bs_bs_banner

Asian Economic Journal 2014, Vol. 28 No. 2, 181–199 181

Vertical and Horizontal Educational Mismatches

of Female Graduates in Taiwan*

Hung-Lin Tao and Chia-Yu Hung

Received 30 October 2012; accepted 3 February 2014

This study considers both vertical and horizontal educational mismatches, with the
former referring to overeducation and undereducation, and the latter to the mis-
match between college major and job. It is found that the wage premium of the
vertical educational match is greater than that of the horizontal educational match.
A better vertical match augments the wage premium of an improvement in the
horizontal match, and vice versa. The horizontal educational mismatch appears to
be an extended scenario of overeducation because graduates from colleges with
low rankings have a higher probability of being vertically overeducated as well as
horizontally mismatched. Graduates from highly-ranked colleges are privileged to
not only have high earnings but also to have low probabilities for the vertical and
horizontal mismatches. These low probabilities indirectly raise their earnings. The
indirect effects of academic characteristics on earnings are calculated. The
approach we propose provides an insight into how academic characteristics com-
prehensively influence earnings.

Keywords: educational mismatches, academic characteristics, college graduates,

earnings.
JEL classification: I20.
doi: 10.1111/asej.12032

I. Introduction
Over the past two decades since 1987, higher education in Taiwan has gone
through a phase of rapid expansion. The rapid expansion in higher education has
resulted in Taiwan having to meet the associated costs in at least three ways. First,
any new public colleges have been founded by the government, and the govern-
ment is responsible for providing the funds to operate these public colleges. The
official investment, however, seems to have been in vain. Second, many high
school graduates are keen to attend colleges. They pay their tuition fees and spend
4 years of their time earning a degree. Unfortunately, the payoff in the labor
market is not commensurate with students’ tangible and intangible costs (fore-
gone earnings) of earning a college degree. Third, a college degree increases

*Tao (corresponding author): Department of Economics, Soochow University 56 Kuei-Yang

Street, Sec. 1, Taipei 100, Taiwan. Email: hltao@scu.edu.tw. Hung: Department of Economics,
National Dong Hwa University, No. 1, Sec. 2, Da Hsueh Road, Shoufeng, Hualien, Taiwan. The
authors would like to express gratitude to the National Taiwan Normal University for providing data
from the Integrated Higher Education Data Base System in Taiwan.

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
ASIAN ECONOMIC JOURNAL 182

graduates’ expectations in their job search. Because the demand for graduates has
not risen sufficiently quickly, many graduates either take on a mismatched job or
become unemployed. College graduates frequently accept jobs that do not require
a college degree or do not match their specialties. As a result, human resources in
Taiwan are severely misallocated.
The issues related to overeducation in many countries, including Taiwan, have
been intensively examined in recent decades, with the focus in the past few years
having been on explaining the prevalence of overeducation.1 The impact of
Taiwan’s overeducation on wages is discussed in Lin and Wang (2005) and Hung
(2008). Both papers conclude that, the earnings resulting from overeducation
(undereducation) are less (more) than those due to adequate education.2 It should
be emphasized that in the existing research, an educational mismatch often refers
to workers having either more or less education than that which is required for the
job. The numbers and sizes of colleges are increasing in most countries, with
college graduates accounting for a growing portion of the labor force. A vertical
educational mismatch (i.e. overeducation or undereducation), which results in a
misallocation of educational resources among the labor force, is not the only
concern. A horizontal educational mismatch, that is, where there is a mismatch
between a person’s major and the available jobs, has also surfaced as an emerging
challenge that is being faced by educational authorities. Ignoring or not control-
ling for a horizontal educational mismatch might lead to problematic conclusions
in regard to the impact of overeducation on wages. Variables for the horizontal
educational mismatch that are omitted could bias the estimators if they are
correlated with the vertical educational mismatch variables. Although some
studies, for instance, Allen and van der Velden (2001) and Di Pietro and Urwin
(2006), do control for the horizontal educational mismatch in their regression
models, the horizontal educational mismatch is an issue that was not emphasized
until Robst (2007). Robst (2008) further incorporates vertical educational mis-
match to complete the exploration of educational mismatches.
Similar to Robst (2008), one of the purposes of the present study is to simul-
taneously investigate the wage effects of the vertical and horizontal educational
mismatches of college graduates. The incorporation of both types of educational
mismatch enables us to compare the wage penalties inflicted by the vertical and
the horizontal educational mismatches, and shows us which of these two types of
educational mismatches is more serious. This paper differs from Robst (2008) in

1 Three theories are typically employed when explaining the existence of overeducation. They are
the assignment theory, the heterogeneous skill theory and the human capital theory. Studies usually
use the relationship between overeducation and being too highly skilled to verify whether the
assignment theory or the heterogeneous skill theory can provide an explanation for the existence of
overeducation. See Allen and van der Velden (2001), Hartog (2000), Dolton and Vignoles (2000), Di
Pietro and Urwin (2006) and Green and McIntosh (2007) for more detail.
2 For example, college graduates in a job requiring a high school diploma are overeducated, and
those in a job requiring a master’s degree are undereducated. Therefore, given the same educational
attainment, undereducated graduates might have more human capital than overeducated graduates.

