Universiteit van Amsterdam
Research Master Social Sciences
A Vocational Decline? The Influence of School-to-work
Linkage on Employment over the Life-course
Research Master’s Thesis
Andrea Forster
Amsterdam, 31 July 2015
Student number: 10635114
Contact: [email protected]
Supervisor: dr. Thijs Bol
Second reader: prof. dr. Herman G. van de Werfhorst
Abstract: Vocational education is seen as beneficial for the labour market allocation of young people.
However, recent literature points at disadvantages later in the life-course. This paper re-evaluates the
findings for the Netherlands with an improved measure of vocationality. For this purpose, a gradual
measure for the linkage strength between education and occupation is introduced and subsequently used
to predict labour market outcomes over the life-course. We can show that for the Dutch labour market
linkage strength is a better predictor for employment probabilities than previous measures of vocational
education. In the life-course analysis we find that the benefits of vocationality disappear later in the career.
However, while employment probabilities converge, vocational education never turns into a penalty.
Key words: Vocational Education, Labour Market, Segregation Indices, Netherlands
Contents
1 Introduction
1
2 Vocational Education and Labour Market Outcomes
2.1 School-to-work Transition . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Life-course . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
3
5
3 The Measurement of Vocationality
3.1 The Traditional Dichotomy . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Beyond the Dichotomy: Linkage Strength . . . . . . . . . . . . . . . . . . .
6
6
8
4 Data and Methods
4.1 Data and Sub-samples . . . . . . . .
4.2 Determination of Linkage Strength .
4.3 Operationalization of Other Variables
4.4 Logistic Regression Models . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
9
10
14
16
5 Results
18
5.1 Descriptive Results: Heterogeneity of Linkage Strength . . . . . . . . . . . 18
5.2 School-to-work Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3 Life-course . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6 Discussion
29
References
32
Appendices
A
The Educational System in the Netherlands . . . . . . . . . . . . . . .
B
Analysis of Sparse-cell Bias . . . . . . . . . . . . . . . . . . . . . . . .
C
Segregation Analysis: List of Educational and Occupational Categories
D
KHB Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
36
37
40
47
.
.
.
.
.
.
.
.
1
Introduction
The preparation of youth for the labour market is a key responsibility of the educational
system and education plays a major role in the distribution of life chances through this allocative function. A central feature of the educational system for labour market allocation
of graduates is its orientation towards vocational education. This form of occupationally
oriented training has been praised by policy makers as an efficient way of lowering youth
unemployment (OECD and ILO, 2014; Biavaschi et al., 2013).
Most research has confirmed these benefits of vocational education for the transition
from school to the first job (Shavit and Müller, 1998; Müller and Gangl, 2003; Breen,
2005). The main mechanism behind the effect of vocational education is that those
programmes convey occupation specific content to their students that is immediately
valuable for employers (Arum and Shavit, 1995).
However, the effect of occupation specific skills might vary over the life-course. This
makes it important to also look beyond the immediate transition from school to the first
job. Indeed, a recent study by Hanushek et al. (2011) finds that initial employment
benefits turn into disadvantages if one considers the whole labour market career up to the
retirement age. The explanation behind this phenomenon, which they term life-course
vocational decline, is that specific occupational skills become obsolete faster than general
skills later in the career if they are not constantly updated by on-the-job training.
Both, the argument about the school-to-work transition and the hypothesis about
life-course vocational decline identify the occupational specificity of an educational programme as the mechanism through which labour market outcomes are determined. However, none of these lines of research measure this specificity directly. Instead, it is simply
assumed that vocationality is a dichotomy: vocational education leads to high and general education to low specificity. It is much more likely, however, that the vocationality
of educational programmes is gradual. Programmes which are subsumed under the same
category (vocational or general) might vary in their degree of specificity. Some vocational
programmes, like car mechanic training, might indeed convey very specific occupational
skills, whereas for instance commerce oriented programmes might in fact rather teach
general skills even though they are classified as vocational programmes. The same seems
to be true for general education if for example the study of medicine is compared to
social sciences – two programmes labelled as general but with a very different degree of
occupational specificity.
If the occupational specificity is indeed the mechanism through which education gains a
labour market value or leads to life-course decline, not just the categorization as vocational
or general should matter for this outcome but also the variation of specificity within
these broad categories. Therefore, it remains unclear whether the dichotomous measure
1
of vocational education used in previous studies is suited to test arguments about the
consequences of vocationality.
We will contribute to the literature in two ways. First, the life-course hypothesis of
vocational decline will be re-evaluated. So far, the decline has been investigated with
comparative survey data that contain only small samples for individual countries. Using
much more detailed data from the Dutch Labour Force Survey, we will test the hypothesis for a country with high enrolment in very specific vocational programmes in which
strong life-course effects are expected. Thereby, employment probabilities as the most
fundamental labour market outcome will be in focus.
Second, we will perform this test of vocational decline with an improved measure for vocationality. Instead of dichotomizing educational programmes in vocational and general,
this measure captures the vocationality of educational programmes directly. Following
a recent approach by DiPrete et al. (2015), the strength of linkage between educational
programmes and occupational positions will be utilized to measure the occupational specificity of single educational programmes. Thereby, not only levels of education but also
fields of study can be taken into account.
From these two contributions, the subsequent research question follows: Do graduates
from educational programmes that link more strongly to certain occupational positions
experience an initial advantage and subsequent disadvantage in employment probabilities
compared to graduates from programmes with lower linkage?
Answers to this question are relevant to public debates about the value of vocational
content of education. The generally positive evaluation of vocational education in the
school-to-work transition leads to calls for more vocational elements in education and a
tighter coupling of school and work place whenever youth unemployment is on the rise.
However, such policies are only advisable if a highly occupation specific education does
not revert into a disadvantage over the life-course.
To investigate the research question, first, we will introduce the mentioned measure of
linkage strength. With this measure, we will be able to display the heterogeneity in occupational specificity within programmes categorized as vocational and general. Second, an
analysis of the school-to-work transition will show that the new vocationality measure indeed explains labour market allocation better than the traditional dichotomous measure.
Third, we will show that a high linkage strength leads to an initial advantage and subsequent decline in advantage of employment probabilities over the life-course. However,
late disadvantages do not have the potential of outweighing early employment benefits.
Therefore, in the Netherlands, the life-course pattern can be seen as a convergence of
employment probabilities rather than a decline.
2
2
Vocational Education and Labour Market Outcomes
In the following section, theoretical arguments about labour market outcomes of vocational education are presented for (1) the school-to-work transition and (2) the further
labour market career.
2.1
School-to-work Transition
As already mentioned, vocational education is commonly seen as advantageous for the
labour market allocation of graduates (Shavit and Müller, 1998; Müller and Gangl, 2003;
Breen, 2005). This advantage seems to be due to the occupational specificity of this type
of educational programmes.
There are several mechanisms through which educational programmes become occupation specific and through which the benefits of vocational education may operate.
The explanation which is most often used by research relates the mechanism through
which jobs are obtained directly to the type of skills taught in the educational programme.
Vocational degrees are beneficial for obtaining employment because of the specific skills
they entail (Arum and Shavit, 1995; Scherer, 2005; Wolbers, 2007; Van de Werfhorst,
2011). They prepare students for a very narrow set of occupations. It is very likely that,
for example, someone who learns the narrow technical skills to repair cars, will most
often be working in an occupational category where these skills are of use. Vocational
graduates with such specific skills are immediately valuable for employers and do not
require extensive on-the-job training.
Another, related mechanism is that of signalling (Arum and Shavit, 1995). Vocational
degrees signal high immediate productivity to possible employers. Here the focus is not on
the skills which are actually obtained but rather on the productivity which an educational
title signals to employers.
These two mechanisms also imply that vocational education is only beneficial if it
conveys narrow occupation-specific rather than broad practical skills. Broad vocational
skills serve more as a safety net for low achievers and do not have the same benefits as
occupation-specific training. They are above all a signal of low ability and the skills are
not specific enough to replace on-the-job training (Shavit and Müller, 1998).
Next to skill specificity and signalling, vocational training also comes along with
stronger institutional and/or personal networks between graduates and employers which
facilitate the transition from school to work (Rosenbaum et al., 1990). Often, employers
are directly involved in the vocational education system. Then, not the skills obtained
in those programmes are important for hiring but the information that employers gain
about graduates already during the training period.
Yet another reason why the occupational specificity of educational programmes might
be high and the linkage to the labour market close, has to do with credentialism (Bills,
3
1988) and occupational closure (Weeden, 2002; Bol and Weeden, 2014). One form of
closure that directly influences linkage strength between education and occupation is
licensure. Licenses restrict the access to a occupational position by requiring a formal
certificate that allows its practice (Bol, 2014). If a specific educational degree is required
to access a certain position on the labour market, linkage of this educational programme
will necessarily be high independent of the skills taught in it.
Comparative research has shown that in countries with a high proportion of vocational
education, labour market entry is indeed smoother (Shavit and Müller, 1998; Van der
Velden and Wolbers, 2003; Breen, 2005; Wolbers, 2007). However, some scholars doubt
these transition benefits for vocationally trained graduates. While some studies find
that vocational graduates indeed experience advantages compared to their peers with
general training (Shavit and Müller, 1998; Scherer, 2005), other studies conclude that the
transition benefits also extend to other graduates in countries where vocational education
is widespread (Iannelli and Raffe, 2007; Wolbers, 2007). If following the latter results, one
might conclude that a differentiation of the educational system in different school tracks
and not an individual’s enrolment in a vocational programme is responsible for those
beneficial outcomes: the smooth labour market entry is a general benefit for everyone in
such stratified educational systems and is not related to the occupational specificity of
the educational programme (Andersen and Van de Werfhorst, 2010; Levels et al., 2014).
So far, these results do not yet allow a definite conclusion about whether vocational
education is beneficial for the individual in the transition from school to work, and through
which mechanism these benefits on the individual level operate if they exist. We reevaluate the school-to-work transition but replace the dichotomous measure for vocational
education that is used by previous studies, with a direct measure of occupational specificity
for single educational programmes. With this more precise measure of vocationality, we
can show that it is indeed the occupational specificity of educational programmes that
accounts for these transition benefits.
One country with a strong vocational system in which beneficial individual level effects
are expected, is the Netherlands (Shavit and Müller, 1998). Indeed, De Graaf and Ultee
(1998) find lower unemployment rates for vocational graduates compared to generally
trained workers in the Netherlands. Following the argument about the individual benefits
of vocational education for labour market allocation, we assume that, in the Netherlands, occupational specificity has a positive influence on employment probabilities in the
school-to-work transition. This specificity is measured by the linkage strength between
educational programmes and occupations, which leads us to the following hypotheses:
H1.1: The linkage strength of an education to a specific occupation has a positive
influence on employment probabilities early in the career.
4
H1.2: The linkage strength is a better predictor of employment probabilities than a
dichotomous measure of vocational education. If it is controlled for linkage strength,
the dichotomous measure does not significantly add to the prediction of employment.
2.2
Life-course
In addition to the question whether vocational education is beneficial in the transition
from school to the first job, it is relevant to study how workers fare in the rest of their
career. Even if vocational programmes are beneficial for youth in the school-to-work
transition, those advantages might decline over the life course (Hanushek et al., 2011).
However, vocational education can only be evaluated positively if the initial benefits in
labour market allocation do not turn into disadvantages for subsequent employment.
The expected mechanism behind this decline is the development and utilization of
specific skills in the long run. In the course of technological innovations the need for skills
in the labour market changes rapidly, and technical skills quickly become obsolete (Katz
and Murphy, 1992). On-the-job training is required to keep the narrow occupational skills
of vocational workers updated over the course of their labour market career. However,
employers are used to obtain fully trained workers from the educational system and they
are not prepared to invest much in further training. This is especially true in countries
where vocational education is widespread (Hanushek et al., 2011). This lack of training
leads to a deterioration of vocational skills and to a decline of employment probabilities
with increasing age particularly in those countries.
In general, Hanushek et al. (2011) find support for their hypothesis. Initial benefits
of vocational education are reversed at a later point in the labour market career. The
authors report that at an age of 16, vocational graduates are 7 percentage points more
likely to be employed than individuals with general education but that this gap narrows
subsequently and reverses into a disadvantage at the age of 50. Similarly, results for
Austria by Vogtenhuber (2014) indicate that subsequent transitions in the labour market
after the first entrance are less positive for vocational graduates than for generally educated individuals. However, his is unable to capture this effect fully with its sample of
young adults.
We evaluate the decline hypothesis in more detail for the Netherlands. While Hanushek
et al. (2011) focus on a comparative perspective and, thereby, necessarily apply less detail
to the study of single countries, we use very detailed micro-level data which allow us to test
the claims more thoroughly for one country. The Netherlands is a country with a strong
vocational system and high enrolment in a vast number of specific training programmes
for which strong decline effects are expected. However, Hanushek et al. (2011) find no
decline for the Netherlands. This makes the Netherlands an interesting case for further
investigation with more detailed data. Again, we will use the linkage strength measure
5
as a more precise way of operationalizing occupational specificity. Following the decline
argument, the subsequent hypothesis is made:
H2: The positive effect of linkage strength on employment probability declines with
increasing age and turns into a penalty late in the career.
3
The Measurement of Vocationality
The hypotheses formulated in the previous section assume the occupational specificity to
be the mechanism through which education leads to beneficial labour market outcomes in
the school-to-work transition and to a subsequent decline over the labour market career.
