Economics of Education Review 24 (2005) 161–170
Education returns of wage earners and
self-employed workers: Portugal vs. Spain
Inmaculada Garcı´a-Mainar
a,
, Vı´ctor M. Montuenga-Go´mez
b
a
Departamento de Ana´lisis Econo´mico, Facultad de Ciencias Econo´micas y Empresariales,
University of Zaragoza, C/Gran Vı´a no 2, 50005 Zaragoza, Espan˜a, Spain
b
University of La Rioja, Departamento de Economı´a y Empresa, Edificio Quintiliano, C/La Cigu¨en˜a no 60,
26004 Logron˜o (La Rioja), Espan˜a, Spain
Received 12 November 2003; accepted 21 May2004
Abstract
This paper investigates the returns to education in two Southern EU countries, Portugal and Spain, both
characterized bya relativelyhigh self-employment rate. The impact of education on both wage earners and the self-
employed is analyzed by using a comparable data set coming from the European Community Household Panel during
the period 1994–2000. To correct for the abilitybias and recover the education coefficients, an Efficient Generalized
Instrumental Variable technique is applied. Results show that returns are different across the two countries as well as
across the two employment statuses.
r
2004 Elsevier Ltd. All rights reserved.
JEL classification: C13; C23; I21; J31
Keywords: Educational economics; Human capital; Rate of return
1. Introduction
Estimating the returns to education has been one of
the targets of labor economists for decades. Habitually,
cross-sectional information has been used to appraise
the causal effect of schooling on earnings for different
countries (see
). Although most of the earlier studies on
the returns to education applied OLS estimators, the
most recent suggest the use of an instrumental variable
estimation to allow for the endogeneitybias arising in
this type of research (see
, for recent
overviews). The analysis of the returns to education has
been also carried out sorting for several demographic
groups (e.g. male vs. female; white vs. non-white;
immigrants vs. natives; people working for the public
sector vs. for the private sector, etc.), often finding a
differing behavior between the diverse groups under
consideration (see, for example,
).
In this article, we aim to estimate the returns to
education in two Southern EU countries, namely
Portugal and Spain, following the conventional
approach and focusing on the differences between
the self-employed and wage-earner workers. It is often
the case that self-employed workers are excluded from
labor market studies because of difficulties in measuring
and interpreting their earnings. This has meant that only
a few papers have considered the returns to education in
ARTICLE IN PRESS
www.elsevier.com/locate/econedurev
0272-7757/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.econedurev.2004.07.004
Corresponding author. Tel.: +34-976762635; fax: +34-
976761996
E-mail address: igarcia@posta.unizar.es (I. Garcı´a-Mainar).
a waythat distinguishes between the self-employed and
wage earners. Nevertheless, some examples can be cited:
and
for Spain,
and
for the UK,
and
for the US. The evidence, although mixed
across countries and always obtained using cross-
sectional information, tends to show that the self-
employed enjoy higher returns to education than their
wage-earning counterparts.
This paper presents two clear advantages over
previous literature: the data set and the econometric
technique used in the estimation. As regards the
former, data comes from the European Community
Household Panel (ECHP) from 1994 through 2000.
This type of data is superior to cross-section informa-
tion, since panel data is a useful tool when dealing
with the endogeneitybias of education and allowing
for unobserved individual heterogeneity. Moreover,
the information provided bythe ECHP is homo-
geneous across the two countries, given that the
questionnaire is the same and the elaboration process
is coordinated bythe European Statistics Agency
(Eurostat). Despite the existence of various Spanish
and Portuguese studies on the returns to education,
onlythe two mentioned earlier that focus on the case
of Spain distinguish between the self-employed and
wage earners.
1
However, neither of them uses panel
data. Our studyattempts to fill this gap, and this
represents the second advantage of our paper. Specifi-
cally, a modified random effects-type model (the
Hausman–Taylor method) is used to estimate the
returns to education, which allows the possibilityof
controlling for the endogeneitybias and, simulta-
neously, identifying the estimates for the time-invariant
regressors, such as education.
The rest of the paper is organized as follows. In the
next section, the issue of returns to education is
discussed from a theoretical viewpoint. In Section 3,
we describe the empirical model and the estimation
procedure. The database is described in Section 4 with
the employment status and some information about the
education attainments of the workers in these two
countries being also presented. Section 5 offers the
estimated results of the rates of return for the countries
in question. Finally, Section 6 summarizes the most
important conclusions.
2. Education, wage-earners and the self-employed
Our main objective in this paper is to provide an
answer to the following question: does the same level of
education attained byan individual in Portugal and in
Spain implythe same rate of education return for the
self-employed as for the wage earner in these same two
countries? If so, one student acquiring higher education
would obtain similar profitability, other labor charac-
teristics being the same, in each of the two types of
employment. Otherwise, the student will try to engage
him/herself to the job scheme where returns are higher.
2
Essentially, two streams of research have been
followed to justifythe dependence of earnings on
education. First, the human capital theory(
), which argues that education and experience
augments natural abilities that are subsequentlysold in
the labor market. That is to say, education is seen as an
investment in oneself during school and experience as
one made later through on-the-job training. Accord-
ingly, both types of workers, self-employed and wage
earners, would be rewarded for achieving more educa-
tion. In a competitive labor market, returns to education
should be the same for both types. However, in a non-
competitive environment there are some reasons that
mayexplain the existence of different rates of return.