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES 183

at least two ways. First, in addition to their college major, our data provide more
detailed information about the respondents’ academic performance, including
college type (implying college ranking) and college grade. In other words, our
model is able to control the respondents’ ability. With this information in hand,
we will probe into the relationship between academic characteristics and educa-
tional mismatches. Indeed, Battu et al. (1999) and Dolton and Vignoles (2000)
point out that the college type or grade reduces the likelihood of overeducation,
although they do not consider horizontal educational mismatches. Second, the
present study further explores whether bad horizontal educational mismatches
lead to a deterioration in the earnings penalty of vertical educational mismatches,
and vice versa. Moreover, as pointed out above, academic characteristics affect
the likelihood of educational mismatches, and the mismatches cause wages to
decline. Consequently, academic characteristics affect wages in at least two ways.
They directly affect wages and the likelihood of mismatches, which, in turn,
indirectly affects wages. We will explicitly explore these two paths.3
The remainder of this paper is organized as follows. Section II defines the
vertical and horizontal educational mismatches, and introduces the data and
describes their characteristics. Earnings and educational mismatch models are
then specified in Section III. The empirical results are presented and discussed in
Section IV. Section V estimates the direct and indirect wage effects, through
educational mismatches, of academic characteristics. Finally, Section VI
concludes.

II. Definition of Educational Mismatches and Data Sources

II.1 Definition of educational mismatches

Two questions tend to be asked in educational mismatch surveys. One question
asks employed respondents whether their jobs are highly, partly or not related to
their college major. The answer to this question is used to evaluate the extent to
which the major matches the job, and is referred to as a horizontal educational
mismatch. The other question, for example, in the survey that the present study
uses, asks the respondents to specify the appropriate educational attainment
needed to be qualified for the current job.4 There are seven choices: junior high
school, senior high school, vocational senior high school, junior college, college,
masters and PhD. Because all respondents are college graduates, the first four
3 Moreover, this paper focuses on entry workers who have just graduated from college. All
respondents are of roughly the same age. The data that Robst (2008) uses consists of respondents of all
ages. The advantage of our dataset is that it excludes the influence of on-the-job training on earnings.
4 In this study, both horizontal and vertical educational mismatches are based on the respondents’
self-assessments. The data that Robst (2008) uses contain the respondents’ self-assessments based on
horizontal educational mismatches, but do not contain self-assessments of vertical educational mis-
matches. He applies the ‘means of realized matches (MRM)’ proposed by Verdugo and Verdugo
(1989), with required schooling defined as a one-standard deviation range around the mean level of
schooling within an occupation. Groot and Maassen van den Brink (2000) and Hung (2008) show that

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
ASIAN ECONOMIC JOURNAL 184

Table 1 Types of educational mismatch

Type Content Group

0 Overeducation and major not related to job MS0

1 Overeducation and major partly related to job MS1
2 Overeducation and major highly related to job MS2
3 Adequate education but major not related to job MS3
4 Adequate education but major partly related to job MS4
5 Adequate education and major highly related to job MS5A
6 Undereducation and major not related to job MS5a
7 Undereducation and major partly related to job
8 Undereducation and major highly related to job

Groups 6, 7 and 8 are merged to form a new group MS5a due to insufficient observations.

choices are classified as overeducation, ‘college’ as adequate education, whereas