To evaluate these assumptions, a measurement is necessary that captures the occupational specificity of education. In this section we will discuss how previous research
measured if an educational programme was vocational or not, what drawbacks arise from
this measurement and how the operationalization can be improved by using a gradual
measure of occupational specificity.
3.1
The Traditional Dichotomy
It is difficult to find a definition of vocationality that is universally valid as vocational
education takes various forms in different countries: vocational programmes vary from
firm-based training, to dual apprenticeships, to education in specialized schools and vocational curricula in regular high schools. However, in most cases, scholars identify vocational education programmes by the specificity of skills which they convey or by the
closeness of their association with certain labour market positions.
The differentiation between general and specific skills originates from classical work
on human capital (Becker, 1964): specific skills are those which are immediately valuable within one occupation or even only in one firm, general skills in contrast are broad
and applicable in a variety of contexts. An application of the concept of specific and
general skills to the educational system appears in the literature on different varieties
of capitalism. According to Estevez-Abe et al. (2001), two types of skills are conveyed
by the educational system: vocational education teaches mostly industry-specific skills
which lead to specific occupations but are not tied to certain firms. General, mostly
university-bound, tracks convey general skills. Similarly, Shavit and Müller (1998) distinguish general skills from specific vocational skills in their comparative work on the
school-to-work transition. They further differentiate the specificity of skills obtained in
programmes labelled as vocational in broad vocational skills and specific vocational skills.
Broad vocational skills are entirely obtained in schools that most often serve as a safety
net for low-achievers and which do not teach the skills for a specific occupation but rather
general practical skills that can be applied in different occupations. In contrast, specific
6
vocational skills are taught either in a dual system with firm- and school-based training
elements (e.g.in Germany) or in specialized schools (e.g.in the Netherlands) that offer
a wide array of different very specialized training programmes for detailed occupational
titles. Degrees from such programmes serve as a strong productivity signal on the labour
market and employers are often directly involved in this specialized training.
In their work, Shavit and Müller (1998) use the Comparative Analysis of Social Mobility in Industrial Nations (CASMIN) educational classification (Müller et al., 1989; Erikson
and Goldthorpe, 1992). This is one of the major educational classifications used in comparative research and it reflects the separation of general and specific skills which are
taught in academic and vocational programmes respectively. CASMIN differentiates educational degrees in hierarchical levels and also distinguishes the orientation of a programme
towards general or vocational skills within levels (Müller et al., 1989). The original version
of the CASMIN classification, which was developed in the 1970s, includes vocational degrees on elementary and intermediate educational levels. It acknowledges the difficulties
caused by the high diversity of educational programmes in different countries, especially
on the intermediate level, but nevertheless decides for a dichotomous classification of vocational and general programmes (König et al., 1988). A revision of the CASMIN scheme
at the end of the 1990s extends the classification of vocational programmes to the tertiary education level in reaction to the development of professional degrees in the higher
education sector (Brauns and Steinmann, 1997; Brauns et al., 2003).
The most common alternative to the CASMIN scheme, the International Standard
Classification of Education (ISCED), both in the 1997 and the 2011 version, likewise
categorizes education in levels and fields of education and, additionally, applies a dichotomous separation of vocational and general education on the secondary and tertiary levels
of education. It defines vocational education as all those programmes “that are designed
for learners to acquire the knowledge, skills and competencies specific to a particular
occupation, trade, or class of occupations or trades. Such programmes may have workbased components (e.g.apprenticeships, dual-system education programmes). Successful
completion of such programmes leads to labour market-relevant, vocational qualifications”
(UNESCO Institute for Statistics, 2012, p.14). In contrast, general education programmes
are defined “to develop learners’ general knowledge, skills and competencies, as well as
literacy and numeracy skills” (UNESCO Institute for Statistics, 2012, p.14).
Virtually all research in the field of education and comparative stratification uses
these two classifications for vocational and general education. In other words, all previous attempts of defining vocationality of educational programmes use a more or less
dichotomous classification that associates vocational education with specific and general
education with general skills.
7
3.2
Beyond the Dichotomy: Linkage Strength
All definitions outlined in the previous section are based on the presumably high occupational specificity in vocational education. However, none of the classifications measures
this specificity directly. It is assumed that the programmes that are classified as vocational
are more specific than programmes which are classified as general education. However,
it is likely that not all programmes which are classified as vocational entail the exact
same linkage to specific occupations. The same can be assumed for general educational
programmes. This makes the dichotomous measure unsuitable for capturing the varying
degree of vocationality of educational programmes.
These shortcomings of the dichotomous approach to the measurement of vocationality, show the necessity of a measure that more directly captures the gradual nature of
specificity entailed in detailed educational programmes.
DiPrete et al. (2015) introduce a gradual measure of vocationality which is suitable
for this endeavour. In their approach, vocationality is evaluated for single educational
programmes instead of only the two categories of specific and general education. Thereby,
not only educational levels are considered but also detailed educational fields of study.
This makes it possible to assess heterogeneity in the occupational specificity of different
fields within the same educational level.
In this linkage approach, vocationality is measured as the strength with which an educational programme is linked to certain occupational positions. Linkage of an educational
programme is high if a high number of graduates with this specific educational level and
field combination are found in a narrow set of occupational positions. It is low if graduates are spread out over a high number of different occupations. For example, if most
university students of medicine become physicians, linkage of the educational programme
medicine is very high. In contrast, if students of management at an university spread out
over a high number of different occupations, the linkage of this educational programme
and thereby its occupational specificity is considered low. However, in the traditional way
of operationalizing occupational specificity, students from both of these fields would be
classified as having general education as they attended university. This example shows,
that a gradual measurement of vocationality for single educational level-fields captures the
actual specificity of an educational degree more precisely as it addresses all the variation
in specificity between programmes.
In this paper, an approach to the determination of this linkage strength will be chosen
that is very similar to the one introduced by DiPrete et al. (2015). This measurement
uses segregation indices (Reardon and Firebaugh, 2002; Mora and Ruiz-Castillo, 2011;
Alonso-Villar and Del Rı́o, 2010; Frankel and Volij, 2011) which are based on the concept
of entropy. In the context of measuring the association between educational level-fields
and occupations, entropy reflects the amount of information that is gained about the
8
occupational position of an individual if their educational degree becomes known. As
one’s education is expected to contain some information on someone’s occupation, entropy
within an education should be lower than overall entropy. Knowing someone’s education
should therefore lead to a reduction of entropy. The measure of association between
education and occupation that will be used to determine the strength of linkage of an
educational programme in this paper is based on this reduction of entropy.
This linkage strength between an educational programme and an occupation then
serves as an indicator of the vocationality of an educational programme. With this gradual
linkage measure, the questions concerning education and labour market outcomes, which
were presented in Section 2, can be re-evaluated.
4
Data and Methods
4.1
Data and Sub-samples
The Dutch Labour Force Survey (Enquête Beroepsbevolking, EBB) which can be accessed
via Statistics Netherlands (Centraal Bureau voor de Statistiek, CBS) offers high quality
micro-level data that can be used for the determination of linkage between detailed educational programmes and occupations. Furthermore, it provides sufficient information
to investigate labour market outcomes for people of different age. The survey is administered as a rotating panel study of households: each month, a new representative sample
of households in the Netherlands is drawn. Per household up to eight persons from the
age of 15 can participate in the survey. Each respondent in the sample is then approached
for five consecutive interviews over a period of twelve months.
We use the EBB rounds from 2010 to 2012. To achieve a sample size capable of
precisely measuring linkage strength for a high number of educations and occupations,
observations for these three years are pooled for the analyses. We assume that general
labour market structures in these three years are comparable enough to do so and, additionally, include fixed effects for the survey year in each model. Only one observation per
person is selected for the analysis.1 Different restrictions are made to the sample for the
calculation of the linkage measure and for the analysis of employment probabilities.
First, for the determination of linkage strength all respondents are included who have
complete data on their highest educational degree and on their current labour market
position and who are not enrolled in education at the time of the interview.2 This implies
1
The first interview with complete data is selected for each respondent. Usually the first interview
took place in wave 1 of the survey. However, if in that interview data of the individual was missing on one
of the variables of interest, the first wave with complete data was used. Some individuals participated in
the survey in two of the years. In this case, the first complete wave of the first year was selected.
2
DiPrete et al. (2015) additionally carry out the determination of linkage strength with a smaller
sample of young respondents up to 35 years of age. This procedure captures more precisely the actual
school-to-first-job linkage than a calculation of linkage with the whole sample. However, they find no
9
that the determination of linkage strength is only carried out for individuals who are
currently in employment as otherwise, no occupational data is available for them. Further
research is necessary to investigate the implications of this selection. This is, however,
beyond the scope of the present paper. The final sample has a size of 193,445 respondents.
Second, labour market outcomes are investigated for all respondents no matter their
employment status. Again, respondents are excluded who are in education at the time
of the survey as the focus of interest lies on people who have left full-time education
and are either active in the labour market or out of employment for other reasons than
education. For the investigation of the school-to-work transition, the sample is restricted
to respondents in the age of 16 to 35, following similar restrictions in previous research
on the topic (e.g. Shavit and Müller, 1998; Wolbers, 2007; Vogtenhuber, 2014). The
final sample for this analysis has a size of 60,052 respondents with 29,982 being male
and 30,070 being female. For the life-course outcomes all respondents in a typical age of
labour market activity from 16 to 65 are selected. This again follows previous research
on life-course outcomes (Hanushek et al., 2011). This selection leads to a final sample of
216,004 respondents, whereof 107,047 are men and 108,957 are women.
4.2
Determination of Linkage Strength
Statistical Approach
As mentioned in Section 3.2, the measurement of linkage strength is based on segregation
indices. For this purpose, DiPrete et al. (2015) use one available segregation measure,
the Mutual Information Index (M).3 The M index is an aggregated measure for the vocationality of a national educational system but it is composed of the weighted sum4 of the
linkage of all single education programmes to occupations within this educational system.
This linkage measure for the single educational and occupational categories is called local
segregation. The local segregation (Mg ) for each educational programme can be expressed
by equation 1.
Mg =
X
pj|g ln
j
pj|g
pj
(1)
significant differences between the two measures for linkage strength. The larger sample is better capable
of including a high number of educational programmes as it is less subjected to sparse categories and is
therefore preferred.
3
Compared to previous attempts to measure occupational specificity with another segregation index
– the GINI index (Allen et al., 2000; Vogtenhuber, 2014) – the M index has the advantage of being
strongly decomposable. This means that it can be used to measure the specificity of single educational
programmes but it also can serve as an indicator of the specificity of national educational systems.
4
This means that if an educational programme is strongly linked to an occupation but its size is minor,
it contributes less to the overall linkage in a country than if the category is bigger. For example, a PhD
in medicine is highly specific but only a few individuals obtain this degree. Therefore, the contribution
to national vocationality is low.
10
Thereby, pj|g is the conditional probability of being in a certain occupation j given
one has the educational degree g. This value is multiplied by the logarithm of the ratio
between the conditional probability pj|g and the unconditional probability pj of being
in that specific occupation j across all educations. The result is then aggregated over
all occupations to obtain the local segregation for one specific educational degree. In
substantive terms, local segregation shows how much workers with a specific education
are spread across occupations compared to all workers (DiPrete et al., 2015). A detailed
discussion of the technical properties of the local segregation measure and the Mutual
Information Index can be found in DiPrete et al. (2015) and Mora and Ruiz-Castillo
(2011).
As linkage strength is fundamentally about an association between education and
occupation two variables are necessary for its calculation: detailed educational and occupational categories.
Educational Categories
In the EBB, educational degrees are measured by the Dutch educational classification
(Standaard onderwijsindeling, SOI) which is oriented towards educational levels and fields
in the Dutch educational system.5 Each educational programme is identified by a six-digit
code in which the first two digits represent the level of education and the four remaining
digits indicate the field of study. The two digits for the level of education are further
separable into the main level (first digit) and a sub-level within the main level (second
digit). The four digits for the field of education work in a similar way. We use only the
first two digits for the field in order to avoid empty categories. By truncating the SOI
code, a four-digit code for educational levels-and-fields is obtained: the first two-digits
indicate the level and sub-level and are followed by another two digits which represent
the major field plus one level of sub-fields.
An issue when determining local segregation is the sample size in the single education
cells. Sparse cells potentially inflate the local segregation value as single outliers have a
high impact on the measure if categories are small. For instance, if there are only ten
individuals in an educational category, two individuals which randomly end up in the
same occupation already constitute a high proportion in such an educational category
compared to a cell with several hundreds of observations. Therefore, the potential for
segregation is much higher when there are only few individuals in a category. This issue
decreases the reliability of the measure. DiPrete et al. (2015) conclude from an analysis
of the aggregated M measure with differently sized samples that sample sizes of 100,000
or more are unproblematic for an analysis with a similar detail level of education as we
use in this paper. As our sample meets this criterion, the problem is expected to be less
5
A short overview of the Dutch educational system is presented in Appendix A.
11
severe. However, DiPrete et al. (2015) only investigate the size of the total sample and
not of the single educational categories. Even if the overall sample size is high, individual
cells still can have a low number of observations. As this paper uses local segregation,
the problem is more influential here than in the context of DiPrete et al. (2015) where
the influence of single cells is weighted by their size when aggregating for the whole
educational system. Therefore, an additional analysis with different cell sizes for the
single educational categories is carried out to test the robustness of the linkage measure.