Thus, a significant part of the returns to the self-
employed is not only due to formal education, but rather
to some other abilities (motivation, salesmanship, etc.),
that lead to lower returns to education among this
group. On the other hand, for the case of wage earners,
some of the return on their productivitymaybe
appropriated bythe firm, and thus estimates of
the return to education among the self-employed
should be higher.
Secondly, sorting models also predict higher earnings,
since greater human capital is acquired in order to signal
for higher productivity(
). More specifi-
cally, firms infer productivity from education and
students choose the education level to signal their
productivityto potential employ
ers. Thus, higher
productivity, leading to higher earnings, is not due to
higher education; instead, higher earnings are obtained
bythose who have been able to signal that theyare more
productive byvirtue of the education level theyhave
attained. According to this view, the signaling role of
educational attainment for earnings is less evident in the
case of the self-employed because no employer exists.
One exception would perhaps be the case of certain
professionals, since greater education would signal more
guarantee in the qualityof service theyoffer.
ARTICLE IN PRESS
1
Previous studies of the return to education in Spain are
Barceinas, Oliver, Raymond, and Roig (2001)
and
Garcia, Herna´ndez, and Lo´pez (2001)
. For the case of
Portugal, see
2
In this respect,
find evidence
supporting the idea that the earnings expectations of students
are remarkablyclose to the realizations observed.
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
162
On balance, and given that the human capital and
signaling explanations are not mutuallyexclusive,
nothing can be said in advance about whether returns
to education for the self-employed will be higher or
lower than for wage earners. Similarly, theoretical and
empirical work provides support in favor of higher
returns to experience for both self-employed and wage
earners. The models that predict higher returns for wage
earners are those concerning risk and asymmetric
information aspects. Specifically, wage earners are paid
more the longer theyare in the job to avoid their
incentives to shirk and quit. As the self-employed do not
have anyincentives to do that, returns are expected to be
lower for this group of workers. Bycontrast, the
investment model justifies the view that self-employed
mayobtain higher returns because physical and human
capital investments are not shared with an employer.
For a more detailed exposition of these results, see
and
The analysis of returns distinguishing between self-
employed and wage earners has a different impact
depending upon the degree of presence of self-employ-
ment activities in the labor market. In the European
Union, the self-employment rates (the percentage of self-
employed workers over total employed) have been
decreasing during the last two decades, as shown in
. Marked differences are observed across
countries, with figures above 20% in the Southern
countries, and below 10% in countries such as Germany
or Denmark. In Spain and Portugal, rates are among the
highest, with values close to 20% in 2000. As a result of
this, self-employment has a certain interest both in
Portugal and in Spain because it represents a clear
alternative to paid employment.
3. Empirical model and estimation
This section starts with the discussion of the empirical
model. The specification proposed by
is
followed to obtain the returns to education for both
types of workers in the two countries. Next, the
estimation strategyis presented and the
procedure is described.
3.1. Empirical model
We have chosen a conventional earnings function
approach to determine the returns to education. More
specifically, the log-linear wage equation proposed by
has been extensivelyapplied and will also
be used in this work. Although this specification has
been based on human capital theory(see
), it
can also be interpreted as stemming from signaling-type
models. The earnings reward for more education can
therefore be seen as the combined effect of human
capital accumulation and the effect of being identified as
a graduate rather than as a dropout.
Thus, the estimated model takes the following form:
ln w
it
¼
a
i
þ
b Edu
i
þ
m
1
Exp
it
þ
m
2
Exp
2
it
=100
þ
X
it
d þ Z
i
g þ v
it
;
(1)
where i and t stand for N individuals and T time periods,
respectively. w denotes earnings, Edu, the education
variable/s (considered time-invariant), Exp experience
(time-varying), X a set of time-varying regressors and Z
a set of time-invariant regressors.
The dependent variable is the natural log of net
earnings, where net earnings are defined as gross
earnings less tax. The education variables are of two
types, which will be used alternatively. The habitual
measure used is the years of schooling, represented by
Edu, which is a continuous variable. However, measur-
ing education in such a waypresents some flaws. First, it
restricts the annual marginal effect of schooling to being
the same regardless of education level. Secondly, if
information is provided bythe number of y
ears
effectivelystudied, it attributes more education to
students who failed and repeated some year than to
the same individual who passed the year successfully.
Alternatively, if information is provided by the level of
education attained, then the assignment of years to each
education categorymaylead to measurement errors in
the years of schooling. Thirdly, if we believe that sorting
models have some relevance in inducing individuals to
acquire formal education, as is verylikelyin the case of
Spain and Portugal (where the salaryprofiles are largely
ARTICLE IN PRESS
Table 1
Self-employment rates in EU member states 1987–2000
1987
1995
2000
Austria
—
10.8
10.5
Belgium
15.3
15.4
13.6
Denmark
9.2
8.4
8.0
Finland
—
14.3
12.6
France
12.7
11.6
10.0
Germany9.1
9.4
9.7
Greece
35.4
33.8
31.3
Ireland
21.8
20.8
16.5
Italy24.4
24.5
23.6
Luxembourg
9.2
10.0
8.7
Netherlands
10.1
11.5
10.0
Portugal
27.2
25.8
20.2
Spain
23.5
21.8
18.0
Sweden
—
11.3
9.8
United Kingdom
12.5
13.0
10.9
EU
15.9
15.0
13.6
Note: Percentage of self-employed workers over total of
employed.