‘masters’ and ‘PhD’ are classified as undereducation. Overeducation, adequate
education and undereducation are used to construct the degree of the vertical
educational mismatch. As a result, there are three degrees of horizontal educa-
tional mismatch as well as three degrees of vertical educational mismatch. Con-
sidering the vertical and the horizontal educational mismatch together, there are
nine types of educational mismatch, as presented in Table 1. Because of insuffi-
cient observations for undereducation, as in Robst (2008), all of the three groups
classified as undereducation are merged into a single group.
Because it is unclear at this moment how to assign the degree of educational
mismatch for each group represented in Table 1, we propose that the degree of
educational mismatch for these groups be based on the rankings of their wages.
The empirical results from Allen and van der Velden (2001) and Di Pietro and
Urwin (2006) provide us with preliminary information regarding the wage
ranking. First, in model 4 of table 4 in Allen and van der Velden (2001),
overeducation, undereducation, and ‘skill outside own field’ are considered.
Overeducation and undereducation capture the effects of the vertical educational
mismatch, while ‘skill outside own field’ captures the effect of the horizontal
educational mismatch. Their results show that, given the same educational attain-
ment, there are significantly positive returns to undereducation and significantly
negative returns to overeducation. In addition, the absolute effect of overeduca-
tion is greater than that of undereducation, and the latter is greater than the
insignificant effect of ‘skill outside own field’. Moreover, the former two effects
are significant at the 1-percent level, while the effect of the ‘skill outside own
field’ is not significant. In specification 4 in table 4 in Di Pietro and Urwin (2006),
the vertical educational mismatch (overeducation and undereducation) and the
horizontal educational mismatch both appear in the model. The absolute effect of
the overeducation is the greatest, while the absolute effect of the undereducation

the proportions of mismatches for MRM are lower than those for self-assessments. It is worthwhile
showing the estimation results of an alternative measure of vertical mismatches.

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES 185

is the least and is insignificant. The absolute effect of the horizontal educational
mismatch is significant and greater than that for undereducation, but less than that
for overeducation. It appears that existing studies agree that the educational match
improves from type 0 to type 4, as defined in Table 1, but there is no consistent
conclusion regarding the order of type 5 and the regrouping of types 6, 7 and 8.
For this reason, Table 1 does not rank these two groups, but denotes them as MS5A
and MS5a , respectively. The degree of the educational mismatch of these two
groups will be verified by their wages, and the group with the higher wage will be
assigned MS6, while the group with the lower wage will be assigned MS5.

II.2 Data source and statistics

The data used are compiled by the Center for Higher Education Research at
National Tsing Hua University, which is sponsored by Taiwan’s National Science
Council as well as its Ministry of Education. The respondents in the survey were
college graduates who graduated in June 2003, and were surveyed between
August 2004 and January 2005 (i.e. between 1 year and 2 months and 1 year and
7 months after their graduation). The advantage of using the entry wage is that the
impacts of the on-the-job training are completely ruled out. Imagine a college
graduate who takes a masters-level job, who is undereducated, and at the outset is
insufficiently skilled. To make up for the deficiency in capability needed for her
job, she devotes more time to her job, and participates in on-the-job training as
much as she can. Years later, she is finally sufficiently capable of doing her job.
When she was surveyed, she answered that she was undereducated but not
underskilled at the time of the survey, although she was underskilled at the outset.
Because of her capability, her wage is probably similar to that of someone with
adequate education and the right skills. This example emphasizes the possible
problem of using experienced workers as the surveyed sample.
The original data contained 12 263 female and male college graduates, who
made up approximately 8 percent of all college graduates for the year 2003.
Because most male college graduates are obliged to perform 1 year and 10 months
of military service, most of the male college graduates were not in the labor
market when the survey was conducted. A few male graduates who were not in
military service during the survey might have had some health problems and,
therefore, were not a representative sample of male graduates. Consequently, the
male respondents are excluded from the sample. Only the female graduates left in
the dataset are used in this study. To maintain the consistency of all observations,
the college graduates with 2-year college degrees were deleted, leaving only
graduates with 4-year college graduates in the sample. Finally, after deleting
some missing data, the final sample used in this study consisted of 3923 full-time
workers. The variables used in the survey included the respondents’ employment
status, occupation, college major, grade and extracurricular activities.
Table 2 summarizes the relationships between the college graduates’ academic
characteristics and the types of educational mismatch (MS). The mean monthly
© 2014 The Authors
Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
 Table 2 Number of observations and average wage by educational mismatch and academic characteristics
Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
© 2014 The Authors

ASIAN ECONOMIC JOURNAL

Mismatch type Overeducation Adequate Undereducation Observations Mean wage†

Not related Partly related Highly related Not related Partly related Highly related

(MS0) (MS1) (MS2) (MS3) (MS4) ( MS5A ) ( MS5a )

