadings are used to highlight barriers with higher (dark blue) and

lower (light blue) frequencies in each group.

Table A.1. Latent class estimates

Percentage of individuals with selected characteristics, by group

Note: Section 3 describes the indicators and applicable thresholds. Group sizes refer to the target population as defined in Section 1. Colour shadings identify categories with high (dark blue) and lower (light blue) frequencies. Complementary categories (e.g. “high” skills) are omitted. Additional information on model selection and model specification is provided in Annex B.

Source: Authors’ calculations based on EU-SILC 2014.

Table A.2. Characterisation of the latent groups


Group 1 Group 2 Group 3 Group 4 Group 5 Group 6 Group 7 Group 8 Group 9 Group 10 Group 11 Group 12 Group 13Target

Pop

Group Size (Target population=100) 14 13 11 9 9 7 6 6 5 5 5 4 4 100

Low education or professional skills 72 59 61 11 41 44 81 42 60 21 64 63 52 52

No past work experience 34 0 0 2 0 4 7 19 15 0 71 17 0 12

Positive but low relative work experience 59 79 6 37 4 87 46 63 41 12 29 56 20 44

No recent work activity 92 67 80 4 89 98 7 74 100 42 81 83 31 67

Health limitations 30 7 18 17 39 38 12 17 99 8 21 9 16 25

Care responsabilities 6 3 3 0 3 7 1 79 2 58 4 90 1 14

High non-labour income 57 19 10 33 40 18 16 36 19 35 30 24 36 30

High earnings replacements 5 0 5 4 41 1 3 1 45 17 4 1 67 12

Scarce job opportunities 3 100 90 12 1 55 35 1 1 59 100 100 92 46

Core

indicators


Pop

Number of individuals (%) 14 13 11 9 9 7 6 6 5 5 5 4 4 100

Number of individuals (frequency) 1739 1609 1289 1150 1126 819 786 766 668 660 645 494 452 12204

Unstable jobs 5 34 18 57 6 2 67 17 0 44 22 17 66 25

Restricted working hours 5 0 1 33 2 0 35 16 0 15 0 0 2 8

Zero or near-zero earnings 2 1 2 24 4 0 14 4 0 6 0 1 3 5

Women* 100 44 24 53 27 67 80 100 41 62 51 85 33 59

Youth 0 32 0 40 0 8 6 8 1 4 86 28 18 16

Prime age 44 68 79 59 10 63 84 88 67 96 14 71 74 60

Old-age 56 1 21 1 90 29 10 5 32 1 0 1 9 24

Age (average) 56 35 47 33 61 49 44 40 52 39 24 35 41 45

Employed FT 0 0 0 6 0 0 3 0 0 2 0 0 0 1

Employed PT 5 0 1 40 2 0 41 16 0 17 0 0 2 10

Self-employed FT 1 0 1 14 4 0 4 0 0 2 0 0 1 2Self-employed PT 0 0 0 2 0 0 1 1 0 1 0 0 0 0Unemployed 14 99 94 31 20 70 44 19 17 69 82 82 94 54

Retired 3 0 1 0 48 2 0 0 10 0 0 0 1 6

Unfit to work/disable 7 0 1 1 17 8 1 1 57 1 7 0 2 7

Housework 67 0 1 2 5 17 5 58 11 6 4 17 0 17

Other inactive 3 0 0 5 3 3 1 4 6 3 6 1 0 3Employed 7 15 8 81 8 1 64 23 0 37 9 7 28 22

Unemployed 12 83 86 15 15 68 29 16 16 53 71 74 68 45

Inactive 81 2 6 4 77 31 7 62 84 10 20 20 4 33

Length of unemployment spell† 12.7 11.8 12.3 10.8 12.6 12.9 10.3 12.7 12.9 11.2 12.0 12.3 10.6 11.9

Actively seeking a job at the time of the interview 6 80 82 12 3 56 24 9 6 47 65 75 61 61