This sparse-cell analysis is described in Appendix B in more detail. Although further
investigation into the reliability of the linkage measure seems advisable, a minimum cell
size of 120 observations can be chosen as appropriate for the analyses in this paper. By
doing so, we use a slightly more conservative threshold than DiPrete et al. (2015) who
used 100 observations as a minimum.
Using this minimum cell size, the educational categories are prepared for the segregation analysis. Originally, the data contains 193,445 observations distributed over 351
level-fields of education. Educational level-field combinations with a cell size below 120
are identified. These cells are aggregated to a less detailed level by re-coding the fourth
digit (the second field digit) into zero and thus setting the field to Other, only leaving
information about the broader field. This re-coding results in 245 remaining level-fields.
The procedure is repeated for categories which sill remain below the threshold so that the
first field digit is also removed. This re-coding leads to a loss of detail but allows us to
leave those observations in the sample without subjecting them to sparse-cell bias. Nevertheless, the interpretation of the Other categories is less intuitive than for the remaining
level-fields as they are necessarily broader in their content. After this procedure, 194
level-fields of education are left. These are combinations of the levels and fields displayed
Table 1: Educational Levels
SOI
Level
Examples
10
Pre-primary education
Basisschool group 1 and 2
20
Primary education (group 3 and higher)
Basisschool group 3 to 8
31
Lower secondary education, low level
Vocational courses without pre-requisites
32
Lower secondary education, intermediate level
e.g. VMBO basic track
33
Lower secondary education, high level
e.g. VMBO theoretical track, MAVO, HAVO/VWO years 1-3
41
Upper secondary education, low level
e.g. MBO-2
42
Upper secondary education, intermediate level
e.g. MBO-3, HAVO years 4-5
43
Upper secondary education, high level
e.g. MBO-4, VWO years 4-6, pre-university courses
51
Higher education, first phase, low level
e.g. short HBO or Associate Degrees
52
Higher education, first phase, intermediate level
e.g. HBO Bachelor
53
Higher education, first phase, high level
WO Bachelor
60
Higher education, second phase
Master’s degrees HBO and WO
70
Higher education, third phase
Ph.D., Doctorate
Source: EBB 2010–2012
12
Table 2: Educational Fields
01
General education
47
Law, public administration, public order and security
10
Teachers
51
Mathematics, natural sciences
11
Teachers general education
52
Computer science
12
Teachers humanities, social sciences, communication
and arts
60
Engineering
13
Teachers mathematics, natural sciences, agriculture
61
Engineering general
14
Teachers technical subjects and transport
62
Electrical engineering
15
Teachers economics, commercial, management and administration
63
Construction
16
Teachers (health) care, sports and other
64
Metal processing, vehicle and tool manufacturing
17
Teachers with differentiation
65
Process technology
20
Humanities, social sciences, communication and arts
66
Textile and leather processing, other
21
Languages
67
Engineering with differentiation
22
Humanities other
70
Agriculture and environment
23
Social sciences
71
Agriculture
24
Communication, media, information
72
Environment
25
Arts, expression
77
Agriculture and environment with differentiation
27
Humanities, social sciences, communication and arts
with differentiation
80
(Health) care and community services
30
Economics, commercial, management and administration
81
Health care
31
Economics
82
Care, community services
32
Commercial
87
Health care and community services with differentiation
33
Management
90
Hotels, gastronomy, tourism, leisure, transport and logistics
34
Human resource management, personnel
91
Hotels, gastronomy, tourism and leisure
35
Administration, secretarial
92
Transport and logistics
37
Economics, commercial, management and administration with differentiation
97
Hotels, gastronomy, tourism, leisure, transport and logistics with differentiation
41
Law, public administration
98
Other
42
Public order, security
Source: EBB 2010–2012
in Tables 1 and 2. The final list of all 194 combinations between levels and fields used for
the calculation of linkage strength can be found in Appendix C.
Occupational Categories
Occupational information is available in the EBB via the International Standard Classification of Occupations (ISCO, 2008). ISCO codes are four-digit numbers which represent
four levels of detail for occupational categories. The first digit of the ISCO code displays
the major occupational group, the subsequent digits symbolize sub-fields within these major groups. We use the first three digits of the ISCO codes. This decision addresses the
trade-off between including detailed information and having enough cases per occupation
available for the analysis. Following previous research, occupations within the military
(major ISCO group 0) are excluded from all analyses as those categories are hard to compare with civil occupations (Weeden, 2002; Bol and Weeden, 2014). This selection results
13
in 128 occupational categories for the segregation analysis. The list of all occupations
used for the analysis is presented in Appendix C.
Description of the Linkage Measure
Linkage strength is calculated for each educational category (SOI four-digit code) using
equation 1. Weights are applied to the analysis. The sampling weights in the EBB are
rescaled in a way that they amount on average to a value of 1 per observation in the final
sample and thus sum up to the total number of observations.
This analysis results in an index for linkage strength which ranges from 0.178 to 3.494
with an overall mean of 1.280 (SD=0.533). The distribution of the linkage measure for the
different level-fields of education is displayed in Figure 1. The measure is nearly normal
distributed with a few outliers with very strong linkage.6
Figure 1: Distribution of the Linkage Measure
4.3
Operationalization of Other Variables
For the investigation of labour market outcomes, a number of other variables have to
be defined. Summary statistics of all variables for the more extensive sample of 16- to
65-year-old respondents are displayed in Table 3.
The dependent variable is a dichotomous measure of the employment status of individuals. Similar to Hanushek et al. (2011), we do not distinguish between different reasons
for not working. Respondents who are coded as 0 can be unemployed in a narrow sense
but it is also possible that they have left the labour force for reasons like retirement,
inability to work, etc.7
6
These outliers will be discussed in more detail in Section 5.1.
Further studies could address a possible difference between being unemployed and being out of the
labour force in the context of linkage strength. This is, however, beyond the scope of this paper.
7
14
The main independent variable is the linkage strength of each level-field combination
that was obtained in the analyses described in Section 4.2. This linkage value for each
educational programme is matched to the individuals for the analysis of labour market
outcomes using their highest educational degree as identifier.
Additionally, a dichotomous measure of vocational education is obtained to assess
whether linkage strength predicts employment better. This variable is coded using the
International Standard Classification of Education (ISCED, 1997) variable, that is available in the EBB next to the SOI educational classification. Programmes on the secondary
Table 3: Descriptive Statistics for All Respondents Aged 16 to 65 (Full Regression Sample)
Variable
Employment status
Values/Range
0–1
Men
Women
Obs.
Mean
SD
Obs.
Mean
SD
107,047
0.807
0.394
108,957
0.680
0.467
employed
1
86,438
74,052
unemployed
0
20,609
34,905
Linkage strength
0.178 – 3.494
107,047
1.038
0.537
108,957
0.974
0.550
0–1
107,047
0.496
0.500
108,957
0.480
0.500
1
53,098
Vocational education
vocational
general
Age
Level of education
52,270
0
53,949
16 – 65
107,047
43.663
12.775
108,957
56,687
43.655
12.772
1–7
107,047
4.294
1.352
108,957
4.207
1.341
pre-primary
1
1,714
primary
2
6,551
7,473
lower secondary
3
21,721
23,625
upper secondary
4
40,044
41,366
post-secondary
5
3,375
3,259
tertiary
6
32,894
31,062
doctorate
7
749
370
107,047
108,957
Region
1 – 12
1,801
Groningen
1
3,597
3,651
Friesland
2
4,193
4,137
Drenthe
3
3,138
3,158
Overijssel
4
7,149
7,217
Flevoland
5
2,532
2,607
Gelderland
6
12,738
13,003
Utrecht
7
7,686
8,140
Noord-Holland
8
17,478
17,957
Zuid-Holland
9
22,667
23,081
Zeeland
10
2,539
2,586
Noord-Brabant
11
16,072
16,107
Limburg
7,258
7,312
2010 – 2012
107,047
108,957
2010
2010
49,572
50,163
2011
2011
32,552
33,327
2012
2012
24,923
25,467
Year
12
Source: EBB 2010–2012, own calculations
Note: Sampling weights are applied
15
school level which are labelled as vocational or pre-vocational are coded as vocational (1)
contrasted with general education (0). On the tertiary level, professional degrees are
coded as vocational (1), all other degrees are coded as general (0).
Age is one further important variable for the determination of employment across the
labour market career. The age range is defined by the sample restrictions outlined in
Section 4.1 for the different analyses. Following earlier research in the assessment of age
and employment (e.g. Hanushek et al., 2011), a squared term of age is included to account
for a possible non-linear relationship between age and employment probability.
Additionally, as we are interested in the effect of occupational specificity net of the
labour market consequences of different levels of education, we control for level of education. This allows us to look at the effect of linkage strength within educational levels.
Employment probabilities differ considerably between different educational levels which
makes it important to single out this level effect. Again the ISCED classification is used
to obtain an ordinal variable of educational levels with 7 steps.
Next to the educational level, it is also controlled for regional effects as the overall
employment situation differs between provinces. Dummy variables for the 12 provinces
in the Netherlands are included. Finally, fixed effects for the survey year are added to
single out differences in employment in the three years of the EBB that were pooled for
the analyses. A significantly higher proportion of the sample comes from the survey year
2010. This is due to the selection of the first interview with complete data. If respondents
participated in several years, their data from 2010 was used.
4.4
Logistic Regression Models
As the dependent variable is dichotomous, a binary logistic regression design is appropriate
for the analyses. Analyses with different sub-samples are carried out separately for the
school-to-work transition and for life-course outcomes. The specific models for those two
sets of analyses are presented in the following two paragraphs:
The first set of analyses for the school-to-work transition is done with the reduced
age sample (16 to 35 years of age). The first model only contains the influence of linkage
strength on employment. The second model replaces linkage strength with the traditional
dichotomous measure of vocational education. In the third model, both measures – the
dichotomous and the gradual – are added to determine whether linkage strength explains
the effect of vocational education and if the latter predictor loses its significance. In all
three models, it is controlled for age, age squared, educational level, region and survey
year. Equation 2 shows the regression equation for the full model in which pi is the
probability of being employed for individual i.
ln
pi
= β0 + β1 linkagei + β2 voci + controls
1 − pi
16
(2)
In a second set of models, the hypothesis of a vocational decline is investigated with
the sample of respondents in the age from 16 to 65. The first model includes the linkage
strength, age and age squared as main predictors and tests whether linkage strength has
also a positive influence on employment in this full sample. The second model adds an
interaction effect between age and linkage strength which is the main interest of the decline
analysis and allows us to assess whether the effect of linkage on employment differs at
different ages across the labour market career. This makes it possible to test the life-course
hypothesis with the cross-sectional data sample. A negative interaction effect symbolizes
a decline of employment probability for high linkage strength with increasing age and,
thereby, would be an indicator for vocational decline. Again, it is controlled for level of
education, region and survey year in all models. The logistic regression equation for the
probability of being employed (pi ) for the full model is displayed in Equation 3.
ln
pi
= β0 + β1 linkagei + β2 agei + β3 agei × linkagei + controls
1 − pi
(3)
The sampling weights provided by the EBB are used for all models. All analyses
are done separately for men and women. The employment patterns and labour market
careers of men and women differ considerably due to structural gender inequalities and still
existing traditional gender roles. Therefore, it is expected that their life-course outcomes
on the labour market differ in a way that is hard to control for in a joint analysis.
Additionally, separate analyses may give interesting insights in differences between men
and women when it comes to the influence of linkage strength on employment.8
Following recent discussions about the re-scaling of coefficients in logistic regression
which influences the comparability of effect sizes across nested models, all models are
additionally evaluated using the KHB method (Karlson and Holm, 2011; Karlson et al.,
2012). The goal of the KHB method is to obtain differences in the effect of a variable
between models without and with control variables that are unbiased by re-scaling effects
(Breen et al., 2013). The results of the KHB analyses are presented in Appendix D.
No substantively different conclusions arise from using the KHB method instead of regular logistic regression. Therefore, the main analysis is carried out with regular logistic
regression.
8
The regression analyses are run with the full sample. Additionally, a set of regressions is carried out
that excludes the observations of which education was re-coded to Other. No differences arise from these
analyses. Therefore, all observations are used, also those who only have information on a broader field
of study.
17
5
Results
5.1
Descriptive Results: Heterogeneity of Linkage Strength
First, the obtained linkage strength for the different educational programmes is examined.
Figure 2 shows the dichotomous vocational education variable on the y-axis and the
strength of linkage on the x-axis.
The dark grey circles symbolize the single level-fields of education. The light grey
diamonds show the average linkage strength within the vocational and the general category. It can be seen that on average, programmes which are classified as general link even
slightly stronger with a mean of 1.35 than vocational programmes with a mean of 1.29.
However, a t-test shows that this difference is not significant (t=0.6837(164), p = 0.248,
one-sided test). Interestingly, these descriptive results show that – on average – we do not
find more strongly linking educational level-fields in the category that previous research
defined as vocational.
Figure 2: Distribution of Linkage Strength of Vocational and General Programmes9
9
Not all level-fields of education could be identified as clearly vocational or general. This can happen
due to inconsistent response behaviour of individuals. Another reason could be that we do not use educational fields in full detail. Sub-fields of the two-digit fields could be vocational and general, making the
broader category unidentifiable. To produce the figure, the vocationality of the level-fields is harmonized.