College type
Public college (C1) 48 (6%) 54 (7%) 32 (4%) 33 (4%) 125 (16%) 431 (55%) 66 (8%) 789 36 356
Public technical 55 (16%) 78 (22%) 78 (22%) 18 (5%) 50 (14%) 60 (17%) 13 (4%) 352 30 157
college (C2)
Private college (C3) 243 (16%) 314 (21%) 176 (12%) 103 (7%) 275 (18%) 343 (23%) 57 (4%) 1511 30 823
Private technical 302 (24%) 349 (27%) 233 (18%) 54 (4%) 131 (10%) 171 (13%) 31 (2%) 1271 26 768
college
College grade
Score 60–70 34 (31%) 19 (17%) 10 (9%) 8 (7%) 16 (15%) 20 (18%) 3 (3%) 110 28 672
Score 70–80 (S2) 252 (20%) 296 (23%) 188 (15%) 82 (6%) 187 (15%) 233 (18%) 50 (4%) 1288 29 395
Score 80–90 (S3) 260 (12%) 395 (19%) 280 (13%) 91 (4%) 320 (15%) 659 (31%) 100 (5%) 2105 31 539
Score above 90 (S4) 10 (11%) 11 (13%) 7 (8%) 6 (7%) 17 (20%) 33 (38%) 3 (3%) 87 33 033
Forgot score (S5) 92 (28%) 74 (22%) 34 (10%) 21 (6%) 41 (12%) 60 (18%) 11 (3%) 333 28 881
Major
Business (M1) 229 (19%) 408 (33%) 158 (13%) 62 (5%) 204 (17%) 149 (12%) 22 (2%) 1232 27 937
Social science (M2) 23 (12%) 16 (9%) 5 (3%) 13 (7%) 42 (23%) 71 (38%) 16 (9%) 186 31 661
Literature (M3) 107 (12%) 98 (11%) −44 (5%) 42 (5%) 144 (16%) 408 (45%) 55 (6%) 898 33 673
Science (M4) 60 (19%) 80 (26%) 23 (7%) 10 (3%) 52 (17%) 78 (25%) 10 (3%) 313 29 893
Medicine (M5) 45 (11%) 29 (7%) 149 (36%) 11 (3%) 28 (7%) 126 (31%) 21 (5%) 409 37 347
Engineering (M6) 61 (22%) 63 (22%) 37 (13%) 24 (8%) 39 (14%) 43 (15%) 16 (6%) 283 28 061
Other (M7) 68 (20%) 60 (18%) 52 (15%) 21 (6%) 42 (12%) 76 (22%) 20 (6%) 339 27 430
Law 3 (10%) 3 (10%) 0 (0%) 0 (0%) 7 (23%) 16 (52%) 2 (6%) 31 34 293
Media 52 (22%) 38 (16%) 51 (22%) 25 (11%) 23 (10%) 38 (16%) 5 (2%) 232 27 653

186
†
Mean wage is in new Taiwan dollars. One US dollar is approximately 30 new Taiwan dollars.
VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES 187

wages are reported based on academic characteristics and MS type. Note that the
higher values of the subscripts in MS denote better educational matches, although
it is not clear at this moment if MS5A or MS5a represents a better match. Basically,
the mean wage also rises as the value of the MS subscript increases. First, we
group colleges into four types, namely, public colleges, private colleges, public
technology colleges and private technology colleges, according to the ranking in
terms of reputation in descending order. Colleges and technology colleges are,
respectively, similar to the universities and polytechnics presented in Dolton and
Vignoles (2000). The former are more academically-oriented, while the latter are
more vocationally-oriented. In Taiwan, most college freshmen come from senior
high schools, whereas most technology college freshmen come from senior voca-
tional high schools. In general, admission to a college is more competitive than
that to a technology college. The ranking of the mean wage coincides with the
ranking of the college types, while the mean wage for the public technology
colleges is close to that for private colleges. Furthermore, for graduates from
public colleges, the proportions of those falling into better matches are the
highest, while the proportions of those falling into worse matches are the lowest.
Table 2 also shows that the higher the grade, the higher the wage and the higher
the proportion of those falling into better matches. Regarding the performance of
each type of major, we find that the graduates majoring in medicine, law, litera-
ture and social science earn relatively more than those taking other majors. These
graduates also have a higher propensity to have better matches and fall in the
types MS5A and MS5a . In particular, the percentages of those falling into MS5A,
being adequately educated and having majors that are highly related to their jobs
are all above 30 percent. It is also surprising that those with literature and social
sciences majors have higher earnings than those with engineering majors. We
suppose that this is partly because a majority of graduates who major in engi-
neering have graduated from private technology colleges, which have lower
college rankings.5 By contrast, graduates with business or media majors perform
the worst, in that more than 60 percent are overeducated and only 12–16 percent
of them fall into MS5A.