Source: Eurostat Labour Force Survey.
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
163
linked to the education categoryof the worker), then the
level obtained bythe individual is more informative than
the number of years needed to attain it. This is the
reason why, as an alternative to years of schooling, we
also consider educational attainment as a measure of
schooling. Thus, schooling is identified with the level
achieved bya student no matter how manyyears it took
him/her to attain it. Educational attainment therefore
represents the last completed type of schooling and is
classified into three levels (primary, secondary and
higher), which are represented bydummyvariables.
The primarycategoryincludes elementaryand below
elementaryschool; secondaryincludes vocational train-
ing and secondaryschool, whereas higher includes
university(either in short or long cycles i.e. university
diploma or degree). In our estimation, the reference
categoryis primarylevel (not included to avoid
collinearityproblems); Edu2 takes the value 1 when
the individual has attained a secondaryeducation level,
and 0 otherwise; whilst Edu3 takes the value 1 when
he/she has achieved a higher level, and 0 otherwise.
3
These variables are all time-constant.
The experience profile is expressed bythe number of
years worked byan individual, and byits squared value
divided by100. Specifically, experience is measured as
the difference between the current age and the age of
starting work. The coefficient associated to the squared
value is expected to be negative to allow for the
decreasing returns, therebyreflecting the fact that
earnings mayrise due to experience, but at a diminishing
rate. Our database provides us with the information
about the moment in which the individual started to
work. Measuring experience as the difference between
the current age and this information is more appropriate
than the commonlyused age-education-6, because it
reduces the problem associated with measurement
errors. This is of especial relevance in the case of Spain
where people take some time to find their first job.
However, each of these two measures consider experi-
ence as endogenous which represents a potential source
of bias. Manyauthors solved this problem byusing the
age variable as a proxyfor experience because it is an
exogenous variable, but this implies a different inter-
pretation of the coefficient expressing the rate of return
(see
Barceinas, Oliver, Raymond, & Roig, 2001
). The
Hausman–Taylor procedure that is applied in our paper
enables us to consider the possibilityof experience being
treated as an endogenous variable, and so that measure
is the best choice for our study.
The rest of the independent variables, represented in
Eq. (1) by X and Z, contain a set of dummyvariables
indicating gender, marital status, seniority, occupation,
sector of activity, time and whether the worker has taken
some training course. Finally, the a’s represent the
individual-specific effects, which allow us to control for
the unobserved heterogeneity.
3.2. Estimation strategy
The estimation of the Mincerian earnings function to
determine the rate of returns to education is not
problem-free. The presence of measurement errors and
unobserved variables, such as ability, motivation, etc.,
that maybe correlated with schooling, bias OLS
estimates. Specifically, it has been shown that measure-
ment error biases downwards the OLS estimates,
although recent evidence (see
) onlyattributes
at most a 10% gap to this source of bias. Bycontrast,
since schooling and the unobserved abilitymaybe
positivelycorrelated, omitting measures of abilityresults
in the schooling coefficient being biased upwards. The
latest studies seem to coincide in finding that this source
of bias prevails over the measurement-error bias,
resulting in an overall upwards bias of the OLS
estimates (see, for example,
).
Consequently, some effort must be made to alleviate
this abilitybias as much as possible. When a direct
indication of ability, such as IQ tests or information
from twins or siblings, is not available, the most
appropriate approach is to select an instrumental
variable estimator. In this way, schooling is instrumen-
ted byvariables that are correlated with education, but
not with the residuals. Various instruments have been
widelyproposed in the literature. Typical examples are
those known as natural experiments, which include
school reforms and features of the school system, the
proximityof the school to the place of residence (and
other supplyside instruments capturing features of the
education system) and the season of birth of the
individual. Other possibilities include the familyback-
ground and the absolute degree of risk aversion.
Excellent surveys on this can be found in
When using anyof the above-mentioned instruments
(or a combination of them) a common result emerges,
namelythat IV estimates are of a higher order than
OLS, which is somewhat paradoxical, given that OLS
estimates have alreadybeen found to be biased upwards.
This would implythat IV estimates are even further
biased upwards.
provides a thorough
discussion on this result.
Instrumental variable estimation is, however, some-
times difficult to implement. Some problems that
appears are, firstly, the number of possibilities for
choosing instruments is verywide and the estimates
coming from the use of these alternative instruments are
ARTICLE IN PRESS
3
In this case, the fragment ‘‘b Edu
i
’’ in Eq. (1) would be
represented by‘‘b
1
Edu2
i
+b
2
Edu3
i
’’ such that Edu is a
continuous variable, whereas Edu2 and Edu3 are dummy
variables.
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
164
quite different. Secondly, the use of instruments that are
weaklycorrelated with the endogenous variables may
produce biased estimates even with large samples (see
).
All this generates uncertaintyin the selection of
instruments, which leads us to consider other options.
In this regard, the availabilityof panel data opens up the
number of possibilities for dealing with endogeneity.
4
Individual unobserved heterogeneitymaybe eliminated
bymean or time-differencing, i.e. byusing some kind of
within or first-differencing estimators, which eliminate
the correlation between education and unobserved
heterogeneity. However, when operating in this way,
the coefficients of the time constant variables (such as
the level of education) cannot be estimated, because they
disappear when mean or time-differences are con-
structed. The typical alternative to this method is to
consider random effects. However, this approach is not
valid either, because it assumes that no correlation exists
between the regressors and the individual effects, and
this hypothesis is invalidated by the presence of the
abilitybias.