Level-fields in which the mean of the individual values on the vocational variable is above 0.8 are classified as vocational (1). Those who have a mean value below 0.2 are classified as general (0). All cases
with average between 0.2 and 0.8 are not included in the figure. This affects 4 level-fields. Additionally,
18
Furthermore, the distribution of the level-fields along the linkage strength axis is similar for both categories – vocational and general. The heterogeneity in linkage strength
within vocational and general education is much higher than the difference between the
two categories. This represents strong evidence that the vocationality of educational programmes is indeed gradual in nature and that a dichotomization omits this heterogeneity
in linkage occupational specificity.
A closer look at single categories shows that among the level-fields classified as general,
there are more fields with very low linkage but also the highest linking field in the whole
sample is classified as general. The two highest linking fields in the general category
are Law, Public Administration (linkage = 3.172) and Health Care (4.494), both on the
doctorate level. From the law programme, about 65 percent of students are found in ISCO
category 261 (Legal Professionals) and another 10 percent become Regulatory Government
Associate Professionals (ISCO 335). From the health care programme, 69 percent of
students end up in ISCO category Medical Doctors (ISCO 221) with all other occupations
staying far below 10 percent.
Among the vocational programmes, Teachers for Technical Subjects and Transport
(linkage = 3.141) on the level of MBO-2 and the field Transport and Logistics (3.230)
on HBO level have the highest linkage values. Students from the MBO-2 programme
most often enter ISCO categories 516 (Other Personal Service Workers) with 48 percent
and 833 (Heavy Truck and Bus Drivers) with 10 percent. Graduates from HBO Transport and Logistics are most often found in ISCO 315 (Ship and Aircraft Controllers and
Technicians) with 60 percent while all other occupations stay far below 10 percent.
The weakest linking programmes are the general educational programmes in upper
secondary education with HAVO (year 4-5) having a value of 0.178 and VWO (year 4-6)
with a local segregation of 0.212. From these programmes, graduates spread out widely
across occupations with no occupation receiving more than 6 percent of graduates.
Among the vocational programmes, the lowest linkage is attached to the MBO-4 programme Commerce with a value of 0.448 and to the VMBO (theoretical track) programme
of Administration, Secretarial with a local segregation of 0.469. From the commerce programme about 20 percent of graduates become Shop Salespersons (ISCO 522). From the
administration programme, 11 percent are found in the category of Other Clerical Support
Workers (ISCO 411) and another 11 percent in the category of Shop Salespersons (ISCO
522). It can be seen that these percentages are much lower than for the highly linking
educational level-fields.
Like expected, commerce oriented programmes in vocational education show a rather
low linkage whereas health and law oriented training in general education is associated
all those level-fields which do not have information on the fourth-digit or are categorized as Other are
excluded from the figure to avoid bias. This is the case for 24 categories. The data for the graph consists
of the 166 remaining categories.
19
Figure 3: Average linkage strength within vocational and general programmes for different
levels of education10
with a high linkage strength. These programmes do not at all fit the traditional classification of vocational vs. general.
Additionally, it can be seen that the highly linking programmes in general education
are found in very high levels of education (doctorate) whereas the low linking programmes
in vocational education are rather observed at intermediate levels. To gain more insight
in the interplay of level and fields in the contribution to linkage, Figure 3 explores linkage
strength by educational level and by vocational versus general.
In pre-primary and primary education as well as on doctorate level, only programmes
exist which are classified as general. Post-secondary education on the other hand is fully
classified as vocational. All other levels of education offer both, vocational and general
programmes. The average linkage strength is rather low for primary (linkage = 0.561),
lower-secondary general (0.627) and upper-secondary general education (0.226). Preprimary education has a somewhat higher average linkage with 0.902. This is probably
due to the limited occupational options for individuals with such a low level of formal
education. In contrast, vocational programmes on secondary school level have a rather
high average linkage strength with 1.062 for lower and 1.318 for upper secondary educa10
Again the harmonized variable for vocational is used as described for Figure 2. For the educational
levels a similar problem is apparent as the level information is taken from the ISCED classification while
linkage strength is oriented towards SOI categories. The level-fields are related to educational levels by
taking the mean level for each level-field and harmonizing it towards the closest full number for the level.
20
tion. In post-secondary education linkage of vocational programmes is again low with an
average of 0.485. On tertiary level of education, the average linkage strength for general
education programmes (1.428 on tertiary and 1.488 on doctorate level) is almost as high
as for vocational programmes (1.711). In general, average linkage strength is higher on
these higher levels of education in both categories – vocational and general – compared
to lower levels of education. Much of the heterogeneity in linkage strength among general
education programmes seems to originate from higher education. This finding is interesting as many studies on vocational education focus especially on the secondary school
level and come to the apparently correct conclusion for this level that vocational programmes are more specific. However, this conclusion might be driven by the neglect of
the tertiary level of education. If all levels of education are considered, the heterogeneity
of occupational specificity becomes much clearer.
5.2
School-to-work Transition
In a first set of regression models, it is evaluated if linkage strength explains employment
probabilities in the school-to-work transition better than the dichotomous vocational education measure (hypothesis 1). For this analysis, only respondents with an age from 16 to
35 who are not currently enrolled in education are included to capture early labour market
experiences. Linkage strength is standardized within the samples for men and women to
facilitate comparisons across models. Age is centred at 16 to make interpretation more
accessible. The results for men and women are presented in Table 4. The sample for men
consists of 29,982 observations, the sample for women contains 30,070 observations.
In the first model, only the standardized linkage strength is included in addition to the
control variables, level of education, age, region and survey year. The coefficient is positive
and significant for both men and women. This is in line with hypothesis 1.1. An increase
of linkage strength by one standard deviation multiplies the odds of being employed with
e0.154 = 1.166 for men and e0.170 = 1.185 for women. The effect for men and women is
comparable in size, although slightly higher for women. Among the control variables it
can be noted that increasing levels of education consistently lead to higher employment
probabilities. An exception is the highest level (doctorate) which seems not to improve
labour market outcomes. This might be due to the young age of the respondents in this
sample in which doctorates have only been obtained very recently. Age takes a curvilinear
pattern with the coefficient for age being positive and the coefficient for age squared being
negative. Employment probabilities increase with age but this increase stagnates at higher
ages.
In the second model, the traditional dichotomous measure for vocationality is used
instead of the linkage strength. The coefficient here is also positive and significant for
men and women. Being in a vocational in contrast to a general programme, leads to the
21
Table 4: Logistic Regressions for the School-to-work Transition (16–35 Years) with Employment Probability as Dependent Variable
Men
Model 1
Linkage Strength (standardized)
Model 3
Model 1
0.154***
0.138***
0.170***
0.152***
(0.024)
(0.033)
(0.024)
(0.029)
Vocational (ref=general)
Model 2
Women
Model 2
Model 3
0.243***
0.051
0.222***
0.070
(0.050)
(0.069)
(0.048)
(0.057)
0.560***
0.463***
0.549***
0.624***
0.521***
0.613***
(0.159)
(0.159)
(0.160)
(0.157)
(0.156)
(0.157)
1.236***
1.050***
1.202***
1.348***
1.147***
1.302***
(0.145)
(0.148)
(0.153)
(0.143)
(0.145)
(0.148)
upper secondary
1.767***
1.587***
1.729***
2.377***
2.190***
2.322***
(0.144)
(0.148)
(0.152)
(0.141)
(0.146)
(0.148)
post secondary
2.024***
1.816***
1.977***
2.657***
2.468***
2.591***
(0.233)
(0.239)
(0.240)
(0.202)
(0.207)
(0.208)
2.105***
2.194***
2.113***
2.970***
3.090***
2.980***
(0.149)
(0.148)
(0.150)
(0.145)
(0.144)
(0.145)
1.607***
1.893***
1.635***
2.292***
2.698***
2.233***
(0.395)
(0.393)
(0.395)
(0.351)
(0.341)
(0.349)
Age (centred at 16)
0.204***
0.206***
0.203***
0.068***
0.070***
0.066***
(0.014)
(0.014)
(0.014)
(0.013)
(0.013)
(0.013)
Age squared (centred at 16)
-0.005***
-0.005***
-0.005***
-0.003***
-0.003***
-0.003***
(0.001)
(0.001)
(0.001)
(0.001)
(0.001)
(0.001)
-1.572***
-1.601***
-1.568***
-1.451***
-1.486***
-1.441***
(0.176)
(0.176)
(0.176)
(0.171)
(0.171)
(0.172)
yes
yes
yes
yes
yes
yes
Level of Education (ref=pre-primary)
primary
lower secondary
tertiary
doctorate
Constant
Region fixed effects
Year fixed effects
N
Pseudo2
BIC
yes
yes
yes
yes
yes
yes
29,982
29,982
29,982
30,070
30,070
30,070
0.114
0.113
0.114
0.110
0.109
0.110
991,695
992,855
991,668
1,214,902
1,215,745
1,214,809
Source: EBB 2010–2012, own calculations
Note: Logged odds are displayed, standard errors in parentheses. Sampling weights are applied.
Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001
odds of employment being e0.243 = 1.275 times higher for men and e0.222 = 1.249 times
higher for women. This effect is that large since it is the maximum effect of going from
general to vocational while the effect of linkage strength presented above only presents
the effect of an increase by one standard deviation.
Finally, both predictors, linkage strength and the vocational dummy are included in
a third model. The coefficient for linkage strength is still positive and significant at
a similar order of magnitude as before. A one standard deviation increase of linkage
strength multiplies the odds of being employed for men with e0.138 = 1.148 and for women
with e0.152 = 1.164. The coefficient for vocational education is now insignificant and much
smaller (0.051 for men and 0.070 for women). This indicates in line with hypothesis 1.2
22
that the effect of vocational education on employment probabilities explained by the level
of linkage strength of educational programmes. With the linkage strength measure we
capture more relevant variance in employment probabilities than with the simple dummy.
Linkage strength improves the explanation of labour market outcomes.
Also a look at pseudo R2 shows that the dichotomous vocational variable does not
add to the model fit. For men, the pseudo R2 is 0.114 in the first and third model and is
slightly lower with 0.113 in the model in which only the dichotomous vocational variable
is included. For women, a similar pattern is visible with a pseudo R2 of 0.110 in the first
and third model and a value of 0.109 in the second model.
All models are additionally assessed with the KHB technique. The results of this
analysis are presented in Appendix D. The analysis shows that the coefficients for linkage
strength and the vocational dummy are slightly over-estimated when only one variable is
entered and when no correction for re-scaling is applied. However, the changes in the size
of the coefficients are minor and do not lead to substantively different conclusions. The
results of the regular logistic regression are robust in this respect.
To illustrate the substantive meaning of the results, a graph with predicted probabilities of being employed for different levels of linkage strength and mean levels of all other
variables is presented in Figure 4.
Predicted employment probabilities are unsurprisingly high in this sample of young
people who are not currently in education. At a low linkage strength of about 0.2, the
predicted employment probability is at 86.7 percent for men and 78.6 percent for women.
At an average level of linkage strength of about 1.2, the predicted employment probability
for men has increased to 89.3 percent and that for women to 82.8 percent. At a high level
of linkage strength of 2.4, the predicted employment probability for men is 92 percent
and that for women 87 percent. These increases show that the linkage strength of one’s
educational level-field has a considerable influence on employment in this young sample.
From low to high values of linkage strength, the predicted employment probability of men
rises by 5.3 percentage points and that of women even by 10.3 percentage points.
It is interesting to see that linkage has a higher influence for women than for men.
It is known that men and women are highly segregated in different occupations on the
labour market (Alonso-Villar et al., 2012; Hegewisch and Hartmann, 2014) and linkage
may work quite differently in those occupations. It might also well be, that the linkage
strength of women and men differs – a possibility which exploration is beyond the scope
of this paper.
11
The unstandardized measure of linkage strength is used for these graphs. The graph shows predicted
probabilities for the 1st to the 99th percentile of linkage strength and excludes the outliers with extreme
linkage strength.
23
Figure 4: Predicted Probabilities of Being Employed at Different Levels of Linkage
Strength11
(a) Men
(b) Women
24
5.3
Life-course
A second set of regression models evaluates the employment probabilities of individuals
with different educational level-fields over their labour market career (hypothesis 2). In
these models, all respondents from 16 to 65 who are not in education are included. Linkage
strength is again standardized within samples and age centred at 16. Again separate
models for men and women are presented in Table 5. The number of observations in the
sample is 107,047 for men and 108,957 for women.