Primary 42 19 23 7 32 28 24 16 36 6 17 21 14 24

Lower secondary 35 46 39 24 27 38 42 34 38 27 44 41 39 37

Upper secondary 14 18 19 23 16 19 24 25 13 23 15 26 22 19

Tertiary 8 17 18 46 25 14 9 26 13 43 23 12 25 21

Years of education 8.6 10.2 10.2 13.0 10.4 9.7 9.6 11.1 8.9 12.8 10.7 10.1 11.1 10.4

Age

groups*

Main

activity

during the

reference

period

Activity at

the time of

the

Level of

education

(ISCED)


Table A.2. Characterisation of the latent groups (cont.)


Note: Colour shadings identify categories with high (darker) frequencies. The average number of simultaneous barriers per individual is computed for the core indicators in Table A.1 with the exception of recent work experience. Income quintiles refer to the entire population. Poverty risks and material deprivation are calculated with the Eurostat methodology. “Length of unemployment spell” only covers reference period: unemployment spells that started before the start of the reference period are left-censored at the start of the reference period.

* The variable enters as an additional indicator in the latent class model. See Annex B for details.

† Average across observations with strictly positive values.

Source: Authors’ calculations based on EU-SILC 2014.


Pop

Number of individuals (%) 14 13 11 9 9 7 6 6 5 5 5 4 4 100

No work-related skills 34 0 0 2 0 4 7 19 15 0 71 17 0 12

Elementar occupations 28 34 24 11 10 25 61 23 20 12 6 28 21 24

Craft and machine operators 13 28 42 6 36 19 15 6 28 16 4 19 35 21

Clerk and sales 19 28 21 55 31 41 12 41 26 45 14 29 31 30

Technicians et al. 3 6 6 11 10 7 2 5 5 13 3 5 7 6

Professionals 2 4 4 11 10 3 3 4 4 12 2 2 4 5

Managers 1 1 2 3 4 1 1 1 1 2 0 0 1 1

Years of paid work experience†

14 9 25 12 37 12 17 10 22 17 3 10 19 17

Severe health limitations 7 1 2 1 9 8 1 2 37 1 8 1 1 6

Migrant 7 22 17 17 6 13 29 22 8 14 16 38 10 16

Equivalent disposable income (€/year - average) 15022 8709 8630 11483 18477 8927 9130 10673 13399 12844 9957 9114 15792 11753