The literature, therefore, has suggested an alternative
procedure, the Efficient Generalized Instrumental Vari-
ables (EGIV) proposed by
This procedure allows us simultaneouslyto control for
the correlation between regressors and unobserved
individual effects; to identifythe estimates of the time-
invariant covariates, such as education; and, finally, to
avoid the insecurityassociated with the choice of
suitable instruments, since the individual means over
time of all the included regressors can serve as valid
instruments.
Additionally,
the
variance-covariance
structure can be taken into account so as to obtain
more efficient estimators. In short, in the case under
consideration, education is an endogenous, time-invar-
iant regressor, whereas the experience variables are also
endogenous, but time-varying. Since we are interested in
the education coefficients, all the exogenous variables
(either time-invariant or time-varying) plus the indivi-
dual means over time of all the time-varying regressors
can be used as instruments to obtain consistent
and rather efficient estimates of the returns to education.
The recent work by
provides information on the suitabilityof the
Hausman–Taylor procedure in a general framework
where panel data is available and some regressors are
correlated with the individual effects.
The
model can be
represented in its most general form as follows:
ln w
it
¼
a
i
þ
X
0
it
d þ Z
0
i
g þ v
it
;
ð
2Þ
where i ¼ 1;
. . . ; N and t ¼ 1; . . . ; T. The Z
i
are indivi-
dual time-invariant regressors, whereas the X
it
are time-
varying. a
i
is assumed to be iid(0, s
2
a
) and v
it
iid(0, s
2
v
),
both independent of each other and among themselves.
The matrices X and Z can be split into two sets of
variables X ¼ ½X
1
; X
2
and Z ¼ ½Z
1
; Z
2
such that X
1
is
NT k
1
, X
2
is NT k
2
, Z
1
is NT g
1
, and Z
2
is NT
g
2
. The X
1
and Z
1
are assumed exogenous and not
correlated with a
i
and v
it
, while X
2
and Z
2
are
endogenous due to their correlation with a
i
but not
with v
it
. Consequently, the education variable is included
in Z
2
, since it is time-invariant and endogenous. For its
part, experience is a time-varying, endogenous variable
that is included in X
2
.
suggest an instrumental
variables estimator which premultiplies expression (2) by
O
1/2
where O is the variance-covariance term of the
error component a
i
+v
it
,
5
and then performs 2SLS using
as instruments [Q, X
1
, Z
1
]. Q is the within transforma-
tion matrix with X
¼
QX having a typical element
X
n
it
¼
X
it
X
i
and X
i
the individual mean. As
argue, this is equivalent to running 2SLS with
½
X
n
; X
1
; Z
1
as the set of instruments. If the model is
identified, in the sense that there are at least as many
time-varying exogenous regressors X
1
as there are
individual time-invariant endogenous regressors Z
2
, i.e.
k
1
X
g
2
, then this Hausman–Taylor estimator is more
efficient than fixed effects. In our model, the onlytime-
invariant endogenous variable is education, whereas
there are several time-varying exogenous regressors. In
our study, we carry out Hausman-based specification
tests (
) to choose the more appropriate
estimator, as discussed below. For more details see
and
. Some examples of this type of
estimation are
and
. The 8.0 version of Stata includes the Hausman–-
Taylor procedure and is used to obtain the estimates
presented in this paper.
4. Data
The data set we use in this paper is obtained from the
ECHP for the period 1994–2000. This is a unique data
set in the sense that it provides homogeneous informa-
tion for both countries through several years. This
represents a great advantage in estimating returns to
education and making comparisons across the countries.
We select the workers that have provided information
for all the variables used and distinguish between the
ARTICLE IN PRESS
4
Some examples of studies on earnings using panel data are
and
5
The variance-covariance structure of the system can then be
represented by EðUU
0
Þ ¼
s
2
a
ð
i
T
i
0
T
I
N
Þ þ
s
2
v
ð
I
T
I
N
Þ
, where
i
T
is a Tx1 vector containing ones and I
N
(I
T
) is the identity
matrix of rank N (T) and U is an NTx1 vector of disturbances.
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
165
self-employed and wage earners. For the case of Spain,
the samples contain 6652 observations corresponding to
2240 self-employed and 28,651 observations correspond-
ing to 8365 wage earners, whereas for Portugal, the
samples contain 7181 observations corresponding to
2225 self-employed and 28,914 observations correspond-
ing to 6968 wage earners.
6
Some descriptive statistics are
shown in
, in particular the mean values of
earnings (in logs and per hour), experience and
education variables for the period 1994–2000. In Spain
and Portugal the earnings and education are higher for
wage earners than for the self-employed, whilst the
former have fewer years of experience. Comparing both
countries, we can see that earnings are higher in Spain
than in Portugal, as are the years of schooling and the
proportion of workers that have obtained more ad-
vanced education levels, whereas workers enjoymore
experience in Portugal. Globally, this suggests that the
returns to experience maybe of a lower order than the
returns to education. In what follows, we look more
carefullyat some features of these variables.