The first model includes age, age squared and the linkage strength measure next to a
number of controls (educational level, region and survey year). Linkage strength is again
positive and significant for both, men and women. Also in the more extended age sample
Table 5: Logistic Regressions for the Life-course (16-65 Years) with Employment Probabilities as Dependent Variable
Men
Linkage Strength (standardized)
Age (centred at 16)
Age squared (centered at 16)
Women
Model 1
Model 2
Model 1
Model 2
0.063***
0.221***
0.073***
0.267***
(0.012)
(0.026)
(0.011)
(0.024)
0.214***
0.208***
0.118***
0.113***
(0.003)
(0.003)
(0.003)
(0.003)
-0.005***
-0.005***
-0.003***
-0.003***
(0.000)
(0.000)
(0.000)
Linkage Strength × Age
(0.000)
-0.005***
-0.007***
(0.001)
(0.001)
Educational level (ref=pre-primary)
primary
0.630***
(0.082)
(0.081)
(0.076)
(0.076)
lower secondary
1.178***
1.183***
1.406***
1.400***
(0.076)
(0.075)
(0.072)
(0.071)
1.582***
1.584***
2.095***
2.099***
(0.075)
(0.074)
(0.071)
(0.071)
1.651
1.643***
2.300***
2.314***
upper secondary
post secondary
tertiary
doctorate
Constant
BIC
0.865***
(0.089)
(0.089)
(0.083)
(0.083)
1.996***
2.590***
2.596***
(0.076)
(0.076)
(0.072)
(0.072)
2.429***
2.527***
2.653***
2.661***
(0.178)
(0.182)
(0.201)
(0.211)
-1.606***
-1.513***
-1.722***
-1.644***
(0.096)
(0.097)
(0.087)
(0.088)
yes
yes
yes
yes
Year fixed effects
Pseudo R2
0.879***
1.978***
Region fixed effects
N
0.620***
yes
yes
yes
yes
107,047
107,047
108,957
108,957
0.175
0.176
0.163
0.164
3,703,354
3,700,321
4,759,499
4,754,605
Source: EBB 2010–2012, own calculations
Note: Logged odds, standard errors in parentheses. Sampling weights are applied.
Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001
25
of 16 to 65, the linkage of an educational programme has a positive influence on the
probability of being employed. The size of the coefficient is smaller than in the first set
of regressions. Now, an increase of linkage strength by one standard deviation multiplies
the odds of being employed with e0.063 = 1.065 for men and e0.073 = 1.076 for women.
This result is intuitive as now also older respondents are included in the sample and we
expect a less positive influence of linkage strength for them.
The influence of age takes again a curvilinear pattern: The coefficient for age is positive and significant showing that an increase of age by one year leads to e0.214 = 1.239
times higher odds of being employed for men and e0.118 = 1.125 times higher odds for
women. This increase is quite substantive. Interestingly, the increase for women is less
strong indicating different career patterns for men and women. The squared term of
age is negative and significant for both sexes. This shows that even if the employment
probability increases with age, this increase slows down and reverses finally. The effect of
level of education is consistently positive now. Higher levels lead to a higher employment
probability. In contrast to the school-to-work analysis, the doctorate now shows an effect
that is highly positive. It seems that individuals with a doctorate cannot develop their
full labour market outcomes yet if only young respondents are considered. However, when
the larger age sample (up to 65) is used, they have had enough time to find labour market
positions.
To assess whether life-course patterns of employment probabilities also depend on
the level of linkage strength, in Model 2, an interaction term between age and linkage
strength is added to the model. This interaction term is negative and significant (−0.005
for men and −0.007 for women). The main effect of linkage strength is highly positive
and significant. For a 16-year-old male, an increase of linkage by one standard deviation
leads to e0.221 = 1.247 times higher odds of being employed. For a female of the same
age the odds are multiplied with e0.267 = 1.306. The coefficient for age remains more or
less at the same size as in the model without the interaction. The negative interaction
implies that with increasing age, the positive effect of linkage strength on employment
probabilities decreases and reverses to a negative effect at one point of age.
Pseudo R2 is reasonably high in the models but does not get enhanced much by
adding the interaction effect. For men, the value rises from 0.175 to 0.176. For women,
an increase from 0.163 to 0.164 is visible. This indicates that the interaction effect is not
highly important for the fit of the model on the data although it is significant.
Again, the KHB method is used to show the coefficients with a correction for re-scaling
bias. The sizes of the coefficients hardly differ between the models with and without KHB.
The results are again presented in Appendix D.
In principle, the result is in line with hypothesis 2. Strongly linking educational
programmes are beneficial early in the life-course but the benefits decrease over the lifecourse. However, looking at the large size of the positive main effect of linkage strength and
26
the relatively small effect size of the interaction effect, it is still not clear how substantive
those later disadvantages are.
To illustrate the main findings of the table and the influence of linkage strength over
the life-course, two sets of graphs are presented.
Figure 5 shows predicted probabilities of being employed over the life-course for three
different levels of linkage strength: low linkage (10th percentile of the distribution), average linkage and high linkage (90th percentile). It can be seen that employment probabilities rise for all individuals up to the age of about 40 years for men and 35 years for
women and then begin a decline. This curvilinear pattern of age with a peak in the mid
thirties reflects the results of previous studies (e.g. Hanushek et al., 2011).
Looking at the different levels of linkage strength, individuals with higher linkage
start out at a higher level of predicted employment probabilities. At an age of 16, male
graduates of highly linking programmes have a predicted probability of being employed of
63.8 percent, whereas the probability is only 50 percent for graduates from lowly linking
programmes. For women, the respective predicted probabilities are 70.5 and 53.4 percent.
This advantage for individuals from highly linking programmes remains substantive until
about 30 years of age. It shows the strong benefit of vocationality in the transition from
school to work.
With increasing age, the benefit of linkage strength decreases. However, it can be seen
that the predicted employment probabilities of individuals in highly linking programmes
(90th percentile) stay higher than for other individuals up to shortly before 60 years of age.
This is a much later point in time than predicted by Hanushek et al. (2011). After this
time point, a higher linkage strength leads to lower predicted employment probabilities
than a lower linkage if all other variables are kept at their mean. However, as predicted
employment probabilities for all individuals steeply decrease at that age, the level of
linkage strength does not make much difference for employment probabilities. Given the
much stronger differences for different levels of linkage strength earlier in the career, one
can interpret the life course pattern rather as a convergence than a real decline with
disadvantages for graduates from highly linking educational programmes. The patterns
of men and women are similar here, although employment probabilities for women are on
average lower and start to decline much earlier in the life-course than those of men.
27
Figure 5: Predicted Probabilities of Being Employed for Different Levels of Linkage
Strength over the Life-course (Other Variables at Mean)
(a) Men
(b) Women
28
Figure 6: Conditional Marginal Effects of Linkage Strength at Different Ages
(a) Men
(b) Women
A graph of the conditional marginal effect of linkage strength shows the influence of
vocationality more clearly (Figure 6).
The effect of linkage strength is high and positive for young ages but declines and
finally reaches zero at an age between 55 and 60. This point of reversion is very similar
for men and women. However, it can be noted that for women the influence of linkage
strength stays high for a longer period in the life-course. After the age of 60 the effect
of linkage strength becomes slightly negative. For women this negative effect reaches
significance at an age of 65 (p=0.001). Also for men, the effect borders significance
(p=0.048). Although these effects are significant at the very last moments of the labour
market career, the graphs show that while the advantages of vocationality disappear
with increasing age, the effect does not become substantively negative. The largest part
of their career, graduates from highly vocational programmes benefit from their training.
Therefore, we cannot speak of a vocational decline in the sense that vocationality becomes
a penalty if the whole life-course is considered.
6
Discussion
We analysed school-to-work linkages in the Netherlands, resulting in two major findings.
First, a measure for vocationality was introduced to the study of education and labour
market outcomes that builds on the linkage between detailed educational programmes
and labour market positions. With this measure it could be shown that the average occupational specificity of educational programmes which are labelled as vocational is not
different from that in general educational programmes. Instead, differences within these
categories are far bigger than differences between them. Especially on the tertiary level of
education, general education programmes link almost as strongly to occupational positions
as vocational programmes. Furthermore, this gradual measure of vocationality proved to
be a stronger predictor of labour market outcomes than the traditional dichotomous mea-
29
sure of vocational education. With the new measure we better exploit the heterogeneity
in vocationality and could show the labour market outcomes of different educational levelfields in greater detail. Second, the hypothesis about vocational life-course decline was
re-assessed using detailed data from the EBB and the new linkage measure. Results show
that graduates from educational programmes with a high occupational specificity, indeed,
experience a decline of positive outcomes over their life-course compared to their peers
with general education. However, while this decline leads to a convergence of outcomes,
it does not result in disadvantages for individuals with highly linking education. Analyses
with predicted probabilities of employment over the life-course could show that the decline is not strong enough to outweigh initial benefits of occupational specificity. Instead
of a vocational penalty (Hanushek et al., 2011), we find a convergence of labour market
outcomes. Vocational education does not become a burden in later life.
These findings have several implications for the way scholars should look at vocational
education. First, the clear individual level benefits of occupational specificity that are
found for the school-to-work transition confirm earlier studies on the topic (Shavit and
Müller, 1998; Breen, 2005; Scherer, 2005). In the Netherlands, occupational specificity
is relevant for the labour market allocation of students and the effect cannot be reduced
to mere differentiation in the educational system. If occupational specificity is mostly
driven by skills, signalling, networks or closure is still an open question that can be
addressed by further research. Second, the heterogeneities in the occupational specificity
of educational level-fields show that the traditional operationalization of vocationality in
a dichotomy of specific and general education is not suitable for measuring vocationality.
These results also show that we should think about vocationality not in different school
types but focus more on single educational level-fields with their own individual degree
of occupational specificity. University degrees may be as specific as the programmes in
secondary education which were classically defined as vocational programmes. Therefore,
we strongly advise future research to account for this gradual nature of vocationality. In
comparative research, the linkage measure has the potential to build a more consistent
measure of vocationality across countries. It is based on the actual specificity of individual
programmes and can easily be aggregated on the country level. Third, the convergence of
life-course employment probabilities in the Netherlands does not confirm previous research
that would have predicted a decline (Hanushek et al., 2011). There might be several
reasons for this missing decline that should be investigated by further studies. On-the-job
training might effectively prevent the decline of skills in the Netherlands. The Macro-level
structure of the labour market in the Netherlands could be another factor that advantages
vocational skills. If vocational advantage works through the signalling mechanism, it could
also be the case that educational signals simply lose their influence the more experience
an individual has gathered on the labour market. Then, the educational linkage strength
does not predict outcomes anymore for older workers. Finally, also the way in which
30
licensure or other closure mechanisms work in the Netherlands might be responsible for
the convergence. Again, the exact mechanisms are to be determined by further research.
In all of these cases, the clear finding remains: vocational education is beneficial for the
employment of graduates throughout the life-course in the Netherlands.
The consequences of these findings are also relevant for policies surrounding vocational
education. In the Netherlands, occupation specific training can be considered suitable for
increasing employment especially among young people. By fostering a high vocationality
in educational programmes, graduates benefit in the beginning of their career without
facing substantive disadvantages later. This finding also shows that the vocational content
of programmes within educational levels is important when looking for ways to optimize
the employability of graduates. This advice should be especially considered in debates
that seek to focus policy on increasing the educational level of all graduates instead of
reforming the content of programmes within levels. As the heterogeneity of vocationality
within programmes which are categorized as vocational or general is high, it will also
be valuable to look beyond those broad categories when trying to improve occupational
specificity of education. Not the expansion of a single school type like MBO will improve
labour market prospects of youth but the focus on vocationality within single educational
programmes. Of course, this employability might stand in a trade-off to other functions
of education like the enhancement of equality of opportunity or civic participation which
we have not considered in this paper. However, if labour market allocation is in focus,
vocationality seems to have considerable benefits for students.
These findings and their consequences are substantive but the study also has some limitations that should be addressed by further research. First, the most obvious limitation
is that the study only covers one country case. Linkage strength might work differently in
other national contexts which makes it important to extend the findings and re-evaluate
the influence of linkage strength in other countries. Determining the linkage strength for
a number of countries will also enable comparative work with this measurement. Nevertheless, the case of the Netherlands is a typical context for a country with a highly
vocational educational system. Therefore, the results are also indicative for other countries which have a similar structure of training and labour market systems. Second, the
effects we found are embedded in the context of other individual characteristics and lifecourse events that should be considered in more extensive research. For example, in our
research we cannot consider individual skill levels and socio-demographic background of
respondents, two factors that might determine both, the selection into a specific form of
education and the labour market outcomes of respondents. Also, linkage strength might
differ between the sexes and between different ethnic groups in society. Another aspect
are other life-course events during the labour market career. Employment probabilities
are most likely not only determined by education and age but other events that occur
during the time between school graduation and retirement. These factors are hard to
31
capture with our cross-sectional data. Using panel data and applying selection models
in future research could refine our results in this respect. Third, the linkage measure
itself needs further development. It is a powerful measure that can show vocationality
in great detail. However, there are still some reliability issues that need to be resolved.
One problem is the minimum cell size for the educational categories that was already
discussed earlier. Another issue with reliability that is already pointed out by DiPrete
et al. (2015) is the fact that linkage strength depends to a great extent on the level of
aggregation of educational and occupational categories. If instead of four-digit SOI and
three-digit ISCO codes, a higher or lower detail-level is used, the grouping of individuals
in educations and occupations might differ and thereby affect linkage strength. These
problems should be addressed in further research. Finally, future research could extend
the application of the new linkage measure to the study of other labour market outcomes
like occupational status and earnings which we did not yet consider in this paper.
This study changes the way we look at vocational education over the life-course: in
the Netherlands, vocationality does not become a penality for graduates and can be evaluated positively throughout the career. We hope that these results for the Netherlands
lead to further research on the life-course consequences of vocational education in a comparative perspective. Thereby, the gradual way of measuring vocationality of individual
programmes which we applied in this study is a fruitful approach for such an endeavour.