III. The Models

Our first concern is how vertical and horizontal educational mismatches and
academic characteristics impact entry wages. The wage equation, adjusted with a
sample selection term, is presented below.6 As in the literature, the wage loga-
rithm is used:
5 In the full sample, 20, 39, 9 and 32 percent of the graduates are, respectively, from public colleges,
private colleges, public technology colleges and private technology colleges. In the full sample, 283
graduates majored in engineering. In this subsample, 10.6, 22.6, 20.9 and 45.9 percent of them,
respectively, graduated from those four types of colleges. More of the graduates who majored in
engineering are from private technology colleges.
6 A multinomial logit selection model proposed by Lee (1983) is used to correct for the possible
sample selection problem. The first step involves the use of a multinomial logit model of beginning

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
ASIAN ECONOMIC JOURNAL 188

3 5 7 7 6 6
log w = β0W + ∑ βiW Ci + ∑ δ iW Si + ∑ λiW M i + γ W Exa + ∑ π i Oi + ∑ θ i MSi + ∑ ψ i I i
i =1 i=2 i =1 i =1 i =1 i=2
+ η ⋅ lambda + ε W
(1)
In Equation 1,7 C1, C2 and C3 denote public colleges, private colleges and public
technology colleges, respectively. The private technology college is the compari-
son base. S2, S3, S4 and S5 represent college grades, and are introduced in Table 2.
The worst grade, 60 to 70, is the comparison base. M1 to M7 are majors, illustrated
in Table 2, and media and law are the comparison base.8 Exa is the number of
extracurricular activities at the college in which a worker participated. Extracur-
ricular activities sharpen the students’ abilities, such as communication, organiza-
tion and leadership skills. Persico et al. (2004) argue that by participating in social
activities young teens can facilitate the accumulation of human capital from social
adaptability. This might help them to adapt to the labor market. O1 to O7 are seven
occupational dummy variables introduced in the previous section, and the sales-
men form the reference group. In addition, the explanatory variables in the entry
wage model include six educational mismatch dummy variables, MS1 to MS4,
MS5A and MS5a . Ii, i = 1, . . . , 6, is the institutional type introduced in the previous
section, and the non-listed companies make up the reference group. Lambda is the
sample selection term, and η is its regression coefficient. The significance of η
indicates whether the sample selection problem is serious. β, δ, λ, γ, π, θ and ψ are
the regression coefficients of occupations and educational mismatches, respec-
tively. W and ε are the monthly wage and the error term, respectively.
The second model will be specified to explore the relationship between the
degree of educational mismatch and academic characteristics. The types of col-
leges contain information regarding college ranking. Battu et al. (1999), Dolton
and Vignoles (2000) and Green and McIntosh (2007) indicate that academic
characteristics determine educational mismatches. The educational mismatch in
their study refers to the vertical educational mismatch. Although Di Pietro and
Urwin (2006) took college grades into account while estimating the impacts of
vertical educational mismatches on wages, they did not examine whether college
grades affect vertical educational mismatches.

status choice (unemployment, graduate study, part-time worker and full-time worker). The number of
observations for the multinomial logit model is 6725. Because the results of the multinomial logit
model are not our concern, we omit them from the present study. The next section will show that the
sample selection problem is not serious.
7 Because all respondents are women in the same cohort who graduated from college in the same
year, unlike in the general earnings model, this earnings equation does not include sex, age, working
experience or educational attainment.
8 It is not our intention to set two majors as the comparison base. This is because the number of law
major observations is limited (see Table 2), because most technology colleges do not have a law
school, and some of the mismatch groups do not have law observations. The ordered probit model (i.e.
Equation 2) cannot converge if law is an independent major. Therefore, the law major is also included
in the comparison base.

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES 189

The educational mismatch ordered probit model has a dependent variable that
is defined in Table 1. Similar to Robst (2008), the dependent variable pools both
vertical and horizontal educational mismatches. The model is expressed as Equa-
tion 2. Assume that MS* is an unobserved and continuous variable referring to the
quantity of educational mismatches. What can be observed is MS, a discontinuous
variable of the degree of educational mismatch. MS = 0 if overeducation and the
major are not related with the job, and MS = 1, 2, 3, 4, 5 and 6, respectively,
correspond to the same subscript values of MS in Table 1 and to their definitions,
except that groups 5 and 6 are not identified at this moment. The order of the MS
groups will be based on the results of the wage equation. The worse the education
mismatch is in terms of the wage penalty, the smaller the value of MS that will be
assigned. Therefore, the educational mismatch ordered probit model is presented
below:
3 5 7
MS * = β0MS + ∑ βiMS Ci + ∑ δ iMS Si + ∑ λiMS M i
i =1 i=2 i =1
7
+γ MS
Exa + ∑ φi Fedui + ε ,
MS

i=2

MS = 0 if MS * ≤ 0, MS = 1 if 0 < MS * ≤ Mu (1) ,

MS = J if Mu ( J − 1) < MS * ≤ Mu ( J ) , J = 2, 3, 4, 5 and MS = 6 if Mu (5) ≤ MS *