As regards earnings, the fact that the self-employed
earn less than wage earners, in average terms, deserves
some comments. Self-employed are a heterogeneous
group including a wide range from low-skilled activities
to verysuccessful entrepreneurs, ‘‘superstars’’. This
causes earning dispersion among self-employed to be
higher than among wage earners. Additionally, it must
be borne in mind that we are onlyconsidering the
pecuniaryaspects.
and certain
others present evidence supporting the view that the
self-employed are more satisfied with the employment
situation than wage earners. This points to non-
pecuniaryaspects being veryrelevant in the decision of
workers regarding their employment status. Assessing
both these aspects, pecuniaryand non-pecuniary
, we
would probablyobserve the job valuation of self-
employment to be higher than that of wage employment.
However, this would require working with subjective
information, and that is not the aim of this paper.
With respect to experience, it should be noted that
previous years worked may have a different considera-
tion depending on the employment status. Thus,
promotion in the wage employment sector may mean a
sharp change in the earnings, reflecting in just one year
the accumulated returns of previous experience, with
this possibilitybeing not an option in the case of the self-
employed. Besides, some studies (
) have found that the experience
returns depends on whether the worker keeps working in
the same sector or switches to the other. Nevertheless,
since the focus of this studyis to analyze the returns to
education (for both groups of workers in the two
countries) experience is considered equallyin the two
employment sectors regardless of which sector the
worker was previouslyenrolled in.
Looking at the educational structure of both coun-
tries, we can observe certain characteristics. In Portugal,
only10% of the self-employed have an education level
higher than elementary, with this proportion rising to
20% for wage earners. In Spain, two thirds of self-
employed have only attained a primary level of
education, with the other third being equallydistributed
among those that have obtained a secondaryor a higher
level of education. As regards wage earners, the
proportion of primaryeducation in Spain is ‘‘only
’’
one half, with almost one third holding a university
degree. All this suggests that most of the workers in
these two countries suffer from low levels of education,
with this proportion being clearlyhigher among the self-
employed. In other words, graduate workers prefer to be
salaried.
ARTICLE IN PRESS
Table 2
Mean values of earnings, experience and years of schooling and percentage of workers in the three educational levels (1994–2000)
Spain
Portugal
Self-employed
Wage earners
Self-employed
Wage earners
Earnings/hour
a
4.51 (4.70)
5.59 (3.63)
2.27 (2.69)
3.32 (2.77)
Experience
26.44 (13.71)
19.53 (12.48)
31.96 (15.89)
19.98 (13.56)
Education
Primary
b
65.6
50.3
90.2
79.1
Secondary
b
16.9
20.2
6.1
13.3
Higher
b
17.5
29.5
3.7
7.6
Years of schooling
11.88 (6.78)
13.46 (7.00)
9.67 (7.01)
10.92 (6.84)
Notes: Standard deviation between parentheses.
a
Expressed in euros.
b
Proportion of the population in the respective group in the countrythat have attained this educational level.
6
When the education is measured byyears of schooling the
samples are a bit smaller because there is some missing
information in the dataset for this variable.
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
166
In summary, given that the earnings, experience and
educational attainment structures as well as the job
valuation and the role of experience are quite different
between the self-employed and wage earners, this
influences our investigation as to whether the returns
to education differ between these two categories of
workers in the two countries in question.
To conclude with the data analysis, two final
considerations are worthyof note with regard to the
estimation strategyset up in Section 3. Firstly, Mincer’s
equation is typically estimated for sub-populations with
given characteristics (say, as in our own case, wage
earners or self-employed). As a consequence, the sample
maybe not representative of the whole population. In
this paper, we do not correct explicitlyfor this selection
bias. Instead, and so as to alleviate it as much as
possible, we have included a large set of regressors in the
wage equation in order to control for several factors that
might influence the workers’ decisions (see,
;
).
Secondly,
the
attrition
problem
characterizing
panel data information has special relevance in our
study. The attrition rates in our samples for Spain
and Portugal are, respectively, 58% and 54% for the
self-employed and 51% and 41% for wage earners.
These values are quite high in comparison with other
surveys and probably attributable to two factors.
Firstly, attrition arises from the fact that some
information of individuals is not available for all
the periods or they leave the survey. Secondly, in
our particular case, attrition can be derived from
the fact that individuals mayswitch their employ
-
ment situation during the period, that is to say, from
their initial situation (wage earner or self-employed)
to the other employment situation, to unemployment
or to inactivity. Considering that the first source of
attrition is random and dominates, the bias influence
is much less than that inferred from the global attrition
rates. For its part, the procedures habituallysuggested
in the literature to correct for the bias arising
from attrition (see
, for a summary),
applyfirst differences to remove unobserved hetero-
geneity. This hinders estimation of the coefficients
of time-invariant regressors, which constitute the
main target of our study, and thus they are not
considered.
5. Results
In this section, we present two sets of results. The first
one is obtained byapplying a random effects estimator,
which while not providing consistent estimates for
education, is the most common method used, and
enables comparison with previous studies. Secondly,
we consider the possibilityof endogeneityin some of the
regressors and thus applythe Hausman–Taylor method,
checking for such endogeneitywith a pair of Hausman
tests.
The results of the random effects estimation are
shown in
. Here, we onlydisplaythe estimated
coefficients corresponding to the variables of interest.
We can observe that all the coefficients are significant at
the 5% level and have the right signs. With respect to the
experience variables, on-the-job training increases hu-
man capital accumulation along the life cycle, as
expected, attaining the maximum return when the
worker has around 30 years of experience. In Portugal,
returns are somewhat lower with the maximum attained
just before reaching 30 years of experience, whereas in
Spain this maximum is attained once this time has been
exceeded.