References
Allen, J., P. Boezerooy, E. De Weert, and R. Van der Velden (2000). Higher Education
and Graduate Employment in the Netherlands. European Journal of Education 35 (2),
211–219.
Alonso-Villar, O., C. Del Rio, and C. Gradin (2012). The Extent of Occupational Segregation in the United States: Differences by Race, Ethnicity, and Gender. Industrial
Relations: A Journal of Economy and Society 51 (2), 179–212.
Alonso-Villar, O. and C. Del Rı́o (2010). Local Versus Overall Segregation Measures.
Mathematical Social Sciences 60 (1), 30–38.
Andersen, R. and H. G. Van de Werfhorst (2010). Education and Occupational Status in
14 Countries: The Role of Educational Institutions and Labour Market Coordination.
The British Journal of Sociology 61 (2), 336–355.
Arum, R. and Y. Shavit (1995). Secondary Vocational Education and the Transition from
School to Work. Sociology of Education 68 (3), 187.
Becker, G. S. (1964). Human Capital: A Theoretical and Empirical Analysis with Special
Reference to Education. New York: Columbia University Press.
32
Biavaschi, C., W. Eichhorst, C. Giulietti, M. Kendzia, A. Muravyev, J. Pieters,
N. Rodrı́guez-Planas, R. Schmidl, and K. Zimmerman (2013). Youth Unemployment
and Vocational Training. Background Paper for the World Development Report 2013.
Technical report, Worldbank.
Bills, D. B. (1988). Educational Credentials and Promotions: Does Schooling Do More
than Get You in the Door? Sociology of Education, 52–60.
Bol, T. (2014). Economic Returns to Occupational Closure in the German Skilled Trades.
Social Science Research 46, 9–22.
Bol, T. and K. A. Weeden (2014). Occupational Closure and Wage Inequality in Germany
and the United Kingdom. European Sociological Review , 1–16.
Brauns, H., S. Scherer, and S. Steinmann (2003). The CASMIN Educational Classification
in International Comparative Research. In J. H. P. Hoffmeyer-Zlotnik and C. Wolf
(Eds.), Advances in Cross-National Comparison. A Working Book for Demographic
and Socio-economic Variables, pp. 221–244.
Brauns, H. and S. Steinmann (1997). Educational Reform in France, West-Germany,
the United Kingdom and Hungary: Updating the CASMIN Educational Classification.
Citeseer.
Breen, R. (2005). Explaining Cross-national Variation in Youth Unemployment: Market
and Institutional Factors. European Sociological Review 21 (2), 125–134.
Breen, R., K. B. Karlson, and A. Holm (2013). Total, Direct, and Indirect Effects in Logit
and Probit Models. Sociological Methods & Research 42 (2), 164–191.
De Graaf, P. M. and W. C. Ultee (1998). Education and Early Occupation in the Netherlands around 1990: Categorical and Continuous Scalres and the Details of a Relationship. In Y. Shavit and W. Müller (Eds.), From School to Work. A Comparative Study
of Educational Qualifications and Occupational Destinations.
DiPrete, T. A., T. Bol, C. Ciocca, and H. Van de Werfhorst (2015). School-to-Work
Linkages in the United States, Germany and France. Under review .
Erikson, R. and J. H. Goldthorpe (1992). The Constant Flux: A Study of Class Mobility
in Industrial Societies. Oxford: Clarendon Press.
Estevez-Abe, M., T. Iversen, and D. Soskice (2001). Social Protection and the Formation
of Skills: A Reinterpretation of the Welfare State. In Varieties of Capitalism. The
Institutional Foundations of Comparative Advantage. Oxford University Press.
33
Frankel, D. M. and O. Volij (2011). Measuring School Segregation. Journal of Economic
Theory 146 (1), 1–38.
Hanushek, E. A., L. Woessmann, and L. Zhang (2011). General Education, Vocational
Education, and Labor-market Outcomes over the Life-cycle.
Hegewisch, A. and H. Hartmann (2014). Occupational Segregation and the Gender Wage
Gap: A Job Half Done. Technical report, Institute for Women’s Policy Research
(IWPR).
Iannelli, C. and D. Raffe (2007). Vocational Upper-Secondary Education and the Transition from School. European Sociological Review 23 (1), 49–63.
Karlson, K. B. and A. Holm (2011). Decomposing Primary and Secondary Effects: A New
Decomposition Method. Research in Social Stratification and Mobility 29 (2), 221–237.
Karlson, K. B., A. Holm, and R. Breen (2012). Comparing Regression Coefficients Between Same-sample Nested Models Using Logit and Probit A New Method. Sociological
Methodology 42 (1), 286–313.
Katz, L. F. and K. M. Murphy (1992). Changes in Relative Wages, 1963 - 1987: Supply
and Demand Factors. The Quarterly Journal of Economics 107 (1), 35–78.
König, W., P. Lüttinger, and W. Müller (1988). A Comparative Analysis of the Development and Structure of Educational Systems. Methodological Foundations and the
Construction of a Comparative Educational Scale. CASMIN Working Paper.
Levels, M., R. Van der Velden, and V. DiStasio (2014). From School to Fitting Work.
How Education-to-job Matching of European School Leavers is related to Educational
System Characteristics. Acta Sociologica 57 (4), 341–361.
Müller, W. and M. Gangl (2003). Transitions from Education to Work in Europe: The
Integration of Youth into EU Labour Markets. Oxford University Press Oxford.
Müller, W., P. Lüttinger, W. König, and W. Karle (1989). Class and Education in
Industrial Nations. International Journal of Sociology, 3–39.
Mora, R. and J. Ruiz-Castillo (2011). Entropy-based Segregation Indices. Sociological
Methodology 41 (1), 159–194.
OECD and ILO (2014). Promoting better Labour Market Outcomes for Youth. Report on
Youth Employment and Apprenticeships prepared for the G20 Labour and Employment
Ministerial Meeting Melbourne, Australia, 10-11 September 2014. Technical report,
OECD, ILO.
34
Reardon, S. F. and G. Firebaugh (2002). Measures of Multigroup Segregation. Sociological
methodology 32 (1), 33–67.
Rosenbaum, J. E., T. Kariya, R. Settersten, and T. Maier (1990). Market and Network
Theories of the Transition from High School to Work: Their Application to Industrialized Societies. Annual Review of Sociology 16 (1), 263–299.
Scherer, S. (2005). Patterns of Labour Market Entry - Long Wait or Career Instability? An
Empirical Comparison of Italy, Great Britain and West Germany. European Sociological
Review 21 (5), 427–440.
Shavit, Y. and W. Müller (1998). From School to Work. A Comparative Study of Educational Qualifications and Occupational Destinations. ERIC.
UNESCO Institute for Statistics (2012). International Standard Classification of Education: ISCED 2011. Montreal, Quebec: UNESCO Institute for Statistics.
Van de Werfhorst, H. G. (2011). Skill and Education Effects on Earnings in 18 Countries:
The Role of National Educational Institutions. Social Science Research 40 (4), 1078–
1090.
Van der Velden, R. and M. Wolbers (2003). The Integration of Young People into the
Labour Market: The Role of Training Systems and Labour Market Regulation. Oxford
University Press.
Vogtenhuber, S. (2014). Explaining Country Variation in Employee Training: An Institutional Analysis of Education Systems and Their Influence on Training and Its Returns.
European Sociological Review , 1–14.
Weeden, K. A. (2002). Why Do Some Occupations Pay More than Others? Social Closure
and Earnings Inequality in the United States. American Journal of Sociology 108 (1),
55–101.
Wolbers, M. H. J. (2007). Patterns of Labour Market Entry: A Comparative Perspective
on School-to-Work Transitions in 11 European Countries. Acta Sociologica 50 (3), 189–
210.
35
Appendices
A
The Educational System in the Netherlands
An overview of the Dutch educational system is presented in Figure A1. The figure
shows the different school types and in parentheses the respective SOI level. The full
names of the Dutch school types are presented in table A1. From age 0 to 4, children
attend different form of pre-primary institutions. At an age of four years they enter
primary school (basisschool ) of which the first two grades are still considered as preprimary education on SOI-level 10. The remaining six years of basisschool are subsumed
under level 20 (primary education).
From an age of twelve, the Dutch educational systems is differentiated in different
tracks. In the first three to four years of the tracked system, VWO, HAVO and the
theoretical track of VMBO are considered as lower secondary education, high level (33).
The more basic VMBO tracks are considered as lower secondary intermediate (32) in the
SOI classification. Vocational courses which are termed Praktijkonderwijs are considered
as low level lower secondary education (31).
In upper secondary education, the last three years of VWO as well as MBO level 4
programmes (middle-management training) are considered as high level (43), the two final
years of HAVO and MBO level 3 programmes (professional training) are categorized as
intermediate (42), and MBO level 2 programmes (basic vocational training) are seen as
low level (41). While VWO and HAVO are considered as general education, the different
MBO programmes are classified as vocational in the dichotomous measure of vocationality.
Figure A1: Educational System of the Netherlands
0
1
2
3
4
Kindercentra,
Peuterspeelzalen (10)
5
6
Basissch.
(10)
7
8
9
Age
10 11
Basisschool (20)
12
13
14
15
16
VWO (33)
HAVO (33)
17
18
19
VWO (43)
HAVO
(42)
20
1
2
years of education
6 -- 10
3
4
5
WO bachelor
(53)
WO master (60)
WO (70)
HBO bachelor (52)
Short
HBO (51)
VMBO theoretical
(33)
VMBO basic (32)
MBO-4 (43)
MBO-3 (42)
MBO-2
(41)
Praktijkonderwijs (31)
Note: The numbers in parentheses are the level as used in SOI and in the level-field code in this paper.
36
Table A1: School Types in the Dutch Educational System
Abbreviation
Full name
Translation
VMBO
Voorbereidend middelbaar beroepsonderwijs
Preparatory intermediate vocational
education
HAVO
Hoger algemeen voortgezet onderwijs
Higher general continued education
VWO
Voorbereidend wetenschappelijk onderwijs
Preparatory academic education
MBO
Middelbaar beroepsonderwijs
Intermediate vocational education
HBO
Hoger beroepsonderwijs
Higher vocational education
WO
Wetenschappelijk onderwijs
Academic Education, University
In higher education it is distinguished between scientific WO bachelor programmes
(53) and professional HBO programmes (51 and 52). The master (60) and doctorate
levels (70) are not differentiated in different sub-levels.
B
Analysis of Sparse-cell Bias
When obtaining the local segregation of categories, the number of observations in a certain
level and field of education plays a critical role. If sizes of level-field cells are very low, the
local segregation is likely to be overestimated. Single cases are much more influential in
those small cells. If, for example, in a cell with only 10 observations, two by coincidence
have the same occupation, this already constitutes 20 percent clustering in one occupation.
This leads to a high fluctuation of the linkage strength measure due to randomness and
overall likelihood to observe stronger linkages for the cell. This, of course, is a problem
for the reliability of the local segregation measure. To gain insight in the nature of this
problem and when it affects the outcome of the local segregation measure, an analysis is
done on the question what minimum cell size is required to obtain a reliable measure of
local segregation.
For this purpose, the local segregation is simulated for different cell sizes. From the
original level-fields, random samples of different sizes are drawn and used as a sort of
forced cell size. Starting with a forced cell size of 500, samples are drawn in steps of 20
and going down to a forced cell size of 20. For each of these forced cell sizes a sample is
drawn 10 times. For all these artificially lowered cell sizes, local segregation is calculated.
As in each of the 10 repetitions, the sample should be different, this analysis results in 10
different local segregation measures for each forced cell size from 20 to 500 and for each of
the 351 level-fields.12 This analysis results in a data set with 250 local segregation values
for each level-field (25 sample sizes x 10 repetitions). This data is used to determine the
minimum cell size acceptable for the segregation analysis.
12
For cells which were already smaller than 500 in the beginning, the 10 repetitions do not lead to 10
different values as long as the forced cell size exceeds the actual full cell size.
37
To determine at what cell size the local segregation measure starts to get inflated, the
mean of the 10 repetitions in each forced cell size is compared to the mean at cell size 500,
a cell size at which inflation should not be a big problem anymore. A further question is,
when a local segregation value should be considered as inflated or unreliable. Reliability
is defined as a maximum of deviation from the mean local segregation for sample size 500.
A lower ratio, thereby, means a stricter criterion for reliability. For example, if the ratio
is 1.1., each exceeding of this ratio is considered an unreliable local segregation measure.
Different ratios of deviation from the mean at cell size 500 are displayed to evaluate this
question with different levels of strictness. The results are shown in Figure B1.
Figure B1: First Exceeding of Ratios (from 500 downwards)
(a) All cells
(b) All cells larger than 500
The figure shows the cumulative percentage of level-fields of which local segregation
exceeds a certain ratio (1.1, 1.2, 1.5, 2.0) at a certain cell size compared to the local
38
segregation of the same level-field at cell size 500. Sub-figure (a) shows this percentage
for all 351 level-fields in the data set, Sub-figure (b) is restricted to those 90 level-fields
which are larger than 500 observations in the original data. In Sub-figure (a), it can be
seen that if a ratio of 2.0 or 1.5 is chosen, the measure only becomes unreliable if the cell
size is lower than 100. However, if 1.2 is chosen as a ratio, only cell sizes larger than 200
can be considered as reliable. For a ratio of 1.1, even a cell size of about 360 or higher
would be necessary if no deviations were to be tolerated. In Sub-figure (b), the 1.5 ratio
starts to get exceeded more often if cells are smaller than 120. For ratio 1.2, again almost
300 cases per cell would be necessary. And a ratio of 1.1 is only accomplished with 400
cases or more.