(2)

In Equation 2, the coefficients of the explanatory variables, which are in Equation

1, are superscripted with ‘MS’ to differentiate them from those in Equation 1.
Fedu refers to the father’s education, which is classified into seven categories.
Fedu1 relates to elementary school and is the reference group. Fedui, i = 2, . . . , 7,
represent junior high school, senior high school, junior college, college, graduate
school and missing, respectively. The Mus are unknown parameters of the thresh-
olds to be estimated. It should be noted that the academic variables are included
in both Equation 1 and 2. The coefficients of the academic variables in Equation 2
and the coefficients of the educational mismatches in Equation 1 can be used to
estimate the indirect wage effects of the academic characteristics. The coefficients
of the academic variables in Equation 1 provide direct wage effects of the aca-
demic characteristics.

IV. Empirical Results and Discussions

IV.1 Determination of the entry wage

The explanatory variables in the entry wage model include previous variables
related to academic characteristics, six educational mismatch dummy variables,
seven occupational dummy variables, six institutional variables, and the sample
© 2014 The Authors
Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
ASIAN ECONOMIC JOURNAL 190

selection term, lambda. The regression results are presented in Table 3. First of
all, the coefficient of lambda is not significantly different from zero, indicating
that the sample selection problem is not serious. Hence, we re-estimate the wage
model without the sample selection term and thereafter we will only focus on
this OLS regression when interpreting the results. The results show that, in

Table 3 Wage equation of sample selection and OLS models

Model Sample selection OLS

Variable Coefficient t-ratio Coefficient t-ratio

Constant 9.971 75.369* 10.034 341.108*

Public college (C1) 0.159 3.616* 0.179 14.068*
Public technical college (C2) 0.088 4.056* 0.080 5.810*
Private college (C3) 0.103 10.742* 0.102 11.023*
Score 70–80 (S2) 0.009 0.324 0.016 0.740
Score 80–90 (S3) 0.022 0.690 0.034 1.527
Score above 90 (S4) 0.034 0.686 0.053 1.635
Forgot score (S5) 0.001 0.017 0.018 0.733
MS1 0.021 1.775* 0.021 1.768*
MS2 0.059 4.269* 0.059 4.243*
MS3 0.060 3.377* 0.061 3.374*
MS4 0.084 6.359* 0.085 6.344*
MS5A 0.142 10.825* 0.142 10.767*
MS5a 0.128 6.377* 0.128 6.347*
Extracurricular activities (Exa) 0.005 0.856 0.003 0.721
Business (M1) 0.054 2.575* 0.047 3.033*
Social science (M2) 0.026 1.004 0.019 0.878
Literature (M3) 0.079 2.788* 0.068 4.067*
Science (M4) 0.052 1.794* 0.063 3.327*
Medicine (M5) 0.258 7.625* 0.244 12.911*
Engineering (M6) 0.032 1.029 0.044 2.211*
Other (M7) −0.017 −0.838 −0.020 −1.092
Managers (O1) 0.130 4.626* 0.130 4.598*
Professional (O2) 0.344 8.403* 0.345 8.370*
Teachers (O3) 0.034 1.788* 0.034 1.751*
Associate professionals (O4) 0.054 3.677* 0.054 3.655*
Technicians (O5) −0.038 −2.371* −0.038 −2.348*
Clerks and other staff (O6) −0.054 −4.185* −0.054 −4.154*
Others (O7) −0.002 −0.146 −0.002 −0.143
Listed company (I2) 0.083 7.781* 0.083 7.718*
Public organization (I3) 0.077 4.335* 0.076 4.302*
School (I4) 0.114 6.938* 0.114 6.873*
Self-employed (I5) −0.008 −0.674 −0.008 −0.673*
Others (I6) 0.034 2.497* 0.034 2.478*
Lambda −0.022 −0.605
R2 0.412 0.412
Number of Observations 3923 3923

*denotes significance at the 10% level.