Bottom quintile 18 52 50 34 15 50 44 33 24 25 45 45 20 35

Second quintile 21 23 26 21 15 26 30 30 23 23 23 27 23 23

Third quintile 23 13 13 19 19 13 14 20 22 24 15 16 17 17

Fourth quintile 23 7 6 15 21 7 9 11 18 16 9 9 18 13

Top quintile 16 5 4 11 30 4 4 6 12 12 8 2 21 11

AROPE (eurostat methodology) 20 55 55 37 17 54 49 35 27 28 48 47 24 38

No material deptivation 83 57 58 81 85 62 65 75 69 78 62 63 77 71

Deprived 10 22 23 11 9 20 20 13 18 14 22 21 15 16

Severe 7 21 19 8 6 18 15 12 14 8 16 16 8 13

Sickness and disability recipients (%), 16 2 3 6 31 12 3 6 69 5 8 3 12 13

they receive, in average†

8379 .. .. 5519 15594 7231 .. .. 14022 .. .. .. .. 11153

Unemployment benefits recipients (%), 13 45 69 34 26 35 43 16 12 59 16 41 81 36


4125 3604 5300 4092 8678 3896 3222 2935 4407 6526 2996 3827 11306 5223

Social Assistance recipients (%), 5 10 8 3 4 7 4 4 9 5 7 4 3 6


4100 3918 4272 .. .. .. .. .. 4144 .. .. .. .. 4097

Housing Benefits recipients (%), 0 2 3 1 1 3 1 1 0 2 1 1 1 1


.. .. .. .. .. .. .. .. .. .. .. .. .. 1,562

Family-related benefits recipients (%), 2 3 3 2 1 3 4 8 3 11 4 6 3 4


.. .. .. .. .. .. .. .. .. 2327 .. .. .. 2724

Old-age Benefits recipients (%), 3 0 1 0 37 1 0 1 3 0 0 0 1 4


.. .. .. .. 21934 .. .. .. .. .. .. .. .. 19679

Single 6 7 15 11 16 13 9 0 12 0 1 0 15 9

Couple without children 35 21 27 28 40 31 23 0 32 2 13 0 29 24

Couple with children 12 25 23 17 7 12 28 79 17 78 13 77 22 27

2+ adults without children 35 24 21 29 31 29 19 0 24 2 40 0 24 24

2+ adults with children 11 21 13 14 5 13 17 18 14 11 31 16 9 15

Lone parents 1 1 1 1 1 2 5 2 2 6 2 7 1 2

Have children* 2 32 19 11 3 1 28 99 19 95 20 100 18 28

Number of children†

.. 1.5 1.4 1.4 .. .. 1.4 1.5 1.2 1.5 1.5 1.5 1.5 1.5

Age of the youngest child†

.. 5 6 5 .. .. 6 5 5 5 5 4 5 5

Live in rural area* 27 30 27 26 23 27 34 32 30 24 28 30 36 28

Northeast 11 8 7 9 12 10 6 11 13 7 9 7 8 8

Northwest 10 6 8 10 10 5 7 9 6 8 6 3 8 6

Madrid 12 9 12 14 14 9 10 13 9 16 10 9 15 11

Central 13 12 13 12 11 14 13 14 12 10 14 12 13 13

Est 25 26 30 26 29 26 27 22 22 28 25 29 25 27

South 24 32 26 24 20 27 34 26 30 26 32 30 28 29

Canary Islands 6 8 5 5 4 9 4 5 8 4 5 9 3 6

Household with other working household members 56 44 35 58 38 39 49 80 37 74 59 77 48 51

Number of simultaneous barriers 2.7 2.7 1.9 1.2 1.7 2.5 2.0 2.6 2.8 2.1 3.2 3.6 2.8 2.3

Regions

(NUTS1)

Position in the

income

distribution

Material

deprivation

(Eurostat)

Benefits -

Recipiens and

average

amounts

(€/year)

Household

type

Work-related

skills (ISCO)


ANNEX B

LATENT CLASS ANALYSIS AND MODEL SELECTION

The segmentation method used in this note is Latent Class Analysis (LCA). This method exploits the

interrelations of an array of indicators through a fully-specified (i.e. parametric) statistical model for

organising the target population into homogeneous groups. In the present framework, the indicators

represent employment barriers and the statistical algorithm therefore identifies population sub-groups

sharing similar barriers to employment, e.g. “low skills, low work experience and weak financial work

incentives” for Group 1; “low skills, low work experience and scarce job opportunities” for Group 2, etc.

LCA has three main advantages relative to other common segmentation (or “clustering”) methods:

1) Formal statistical tests guide the selection of the optimal number of groups and other model’s features;

2) LCA does not allocate individuals into specific groups in a deterministic way but, instead, provides

probabilities of group membership, thus reducing possible classification errors in any post-estimation

analysis; 3) LCA deals easily with common data-related issues such as missing data and complex survey

designs.

Latent Class Analysis does not automatically provide an estimate of the optimal number of latent

classes. Instead, models with different number of classes are estimated sequentially and the optimal model

is chosen based on a series of statistical criteria. To summarise, the model selection process starts with the

definition of a standard latent-class model that is repeatedly estimated for an increasing number of latent

classes (Step 1).14

The choice of the optimal number of classes is primarily based on goodness-of-fit and

error-classification statistics (Step 2, see also Figure B.1), and then on the analysis of potential

misspecification issues (Step 3). Fernandez et al. (2016) describes these steps in details and provides

guidelines for practitioners interested in adapting the approach to specific analytical needs or data.