This result stands in an obvious trade-off with the amount of data per cell that is
available. To obtain a reliable measure and to be able to evaluate a high number of
educational and occupational categories, a minimum cell size of 120 is chosen for this
paper - thereby, the local segregation measure remains at least largely within the deviation
ratio of 1.5.
39
C
Segregation Analysis: List of Educational and Occupational Categories
Table C1: Educational Categories, Linkage Strength and Frequency
Level
Field
SOI
Freq.
Pre-primary education
Primary education
Lower secondary education, low level
Lower secondary education, low level
Lower secondary education, low level
Lower secondary education, intermediate
Lower secondary education, intermediate
Lower secondary education, intermediate
Lower secondary education, intermediate
General education
General education
General education
Languages
Other
Administration, secretarial
Engineering
Construction
Metal processing, vehicle and tool manufacturing
Care, community services
Other
General education
Humanities, social sciences, communication
and arts with differentiation
Commercial
Administration, secretarial
Engineering general
Electrical engineering
Construction
Metal processing, vehicle and tool manufacturing
Process technology
Textile and leather processing, other
Agriculture
Care, community services
Hotels, gastronomy, tourism and leisure
Transport and logistics
Hotels, gastronomy, tourism, leisure, transport
and logistics with differentiation
Other
Teachers general education
Teachers technical subjects and transport
Teachers (health) care, sports and other
Commercial
Management
Administration, secretarial
Public order, security
Mathematics, natural sciences
Computer science
Electrical engineering
Construction
Metal processing, vehicle and tool manufacturing
Process technology
Agriculture
1001
2001
3101
3121
3198
3235
3260
3263
3264
1,263
7,862
670
682
240
141
130
127
160
0.902
0.561
0.635
0.965
0.783
0.838
1.148
1.750
1.619
3282
3298
3301
3327
393
363
12,131
157
1.085
0.829
0.281
1.263
3332
3335
3361
3362
3363
3364
1,339
1,276
755
1,157
3,294
3,557
0.576
0.525
0.791
0.883
1.213
0.983
3365
3366
3371
3382
3391
3392
3397
483
259
1,257
5,034
370
535
150
0.844
0.974
1.210
0.833
0.889
1.785
1.051
3398
4111
4114
4116
4132
4133
4135
4142
4151
4152
4162
4163
4164
275
155
138
205
710
174
791
382
215
198
362
1,099
1,039
0.807
2.074
3.141
1.890
0.822
0.706
0.656
2.096
1.325
1.479
1.874
1.660
1.315
4165
4171
459
266
1.342
1.494
level
level
level
level
Lower
Lower
Lower
Lower
secondary education, intermediate level
secondary education, intermediate level
secondary education, high level
secondary education, high level
Lower
Lower
Lower
Lower
Lower
Lower
secondary
secondary
secondary
secondary
secondary
secondary
education,
education,
education,
education,
education,
education,
high
high
high
high
high
high
level
level
level
level
level
level
Lower
Lower
Lower
Lower
Lower
Lower
Lower
secondary
secondary
secondary
secondary
secondary
secondary
secondary
education,
education,
education,
education,
education,
education,
education,
high
high
high
high
high
high
high
level
level
level
level
level
level
level
Lower secondary education, high level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
Upper secondary education, low level
40
Linkage
Level
Upper
Upper
Upper
Upper
Upper
secondary
secondary
secondary
secondary
secondary
education,
education,
education,
education,
education,
low
low
low
low
low
level
level
level
level
level
Upper secondary education, low level
Upper secondary education, intermediate level
Upper secondary education, intermediate level
Upper secondary education, intermediate level
Upper secondary education, intermediate level
Upper secondary education,
Upper secondary education,
Upper secondary education,
Upper secondary education,
Upper secondary education,
Upper secondary education,
intermediate level
intermediate level
intermediate level
intermediate level
intermediate level
intermediate level
Upper secondary education,
Upper secondary education,
Upper secondary education,
Upper secondary education,
Upper secondary education,
intermediate level
intermediate level
intermediate level
intermediate level
intermediate level
Upper secondary education,
Upper secondary education,
Upper secondary education,
Upper secondary education,
intermediate level
intermediate level
intermediate level
intermediate level
Upper secondary education,
Upper secondary education,
Upper secondary education,
Upper secondary education,
Upper secondary education,
intermediate level
intermediate level
intermediate level
intermediate level
intermediate level
Upper secondary education, intermediate level
Upper secondary education, high level
Upper secondary education, high level
Upper secondary education, high level
Upper secondary education, high level
Upper secondary education, high level
Upper secondary education, high level
Upper
Upper
Upper
Upper
Upper
Upper
secondary
secondary
secondary
secondary
secondary
secondary
education,
education,
education,
education,
education,
education,
high
high
high
high
high
high
level
level
level
level
level
level
Upper secondary education, high level
Upper secondary education, high level
Field
SOI
Freq.
Health care and community services
Care, community services
Hotels, gastronomy, tourism and leisure
Transport and logistics
Hotels, gastronomy, tourism, leisure, transport
and logistics with differentiation
Other
General education
Teachers
Humanities, social sciences, communication
and arts
Humanities, social sciences, communication
and arts with differentiation
Commercial
Management
Administration, secretarial
Law, public administration
Public order, security
Law, public administration, public order and
security
Computer science
Engineering general
Electrical engineering
Construction
Metal processing, vehicle and tool manufacturing
Process technology
Engineering with differentiation
Agriculture
Agriculture and environment with differentiation
Health care
Care, community services
Hotels, gastronomy, tourism and leisure
Transport and logistics
Hotels, gastronomy, tourism, leisure, transport
and logistics with differentiation
Other
General education
Teachers general education
Teachers (health) care, sports and other
Humanities, social sciences, communication
and arts
Arts, expression
Humanities, social sciences, communication
and arts with differentiation
Commercial
Management
Human resource management, personnel
Administration, secretarial
Law, public administration
Law, public administration, public order and
security
Mathematics, natural sciences
Computer science
4180
4182
4191
4192
4197
163
1,483
654
280
138
1.239
1.180
1.298
2.226
1.197
4198
4201
4210
4220
266
9,941
121
195
0.679
0.198
1.246
0.710
4227
227
1.776
4232
4233
4235
4241
4242
4247
1,305
239
2,077
146
725
316
0.561
0.564
0.633
1.360
1.788
1.096
4252
4261
4262
4263
4264
136
406
437
1,503
1,550
1.575
0.791
1.551
1.471
1.236
4265
4267
4271
4277
226
121
739
132
1.976
1.448
1.446
1.560
4281
4282
4291
4292
4297
1,664
5,592
587
322
193
1.539
1.145
1.542
1.397
1.657
4298
4301
4311
4316
4320
385
4,771
431
454
132
0.842
0.254
1.167
1.224
1.019
4325
4327
598
414
1.141
1.181
4332
4333
4334
4335
4341
4347
5,345
506
215
4,932
474
185
0.501
0.485
0.992
0.812
0.872
0.808
4351
4352
342
684
1.301
1.360
41
Linkage
Level
Upper
Upper
Upper
Upper
secondary
secondary
secondary
secondary
education,
education,
education,
education,
high
high
high
high
level
level
level
level
Upper
Upper
Upper
Upper
Upper
Upper
Upper
Upper
secondary
secondary
secondary
secondary
secondary
secondary
secondary
secondary
education,
education,
education,
education,
education,
education,
education,
education,
high
high
high
high
high
high
high
high
level
level
level
level
level
level
level
level
Upper secondary education, high level
Upper secondary education, high level
Upper secondary education, high level
Upper secondary education, high level
Higher education, first phase, low level
Higher education, first phase, low level
Higher education, first phase, low level
Higher education, first phase, low level
Higher education, first phase, low level
Higher education, first phase, low level
Higher education, first phase, low level
Higher education, first phase, low level
Higher
Higher
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
education, first phase, low level
education, first phase, low level
education, first phase, intermediate
education, first phase, intermediate
education, first phase, intermediate
education, first phase, intermediate
education, first phase, intermediate
Field
SOI
Freq.
Linkage
Engineering general
Electrical engineering
Construction
Metal processing, vehicle and tool manufacturing
Process technology
Textile and leather processing, other
Engineering with differentiation
Agriculture and environment
Agriculture
Health care
Care, community services
Health care and community services with differentiation
Hotels, gastronomy, tourism and leisure
Transport and logistics
Hotels, gastronomy, tourism, leisure, transport
and logistics with differentiation
Other
Commercial
Management
Human resource management, personnel
Administration, secretarial
Engineering
(Health) care and community services
Health care
Hotels, gastronomy, tourism, leisure, transport
and logistics
Transport and logistics
Other
Teachers general education
4361
4362
4363
4364
2,199
1,134
2,915
3,296
1.052
1.190
1.085
0.852
4365
4366
4367
4370
4371
4381
4382
4387
507
675
196
134
2,160
4,255
4,862
1,003
1.363
0.731
1.111
1.324
1.454
1.587
0.972
2.197
4391
4392
4397
1,904
501
254
0.736
1.243
1.505
4398
5132
5133
5134
5135
5160
5180
5181
5190
224
532
239
137
212
161
122
208
137
0.897
0.945
0.668
1.524
1.147
1.571
1.182
2.171
0.895
5192
5198
5211
161
300
4,701
3.230
0.933
2.085
5212
1,587
1.277
5213
420
1.571
5214
520
0.948
Teachers humanities, social sciences, communication and arts
Teachers mathematics, natural sciences, agriculture
Teachers technical subjects and transport
5215
201
1.326
education, first phase, intermediate
Teachers economics, commercial, management
and administration
Teachers (health) care, sports and other
5216
1,147
1.021
education, first phase, intermediate
Education with differentiation
5217
236
2.334
education, first phase, intermediate
Languages
5221
179
1.054
education, first phase, intermediate
Social sciences
5223
758
0.838
education, first phase, intermediate
Communication, media, information
5224
1,072
1.337
education, first phase, intermediate
Arts, expression
5225
1,133
1.346
education, first phase, intermediate
Humanities, social sciences, communication
and arts with differentiation
Economics
5227
188
1.501
5231
1,122
1.128
education, first phase, intermediate
42
Level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
level
Higher
Field
SOI
Freq.
education, first phase, intermediate
Commercial
5232
2,342
0.740
education, first phase, intermediate
Management
5233
1,682
0.641
education, first phase, intermediate
Human resource management, personnel
5234
1,092
1.304
education, first phase, intermediate
Administration, secretarial
5235
1,205
1.623
education, first phase, intermediate
5237
925
0.908
education, first phase, intermediate
Economics, commercial, management and administration with differentiation
Law, public administration
5241
571
1.072
education, first phase, intermediate
Public order, security
5242
210
2.290
education, first phase, intermediate
5247
239
0.857
education, first phase, intermediate
Law, public administration, public order and
security
Mathematics, natural sciences
5251
924
1.293
education, first phase, intermediate
Computer science
5252
807
2.140
education, first phase, intermediate
Engineering general
5261
358
1.115
education, first phase, intermediate
Electrical engineering
5262
1,142
1.436
education, first phase, intermediate
Construction
5263
1,254
1.390
education, first phase, intermediate
5264
1,173
1.164
education, first phase, intermediate
Metal processing, vehicle and tool manufacturing
Process technology
5265
460
1.028
education, first phase, intermediate
Textile and leather processing, other
5266
160
1.198
education, first phase, intermediate
Engineering with differentiation
5267
318
0.993
education, first phase, intermediate
Agriculture
5271
562
0.915
education, first phase, intermediate
Environment
5272
140
1.662
education, first phase, intermediate
5277
202
1.002
education, first phase, intermediate
Agriculture and environment with differentiation
Health care
5281
4,101
1.909
education, first phase, intermediate
Care, community services
5282
3,477
0.903
education, first phase, intermediate
5287
951
1.729
education, first phase, intermediate
Health care and community services with differentiation
Hotels, gastronomy, tourism and leisure
5291
702
0.820
education, first phase, intermediate
Transport and logistics
5292
751
0.908
education, first phase, intermediate
Hotels, gastronomy, tourism, leisure, transport
and logistics with differentiation
Other
5297
615
0.700
5298
129
1.238
5320
136
0.939
education, first phase, intermediate
education, first phase, high level
Humanities, social sciences, communication
and arts
43
Linkage
Level
Field
SOI
Freq.