7
With respect to the education variables, we first
analyze the dummies for secondary and higher educa-
tion and then for years of schooling. The higher
education coefficients are larger than those correspond-
ing to the secondarylevel, indicating that returns to
education in both levels would appear to exist. In Spain,
the returns are higher for wage earners than for the self-
employed, whilst in Portugal returns are higher for wage
earners with higher education but higher for the self-
employed with a secondary education level. When
comparing both countries, we find that returns are
higher in Portugal. When education is measured by
years of schooling, no differences exist between the self-
employed and wage earners in Spain, even though
the rate of return is reduced (1.3%). Bycontrast, in
Portugal, the returns to education of wage earners
is more than three times that of the self-employed, but
still verylow (2.4%). In both countries, these values are
considerablylower than those reported in previous
studies using cross-sectional information (between 5 and
8%; see, for instance,
shows the results of the EGIV estimation,
which considers the endogeneityof experience and
education (and some other variables). We have pre-
viouslyapplied Hausman tests to investigate the
endogeneityof these variables. Following
, a first Hausman test is the standard one
to distinguish between the random and fixed effects
estimators. In all cases, the random effects hypothesis is
rejected in favor of the fixed effect estimator (see row H1
in
). A second Hausman test contrasts the
Hausman–Taylor against the fixed effects model.
Although the fixed effects estimator is not an option in
our study, since it does not allow the estimation of the
coefficients of the time invariant regressors, it is useful in
order to test the strict exogeneityof the regressors that
ARTICLE IN PRESS
7
The point where experience stops adding positivelyto
earnings is defined by q ln w=qExp ¼ 0, from the earnings
function (1). This is equal to m
1
þ
m
2
Exp=50 ¼ 0; m
2
o0.
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
167
are used as instruments in the Hausman–Taylor estima-
tion. Thus, when strict exogeneityfor a set of regressors
is rejected, others must be considered in the estimation
to act as instruments. Once the second Hausman test has
identified which are the regressors that are strictly
exogenous, theyare used as instruments in the
Hausman–Taylor estimation, (see row H2).
Comparing the coefficients of
with the
corresponding ones set out in
, we can observe
that in the Hausman–Taylor estimation all coefficients
are higher in absolute values in both countries,
confirming the habitual finding. Focusing now on the
results presented in
, when using anyof the
education variables, whether qualification levels or years
of schooling, we find that returns to education are higher
in Portugal than in Spain. This maybe due to the fact
that the average level of education in Portugal is lower,
which leads to education being considered as a scarce
ARTICLE IN PRESS
Table 4
Estimated coefficients of the Mincerian earnings function byEGIV (Hausman–Taylor)
Spain
Portugal
Self-employed
Wage earners
Self-employed
Wage earners
Experience
0.075**
0.066**
0.063**
0.063**
0.064**
0.061**
0.066**
0.073**
(4.93)
(4.25)
(16.96)
(17.28)
(4.31)
(3.78)
(17.83)
(28.04)
Experience
2
/100
0.127**
0.117**
0.139**
0.142**
0.118**
0.119**
0.103**
0.117**
(4.98)
(4.40)
(24.85)
(23.81)
(5.43)
(4.54)
(24.81)
(23.80)
Secondaryeducation
1.023*
1.694
3.517**
0.420
(2.45)
(1.89)
(5.23)
(0.70)
Higher education
0.696**
1.019**
0.762*
2.201**
(3.63)
(4.29)
(2.00)
(4.46)
Years of schooling
0.049**
0.088**
0.100**
0.095**
(2.86)
(10.10)
(4.14)
(18.23)
Number of observations
a
6652
6280
28651
27481
7181
6000
28914
26065
Hausman test. H1
60.61
58.10
931.17
1147.78
187.69
157.03
1585.12
1690.71
(p-value)
(0.0001)
(0.0002)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
Hausman test. H2
26.01
29.44
30.43
26.21
22.93
34.19
21.94
33.06
(p-value)
(0.4069)
(0.2459)
(0.2088)
(0.3964)
(0.5818)
(0.1038)
(0.6394)
(0.1296)
Notes: t-ratio between parentheses. * indicates that the coefficient is significant at 5%. **indicates that the coefficient is significant
at 1%.
a
Samples when the years of schooling are used are shorter because of missing data for this variable. H1: This tests the random effects
estimator against the fixed effects. H2: This tests the Hausman–Taylor estimator against the fixed effects.
Table 3
Estimated coefficients of the Mincerian earnings function byOLS (random effects)
Spain
Portugal
Self-employed
Wage earners
Self-employed
Wage earners
Experience
0.051**
0.047**
0.053**
0.051**
0.033**
0.031**
0.044**
0.041**
(9.72)
(8.43)
(35.23)
(32.94)
(6.74)
(5.59)
(33.32)
(29.38)
Experience
2
/100
0.079**
0.070**
0.077**
0.079**
0.061**
0.056**
0.072**
0.069**
(8.74)
(7.34)
(26.39)
(25.38)
(8.19)
(6.23)
(29.26)
(25.45)
Secondaryeducation
0.121*
0.216**
0.337**
0.303**
(2.04)
(12.36)
(3.67)
(14.78)
Higher education
0.393**
0.403**
0.599**
0.786**
(5.87)
(21.99)
(4.31)
(27.97)
Years of schooling
0.013**
0.013**
0.007*
0.024**
(3.84)
(13.04)
(2.00)
(21.99)
Number of observations
a
6652
6280
28651
27481
7181
6000
28914
26065
Notes: t-ratio between parentheses. * indicates that the coefficient is significant at 5%. **indicates that the coefficient is significant
at 1%.
a
Samples when the years of schooling are used are shorter because of missing data for this variable.