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES 191

descending order, the wage ranking starts from public colleges, and extends to
private colleges, public technology colleges and then private technology colleges.
The graduates with higher grades do not have significantly higher earnings. More
involvement in extracurricular activities does not increase earnings significantly.
After controlling for the occupations, the graduates with medicine, literature,
science, business and engineering majors earn more than those with the base
major. The occupations of managers, professionals, teachers and associate pro-
fessionals give rise to significantly higher earnings, while the occupations of
technicians and clerks result in significantly lower earnings than for salesmen at
the 5-percent level. The graduates working in listed companies, public organiza-
tions, schools and other institutions have higher earnings than their counterparts
in non-listed companies. However, we have to note that the coefficients here only
estimate the direct wage effects of the academic characteristics. The comprehen-
sive analysis of the wage effects will include the indirect effects later.
Last but not least, Table 3 also demonstrates that the coefficients of MSi are all
significantly positive, and basically the coefficients are increasing as the subscript
values of MS increase. That is, the wages rise with the degree of educational
matches. More importantly, the coefficients of the educational mismatches verify
that MS5a = MS5 and MS5A = MS6 . This confirms and completes our previous
ordering of the educational mismatches.
The wage effects of all types of educational mismatches in Table 3 allow us to
compare the direct effects of these educational mismatches. Table 4 summarizes all
the comparisons. The comparisons are made by using the coefficients of the first
row variable minus the coefficients of the first column variable and we also report
the corresponding t-test for each comparison. All the educational mismatch coef-
ficients are from Table 3. Before interpreting Table 4, recall that MS5a is
undereducation in the vertical educational mismatch but, because of the limitations
of the observations, does not control the horizontal educational mismatch. Hence,
it is suggested that MS5a be temporarily ignored and this is why Table 4 places MS5a
in the far right column. The values in Table 4 in solid-line blocks indicate that they
contain only the effects of the horizontal educational mismatches. For example, the
values in the upper left corner block are all in the same level of vertical educational
mismatch, overeducation, and indicate only the differences among the horizontal
educational mismatches. In this block, for example, the value 0.059 (5.9 percent)
denotes the difference between ‘highly related’ and ‘not related,’ and is controlled
at overeducation. The bottom right block can be interpreted in a similar manner.
Five out of six types of wage differences across horizontal educational mismatches
are different from zero and at least significant at the 10-percent level, as the vertical
educational mismatch is controlled. By contrast, the values in the dotted-line
blocks contain only the effects of the vertical educational mismatches. For
example, in the dotted block in the center, the value 0.064 represents the difference
between adequate education and overeducation, controlled at the partly-related
level. All the other values not in blocks contain both the effects of the vertical and
horizontal educational mismatches.
© 2014 The Authors
Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
 Table 4 Comparisons of all types of educational mismatches

Overeducation Adequate education Undereducation

Partly related Highly related Not related Partly related Highly related

(MS1) (MS2) (MS3) (MS4) ( MS5A = MS6 ) ( MS5a = MS5 )

© 2014 The Authors

Overeducation

Not related
(MS0) 0.021 0.059 0.061 0.085 0.142 0.128
ASIAN ECONOMIC JOURNAL

(t ratio) (1.768)* (4.243)*** (3.374)*** (6.344)*** (10.767)*** (6.347)***

Partly related
(MS1) 0.038 0.040 0.064 0.121 0.107
(t ratio) (2.066)** (5.039)*** (10.496)*** (15.640)*** (18.939)***
Highly related
(MS2) 0.001 0.026 0.083 0.069
(t ratio) (0.076) (1.506) (5.095)*** (4.028)***

Adequate

Not related
(MS3) 0.024 0.082 0.068
(t ratio) (1.553) (8.089)*** (6.850)***
Partly related
(MS4) 0.058 0.044
(t ratio) (3.204)*** (3.530)***
Highly related
( MS5A ) −0.014
(t ratio) −(0.767)

All comparisons are made by using the coefficients of the first row variable minus the coefficients of the first column variable. All the coefficients are from Table 3.

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
192

The solid-line and dotted-line blocks indicate that they contain only the effects of the horizontal and only the effects of the vertical educational mismatches,
respectively. ***, ** and * denote significance at the 1, 5 and 10% levels, respectively.
VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES 193

Table 5 Wage premiums of horizontal and vertical educational matches

Horizontal Vertical condition Horizontal Vertical differentiation

differentiation condition
Overeducation Adequate education Adequate education −
overeducation

Partly related − 0.021 0.024 Not related 0.061

not related (1.768)* (1.553) (3.374)***
Highly related − 0.038 0.058 Partly related 0.064
partly related (2.066)** (3.204)*** (10.496)***
Highly related − 0.059 0.082 Highly related 0.083
not related (4.243)*** (8.089)*** (5.095)***

The values in Table 5 are summarized from Table 4. The alues in the parentheses are t-values.