Higher education, first phase, high level
Higher education, first phase, high level
Higher education, first phase, high level
Languages
Social sciences
Economics, commercial, management and administration
Law, public order
Other
Teachers humanities, social sciences, communication and arts
Teachers mathematics, natural sciences, agriculture
Languages
Humanities other
Social sciences
Communication, media, information
Arts, expression
Economics
Commercial
Management
Human resource management, personnel
Administration, secretarial
Economics, commercial, management and administration with differentiation
Law, public administration
Public order, security
Mathematics, natural sciences
Computer science
Engineering general
Electrical engineering
Construction
Metal processing, vehicle and tool manufacturing
Process technology
Textile and leather processing, other
Agriculture and environment
Agriculture
Health care
Care, community services
Health care and community services with differentiation
Hotels, gastronomy, tourism, leisure, transport
and logistics
Other
Management
Administration, secretarial
Law, public administration
Mathematics, natural sciences
Health care
Other
5321
5323
5330
153
277
298
1.039
0.669
1.012
5341
5398
6012
182
347
473
0.953
0.785
1.995
6013
142
2.415
6021
6022
6023
6024
6025
6031
6032
6033
6034
6035
6037
1,146
624
2,723
357
931
1,040
629
795
175
217
551
1.140
1.442
1.157
1.516
1.409
1.229
0.605
1.000
1.543
2.250
1.241
6041
6042
6051
6052
6061
6062
6063
6064
2,267
214
935
257
509
244
790
324
1.824
2.342
1.272
2.024
1.172
2.077
1.670
1.501
6065
6066
6070
6071
6081
6082
6087
260
205
222
220
933
938
263
1.341
1.417
1.461
1.045
2.032
1.097
1.531
6090
134
1.302
6098
7033
7035
7041
7051
7081
7098
272
312
405
291
355
1,249
461
1.364
1.247
2.352
3.172
1.729
3.494
1.311
Higher education, first phase, high level
Higher education, first phase, high level
Higher education, second phase
Higher education, second phase
Higher
Higher
Higher
Higher
Higher
Higher
Higher
Higher
Higher
Higher
Higher
education,
education,
education,
education,
education,
education,
education,
education,
education,
education,
education,
second
second
second
second
second
second
second
second
second
second
second
phase
phase
phase
phase
phase
phase
phase
phase
phase
phase
phase
Higher
Higher
Higher
Higher
Higher
Higher
Higher
Higher
education,
education,
education,
education,
education,
education,
education,
education,
second
second
second
second
second
second
second
second
phase
phase
phase
phase
phase
phase
phase
phase
Higher
Higher
Higher
Higher
Higher
Higher
Higher
education,
education,
education,
education,
education,
education,
education,
second
second
second
second
second
second
second
phase
phase
phase
phase
phase
phase
phase
Higher education, second phase
Higher
Higher
Higher
Higher
Higher
Higher
Higher
education,
education,
education,
education,
education,
education,
education,
second phase
third phase
third phase
third phase
third phase
third phase
third phase
Source: EBB 2010–2012, own calculations
44
Linkage
Table C2: Occupational Categories - ISCO 3-digit Codes
ISCO
Title
110
Chief executives, senior officials and
legislators
Legislators and senior officials
Managing directors and chief executives
Business services and administration
managers
Sales, marketing and development
managers
Production managers in agriculture,
forestry and fisheries
Manufacturing, mining, construction,
and distribution managers
Information and communications
technology service managers
Professional services managers
Hotel and restaurant managers
Retail and wholesale trade managers
Other services managers
Science and engineering professionals
Physical and earth science professionals
Mathematicians, actuaries and statisticians
Life science professionals
Engineering professionals (excluding
electrotechnology)
Electrotechnology engineers
Architects, planners, surveyors and
designers
Medical doctors
Nursing and midwifery professionals
Traditional
and
complementary
medicine professionals
Veterinarians
Other health professionals
Teaching professionals
University and higher education
teachers
Vocational education teachers
Secondary education teachers
Primary school and early childhood
teachers
Other teaching professionals
Finance professionals
Administration professionals
Sales, marketing and public relations
professionals
Information and communications
technology professionals
Software and applications developers
and analysts
Database and network professionals
Legal professionals
111
112
121
122
131
132
133
134
141
142
143
210
211
212
213
214
215
216
221
222
223
225
226
230
231
232
233
234
235
241
242
243
250
251
252
261
Freq.
123
1,313
1287
ISCO
Title
262
263
264
265
310
Librarians, archivists and curators
Social and religious professionals
Authors, journalists and linguists
Creative and performing artists
Science and engineering associate professionals
Physical and engineering science technicians
Mining, manufacturing and construction supervisors
Process control technicians
Life science technicians and related
associate professionals
Ship and aircraft controllers and technicians
Health associate professionals
Medical and pharmaceutical technicians
Nursing and midwifery associate professionals
Veterinary technicians and assistants
Other health associate professionals
Financial and mathematical associate
professionals
Sales and purchasing agents and brokers
Business services agents
Administrative and specialised secretaries
Regulatory government associate professionals
Legal, social and religious associate
professionals
Sports and fitness workers
Artistic, cultural and culinary associate professionals
Information and communications
technicians
Information and communications
technology operations and user
support technicians
Telecommunications and broadcasting technicians
General office clerks
Secretaries (general)
Keyboard operators
Tellers, money collectors and related
clerks
Client information workers
Numerical clerks
Material-recording and transport
clerks
Other clerical support workers
Travel attendants, conductors and
guides
2057
311
1437
312
16
2547
313
314
456
315
2968
999
2568
301
274
250
320
321
322
324
325
331
57
461
2108
332
333
334
354
1482
335
1313
2182
173
341
342
343
118
1744
4684
774
350
351
2121
1863
14
352
1043
2168
5792
2919
411
412
413
421
1462
2125
422
431
432
2239
1593
441
511
45
Freq.
228
3191
1024
893
355
3145
2064
1098
200
516
57
1528
2692
19
1950
3661
1869
3509
2214
2093
3751
599
768
115
264
206
6525
1847
68
166
3039
1692
3690
3716
421
ISCO
Title
512
513
514
Cooks
Waiters and bartenders
Hairdressers, beauticians and related
workers
Building and housekeeping supervisors
Other personal services workers
Sales workers
Street and market salespersons
Shop salespersons
Cashiers and ticket clerks
Other sales workers
Child care workers and teachers’ aides
Personal care workers in health services
Protective services workers
Market gardeners and crop growers
Animal producers
Mixed crop and animal producers
Forestry and related workers
Fishery workers, hunters and trappers
Building frame and related trades
workers
Building finishers and related trades
workers
Painters, building structure cleaners
and related trades workers
Sheet and structural metal workers,
moulders and welders, and related
workers
Blacksmiths, toolmakers and related
trades workers
Machinery mechanics and repairers
Handicraft workers
Printing trades workers
Electrical equipment installers and repairers
Electronics and telecommunications
installers and repairers
Food processing, wood working, garment and other craft and related
trades workers
Food processing and related trades
workers
Wood treaters, cabinet-makers and
related trades workers
Garment and related trades workers
Other craft and related workers
Mining and mineral processing plant
operators
Metal processing and finishing plant
operators
Chemical and photographic products
plant and machine operators
515
516
520
521
522
523
524
531
532
541
611
612
613
621
622
711
712
713
721
722
723
731
732
741
742
750
751
752
753
754
811
812
813
Freq.
ISCO
Title
938
3491
1739
814
874
816
734
1
337
10421
1502
620
3321
7314
817
Rubber, plastic and paper products
machine operators
Textile, fur and leather products machine operators
Food and related products machine
operators
Wood processing and papermaking
plant operators
Other stationary plant and machine
operators
Assemblers
Locomotive engine drivers and related
workers
Car, van and motorcycle drivers
Heavy truck and bus drivers
Mobile plant operators
Ships’ deck crews and related workers
Domestic, hotel and office cleaners
and helpers
Vehicle, window, laundry and other
hand cleaning workers
Agricultural, forestry and fishery
labourers
Mining and construction labourers
Manufacturing labourers
Transport and storage labourers
Food preparation assistants
Refuse workers
Other elementary workers
815
818
821
831
832
833
834
835
911
2042
2910
1585
323
14
88
4397
912
921
2038
931
932
933
941
961
962
1113
1318
Source: EBB 2010–2012, own calculations
967
3528
179
526
1587
397
236
736
756
501
243
120
78
78
46
Freq.
266
295
421
57
359
411
166
1079
3873
1579
70
5401
327
174
286
1230
3818
812
266
945
D
KHB Models
Tables D1 and D2 present the logistic regressions with KHB technique for both sets of
analyses – school-to-work and life-course.
If a control variable is added in linear regression, the total effect of X on an outcome Y
is composed of the sum of the direct effect of X symbolized by the coefficient of X and the
indirect effect of X via a mediator Z, displayed in the coefficient of Z. This decomposition
of the total effect in direct and indirect effect is not possible in logistic regression models in
which the coefficients depend on the error variance and this error variance differs between
nested models. This varying error variance can be referred to as rescaling. The KHB
technique calculates unbiased comparisons of logit coefficients of the same variable across
Table D1: KHB - School-to-work Transition (DV: Employment Probabilities)
Men
Model 1
Linkage Strength (standardized)
Model 2
0.150***
(0.025)
Vocational (ref=general)
Women
Model 3
Model 1
0.138***
0.164***
(0.033)
(0.025)
Model 2
Model 3
0.152***
(0.029)
0.236***
0.051
0.213***
0.070
(0.050)
(0.069)
(0.047)
(0.057)
0.521***
0.613***
(0.157)
Educational level (ref=pre-primary)
Primary
0.557***
0.463***
0.549***
0.621***
(0.159)
(0.159)
(0.160)
(0.157)
(0.156)
Lower Secondary
1.235***
1.054***
1.202***
1.345***
1.152***
1.302***
(0.145)
(0.148)
(0.153)
(0.143)
(0.145)
(0.148)
1.768***
1.597***
1.729***
2.377***
2.201***
2.322***
(0.144)
(0.148)
(0.152)
(0.141)
(0.145)
(0.148)
2.024***
1.823***
1.977***
2.656***
2.483***
2.591***
(0.233)
(0.238)
(0.240)
(0.202)
(0.206)
(0.208)
2.106***
2.201***
2.113
2.972***
3.096***
2.980***
(0.149)
(0.148)
(0.150)
(0.145)
(0.144)
(0.145)
1.611***
1.901**
1.635***
2.301***
2.721***
2.333***
(0.394)
(0.395)
(0.395)
(0.349)
(0.353)
(0.349)
0.205***
0.206***
0.203***
0.069***
0.070***
0.066***
(0.014)
(0.014)
(0.014)
(0.013)
(0.013)
(0.013)
-0.005***
-0.005***
-0.005***
-0.003***
-0.003***
-0.003***
(0.001)
(0.001)
(0.001)
(0.001)
(0.001)
(0.001)
-1.575***
-1.601***
-1.568***
-1.455***
-1.489***
-1.441***
(0.176)
(0.176)
(0.176)
(0.171)
(0.171)
(0.172)
Upper Secondary
Post Secondary
Tertiary
Doctorate
Age (centred at 16)
Age squared (centred at 16)
Constant
Region fixed effects
yes
yes
yes
yes
yes
yes
Year fixed effects
yes
yes
yes
yes
yes
yes
N
29,982
29,982
29,982
30,070
30,070
30,070
Pseudo2
0.114
0.114
0.114
0.110
0.110
0.110
BIC
991,668
991,668
991,668
1,214,809
1,214,809
1,214,809
Source: EBB 2010–2012, own calculations
Note: Logged odds, standard errors in parentheses. Sampling weights are applied.
Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001
47
nested models, one with and one without a control variable (Karlson et al., 2012). It
de-composites the total effect of a variable into the direct effect due to confounding and
an indirect effect due to re-scaling. It does this by rescaling the coefficients in models with
less predictors by including residuals of the predictors which are missing in the model.
This enhances the comparability of the effect sizes of predictors across nested models.
The coefficients in these KHB models can be compared to the regular logistic regression
tables in the main text. No major differences are found in the size of the coefficients.
Table D2: KHB - Life-course (DV: Employment Probability)
Men
Model 1
Women
Model 2
Model 1
0.079***
0.221***
0.090***
0.267***
(0.013)
(0.026)
(0.011)
(0.024)
Age (centred at 16)
0.212***
0.208***
0.115***
0.113***
(0.003)
(0.003)
(0.003)
(0.003)
Age squared (centered at 16)
-0.005***
-0.005***
-0.003***
-0.003***
(0.000)
(0.000)
(0.000)
(0.000)
Linkage Strength (standardized)
Linkage Strength × Age
Model 2
-0.005***
-0.007***
(0.001)
(0.001)
Educational level (ref=pre-primary)
Primary
Lower Secondary
Upper Secondary
Post Secondary
Tertiary
Doctorate
Constant
0.635***
0.620***
0.900***
0.865***
(0.081)
(0.081)
(0.076)
(0.076)
1.173***
1.183***
1.414***
1.400***
(0.075)
(0.075)
(0.071)
(0.071)
1.577***
1.584***
2.092***
2.099***
(0.074)
(0.074)
(0.071)
(0.071)
1.643***
1.643***
2.299***
2.314***
(0.089)
(0.089)
(0.083)
(0.083)
1.983***
1.996***
2.595***
2.596***
(0.076)
(0.076)
(0.072)
(0.072)
2.478***
2.527***
2.690***
2.661***
(0.181)
(0.182)
(0.212)
(0.211)
-1.560***
-1.513***
-1.675***
-1.644***
(0.088)
(0.097)
(0.097)
(0.088)
Region fixed effects
yes
yes
yes
yes
Year fixed effects
yes
yes
yes
yes
N
107,047
107,047
108,957
108,957
Pseudo R2
0.176
0.176
0.164
0.164
BIC
3,700,321
3,700,321
4,754,605
4,754,605
Source: EBB 2010–2012, own calculations
Note: Logged odds, standard errors in parentheses. Sampling weights are applied.
Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001
48