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
168
resource, and thus more valued. However, this is not the
case of the returns to experience, for which steeper
profiles are observed in Spain for the self-employed, but
not for the wage earners. In both countries, it appears
that when the worker is self-employed, secondary
education produces more returns than higher education.
Particularly, secondary education does not yield any
return to wage earners and onlyhigher education
enables higher earnings to be obtained. Additionally,
secondaryeducation produces higher returns for the
self-employed than for wage earners, whereas the
opposite applies to higher education. These results tend
to reject the linearityin the returns for both groups of
workers, which underlies the years of schooling
approach.
When using the years of schooling, it emerges that
returns to education are higher in Spain for wage
earners, whereas small differences are observed in
Portugal. This result could explain why, despite the fact
that Portugal and Spain both exhibit high self-employ-
ment rates within the EU, these are falling over time, as
shown in
. In the case of Spain, our results show
a greater effect than in those of previous studies, which
tends to confirm the idea that returns to education have
kept increasing throughout the 1990s (see
). The evidence presented here would appear to
support the view that signaling theoryis relevant in
determining individual earnings in Spain, because in
those circumstances where signaling is expected to playa
less important role, i.e. in that of the self-employed,
returns to education are lower when using years of
schooling. In Portugal, however, human capital theory
seems to be prominent since returns are roughlythe
same for both groups of workers.
6. Conclusions
In this paper we have estimated the rates of returns to
education in Portugal and Spain, distinguishing between
the self-employed and wage earners. These two countries
present relativelyhigh self-employment rates within the
EU, and their educational attainment structure is quite
different between the two groups of workers. Informa-
tion from the ECHP for the period 1994–2000 has been
used, with the advantage that this is homogeneous
across both countries. Panel data availabilityhas
allowed us to applyan estimator that provides
consistent estimates of the rates of returns to education.
Education has been represented alternativelybya time
continuous variable, years of schooling, and by dummies
of qualification levels, primary, secondary and higher, to
reduce measurement errors and to check the linearity
assumption.
We have presented two sets of results. First, the
Mincerian earnings equation for both countries and for
the self-employed and wage earners has been estimated
without considering the possible endogeneityof educa-
tion and experience variables. That is to say, we have
carried out a random effects estimation. Secondly, we
have considered endogeneityon these variables by
applying an EGIV estimator (the Hausman–Taylor
procedure). Previously, we have carried out a pair of
Hausman tests to check the validityof our approach and
the choice of the exogenous regressors used as instru-
ments, confirming the suitabilityof this procedure. This
method is veryuseful because it allows us simulta-
neouslyto control for the correlation between regressors
and unobserved individual effects, identifythe estimates
of the time-invariant covariates, and avoid the insecurity
associated with the choice of the instruments.
As regards the most noteworthyresults, we have
found that returns to education are higher in Portugal,
which maybe interpreted as an indication that this less
abundant asset is more valued. In both countries, when
using qualification categories as a measure of education,
it appears that the secondaryeducation level is the best
choice for the self-employed, rather than being a
universitygraduate, as opposed to the case of the wage
earners and what theorypredicts. When education is
measured byyears of schooling, in Spain wage earners
perceive greater returns to education than the self-
employed, with negligible differences in the case of
Portugal. Finally, signaling premises seem to play a
prominent role in determining individual earnings, but
onlyfor the higher education level, even though human
capital influences cannot be excluded in both countries.
To conclude, the answer to the question posed in
Section 2 must be negative; that is to say, we cannot
state that the same level of education attained byan
individual in Portugal and in Spain implies the same rate
of education return for the self-employed as for the wage
earner in these two countries.
Acknowledgements
The authors wish to acknowledge the financial
support provided bythe Spanish Ministryof Science
and Technology, Project SEC2002-01350. Similarly,
their thanks go to an anonymous referee for the many
useful comments, guidance and advice given on an
earlier version of this paper.
References
Alba-Ramirez, A. (1994). Self-employment in the midst of
unemployment: The case of Spain and the United States.
Applied Economics, 26, 189–204.
ARTICLE IN PRESS
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
169
Alba-Ramirez, A., & San Segundo, M. J. (1995). The returns to
education in Spain. Economics of Education Review, 14,
155–166.
Baltagi, B. H., Bresson, G., & Pirotte, A. (2003). Fixed effects,
random effects or Hausman–Taylor? A pretest estimator.
Economics Letters, 79, 361–369.
Barceinas, F., Oliver, J., Raymond, J. L., & Roig, J. L. (2001).
Spain. In C. Harmon, I. Walker, & I. Westergrad-Nielsen
(Eds.). Education and earnings in Europe: A cross country
analysis of the return to education. Cheltenham, UK:
Edward Elgar.
Becker, G. S. (1964). Human Capital. New York: National
Bureau of Economics Research.
Blanchflower, D. (2000). Self-employment in OECD countries.