Figure B.1 summarises graphically Step 2 outlined above for the Spanish SILC 2014; The blue bars

show the percentage variations of the Bayesian Information Criterion (BIC, Schwartz 1978)15

for

increasing numbers of latent groups, whereas the black line shows, for the same groups, the classification

error statistics (Vermunt and Magdison, 2016).16

In general, a smaller value of the BIC indicates a more

optimal balance between model fit and parsimony, whereas a smaller value of the classification error

statistics means that individuals are well-classified into one (and only one) group. In Figure B.1 the BIC is

minimised for a model with 16 classes and the classification error of 22% indicates that the model provides

a good representation of the heterogeneity in the underlying data.

14. A standard latent class model means that the likelihood function is derived under the so-called Local

Independence Assumption (LIA). See Fernandez et al. (2016) for details.

15. The BIC summarises into a single index the trade-off between the model’s ability to fit the data and the

model’s parametrization: a model with a higher number of latent classes always provide a better fitting of

the underlying data but at the cost of complicating the model’s structure.

16. The classification error shows how-well the model is able to classify individuals into specific groups. To

understand the meaning of the classification error index it is important to keep in mind that LCA does not

assign individuals to specific classes but, instead, estimates probabilities of class membership. One has

therefore two options to analyses the results: allocate individuals into a given cluster based on the highest

probability of class-membership (modal assignment) or weighting each person with the related class-

membership probability in the analysis of each class (proportional assignment). The classification error

statistics is based on the share of individuals that are miss-classified according to the modal assignment.


Figure B.1. Selection of the optimal number of latent classes

Post-estimation tests based on the Bivariate Residuals (Vermunt and Magdison, 2005) show for the

16-class model some residual within-group correlation between 11 pairs of indicators. This indicates that

the model violates to some extent the Local Independence Assumption (LIA).17

Increasing the number of

latent classes always reduces the residual dependencies between indicators. For instance, the 17-class

model has only four local dependencies, but this comes at the cost of a higher classification error (25%).

Following Fernandez et al. (2016) and Vermunt and Magdison (2005) the residual dependencies

between indicators is addressed with the so-called direct effects; these are ad-hoc terms that enter the

specification of the likelihood function to model explicitly the joint probabilities of pairs of indicators

conditional on group membership. The inclusion of direct effects eliminates any residual correlation

between the relevant pair of indicators but it also requires repeating the model selection process, as the new

baseline model with local dependencies may lead to a different optimal number of classes. For the new

baseline model with direct effects the BIC now points to the 13-class model, which therefore becomes the

favoured solution.18

17. The LIA shapes the algebraic specification of the model and, in practice, requires the indicators to be

pairwise independent within latent groups. Bivariate residuals are Pearson chi-squared tests comparing the

observed associations between pairs of indicators with the expected association under the assumption of

local independence; large differences between estimated and observed associations signal violations of the

LIA.

18. Age, gender and regional differences define labour market segments that are worth including in the latent

class model to account for differences between and within these groups. Fernandez et al. (2016) discusses

three possibilities for including additional variables in the model’s specification. In SILC-2014 for Spain

the favoured specification in terms of lower classification error, interpretation of the results and

specification tests includes age, gender and a dummy for whether an individual has children as active

covariates. Figure B.1 is based on a model that already includes information on age (three categories:

18-29, 30-54, 55-64), gender and degree of urbanisation (three categories).

0.00

0.05

0.10

0.15

0.20

0.25

0.30

-2.5%

-2.0%

-1.5%

-1.0%

-0.5%

0.0%

0.5%

4 5 6 7 8 9 10 11 12 13 14 15 16 17

Number of latent groups

% Var. BIC Class. Err (RX)

Documents

James Browne and Rodrigo Fernández - OECD · FACES OF JOBLESSNESS IN SPAIN: ANATOMY OF EMPLOYMENT BARRIERS James Browne and Rodrigo Fernández