Labour Economics, 7, 471–505.
Bound, J., Jaeger, D., & Baker, R. (1995). Problems with
instrumental variables estimation when the correlation
between the instruments and endogenous explanatory
variable is weak. Journal of the American Statistical
Association, 90, 443–450.
Brunello, G. (2002). Absolute risk aversion and the returns to
education. Economics of Education Review, 21, 635–640.
Card, D. (1994). Earnings, schooling and ability revisited.
Working Paper No. 4832. Cambridge, MA: National Bureau
of Economic Research.
Card, D. (1999). The causal effect of education on earnings. In
O. Ashenfelter, & D. Card (Eds.). Handbook of Labour
Economics, Vol. 3B. Amsterdam: Elsevier.
Card, D. (2001). Estimating the returns to schooling: Progress
on some persistent econometric problems. Econometrica,
69(5), 1127–1160.
Evans, D. S., & Leighton, L. S. (1989). Some empirical aspects
of entrepreneurship. American Economic Review, 79(3),
519–535.
Fairlie, R. W., & Meyer, B. D. (1996). Ethnic and racial self-
employment differences and possible explanations. Journal
of Human Resources, 31, 757–793.
Garcia, J., Herna´ndez, P. J., & Lo´pez, A. (2001). How wide
is the gap? An investigation of gender wages differences
using
quantile
regression.
Empirical
Economics,
26,
149–168.
Hamilton, B. H. (2000). Does entrepreneurship pay? An
empirical analysis of the return to self-employment. Journal
of Political Economy, 108(3), 604–631.
Hansen, J., & Wahlberg, R. (1997). The return to education and
the gender wage gap: Results from Swedish panel data.
Working Paper No. 2297, Centre de recherche et de´vel-
oppement en e´conomique, Universityof Montreal.
Hausman, J. A. (1978). Specification Tests in Econometrics.
Econometrica, 50, 749–759.
Hausman, J. A., & Taylor, W. E. (1981). Panel data
and unobservable individual effects. Econometrica, 49,
1377–1398.
Kalwij, A. (2000). Estimating the economic return to schooling
on the basis of panel data. Applied Economics, 32, 61–71.
Kawaguchi, D. (2002). Human capital accumulation of salaried
and self-employed workers. Labour Economics, 10, 55–71.
Mincer, J. (1974). Schooling, experience and earnings. New
York: Columbia UniversityPress for the National Bureau
of Economics Research.
Monks, J. (2000). The returns to individual and college
characteristics. Evidence from the National Longitudinal
Surveyof Youth. Economics of Education Review, 19(2),
279–289.
Polachek, S. W., & Kim, M. K. (1994). Panel estimates of the
gender earnings gap: Individual-specific intercept and
individual-specific slope models. Journal of Econometrics,
61, 23–42.
Psacharopoulos, G. (1994). Returns to investment in education:
A global update. World Development, 22(9), 1325–1343.
Rees, H., & Shah, A. (1986). An empirical analysis of self-
employment in the UK. Journal of Applied Econometrics, 1,
95–108.
Rosholm, M., & Smith, N. (1996). The Danish gender wage gap
in the 1980s: A panel data study. Oxford Economics Papers,
48, 254–279.
Spence, M. (1973). Job market signaling. Quarterly Journal of
Economics, 87, 355–374.
Staiger, D., & Stock, J. H. (1997). Instrumental variables
regressions with weak instruments. Econometrica, 65(3),
557–586.
Stanovnik, T. (1997). The returns to education in Slovenia.
Economics of Education Review, 4, 443–450.
Taylor, M. P. (1996). Earnings independence or unemployment:
Whybecome self-employed? Oxford Bulletin of Economics
and Statistics, 58, 253–266.
Trostel, P., Walker, I., & Woolley, P. (2002). Estimates of the
economic return to schooling for 28 countries. Labour
Economics, 9, 1–16.
Vieira, J. A. (1999). Returns to education in Portugal. Labour
Economics, 6(4), 535–541.
Vila, L., & Mora, J. G. (1998). Changing returns to education
in Spain during the 80’s. Economics of Education Review,
17(2), 173–178.
Webbink, D., & Hartog, J. (2004). Can students predict starting
salaries? Yes!. Economics of Education Review, 23, 103–113.
Williams, D. (2000). Consequences of self-employment for men
and women in the Unites States. Labour Economics, 7,
665–687.
Willis, R. J. (1986). Wage determinants: A surveyand
reinterpretation of human capital earnings function. In O.
Ashenfelter, & R. Layard (Eds.). The Handbook of Labor
Economics 1 (pp. 525–602). Amsterdam: North Holland-
Elsevier.
Wooldridge, J. M. (2002). Econometric Analysis of Cross-
Section and Panel Data. Cambridge, MA: MIT Press.
ARTICLE IN PRESS
I. Garcı´a-Mainar, V.M. Montuenga-Go´mez / Economics of Education Review 24 (2005) 161–170
170
Document Outline
Wyszukiwarka
Podobne podstrony:
panele v2 streszczenie
panele v2 prezentacja
panele v3 artykul
duration analysis v2 artykul
markov v2 artykul
nieparametryczne v2 artykul
panele v1 artykul
DTC v2
dodatkowy artykul 2
ARTYKUL
Elektro (v2) poprawka
l1213 r iMiBM lakei v2
logika rozw zadan v2
więcej podobnych podstron