Given the type of educational mismatch being compared in the first column, it
can be easily observed that the values rise from column MS1 to column MS5A for
all rows. Table 5 summarizes those values in the solid-line and dotted-line blocks
in Table 4 to further compare the magnitudes of the effects of the vertical and the
horizontal educational matches. The left part of Table 5 shows that given the
vertical match (the top row), the wage premium increases with an improvement in
the horizontal match. The results emphasize that a better vertical educational
match enlarges the wage premium of an improvement in the horizontal educa-
tional match. Note that the wage premium of ‘highly related’ to ‘not related’ (the
bottom row) is the sum of its top two rows; that is, (highly related − partly
related) + (partly related − not related).
The right part of Table 5 indicates that, given the horizontal match (the left
column of the right part), the wage premium rises with the improvement in the
vertical educational match. The results emphasize that a better horizontal educa-
tional match augments the wage premium of an improvement in the vertical
educational match. Therefore, the ‘marginal’ return from improving the educa-
tional match, either vertically or horizontally, is increasing. It is worth waiting for
a better match if one believes that finding a better job is just a matter of time.
Either type of educational match augments the wage premium of the other type of
educational match. Table 5 also demonstrates that the premiums (penalties) of the
vertical educational matches (mismatches) are greater than those of the horizontal
educational mismatches. For example, for a graduate with a job characterized by
overeducation that is also partly related to his/her college major, the wage
premium is 0.038 if his/her horizontal match is improved to ‘highly related’,
given that the vertical match has not been changed. For the same person, the wage
premium is 0.064 if the vertical match is improved to ‘adequate education’, while
the horizontal match has not been changed. Table 5 shows that the wage premium
of the vertical improvement is greater than that of the horizontal improvement. It
appears that the effect of the vertical educational match is more substantial than
that of the horizontal educational match.
© 2014 The Authors
Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
ASIAN ECONOMIC JOURNAL 194

The interpretations of the most right column MS5a (undereducation) in Table 4

are not so straightforward because it contains all kinds of horizontal educational
matches. The coefficient of MS5a is significantly greater than the coefficients of
MS1 to MS4. This means that the coefficient of undereducation MS5a is greater than
the coefficient of overeducation for all kinds of horizontal educational mismatch
and those of adequate education with a major not related to or only partly related
to the job. The difference between MS5a and MS5A is insignificant.

IV.2 Determination of educational mismatch

The order of the dependent variables in the ordered probit model of the educa-
tional mismatch is confirmed above. The independent variables are the academic
characteristics, the father’s educational attainment and the number of extracur-
ricular activities in college. The estimated coefficients and the corresponding
marginal effects of the ordered probit model are presented in Table 6.
First, the coefficients for public colleges, private colleges and public technol-
ogy colleges are all significantly positive. The good ranking in educational
matches starts with public colleges, private colleges, public technology colleges
and then private technology colleges. The ranking of college types in terms of
educational matches accords with the ranking of wage performance reported
above. In particular, it was found that graduates attending colleges with more
academic orientation were more likely to be better matched than those attending
technology colleges. Publicly-established colleges also rank better than
privately-established colleges. In terms of the marginal effect, there is a 35.5-
percent increase in terms of falling in MS6 (adequate and highly related) if the
student graduated from a public college as compared to a private technology
college. To be more specific, the sum of the marginal effects of MS3 to MS6
represents the marginal probabilities of falling in better matches, adequate edu-
cation or undereducation, in terms of the vertical educational mismatch. In con-
trast, the sum of the marginal effects of MS0 to MS2 represents the marginal
probabilities of falling in worse matches; that is, overeducation. Our regression
results provide evidence of a clear trend that there are positive and increasing
marginal probabilities of falling in better vertical matches and negative and
decreasing marginal probabilities of falling in overeducation as the graduates
move from lower to higher ranked colleges; that is, from private technology
colleges to public technological colleges (C2), private colleges (C3) and public
colleges (C1).
Furthermore, the ranking of college types also has an impact on the degrees of
horizontal educational mismatch. By controlling the same level of horizontal
educational mismatch, given the worse matches (i.e. overeducation), the marginal
probabilities are negative and decreasing as the graduates move from lower-ranked
to more highly ranked colleges. For example, compared to private technology
colleges, the marginal probabilities of falling in the worst match (MS0) are −0.155,
−0.035 and −0.071 for public colleges, public technology colleges and private
© 2014 The Authors
Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
 Table 6 Results and marginal effects of ordered probit model of educational mismatches
Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd

VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES

Ordered probit Marginal effects

Overeducation Adequate education Undereducation

Not related Partly related Highly related Not related Partly related Highly related

Coefficient t ratio (MS0) (MS1) (MS2) (MS3) (MS4) (MS6) (MS5)

Constant 0.206 1.585

Extracurricular activities (Exa) 0.091 4.445* −0.020 −0.014* −0.003* 0.000 0.006* 0.027* 0.003*
Public college (C1) 1.014 17.603* −0.155 −0.152* −0.063* −0.013 0.010 0.355* 0.018*
Public technical college (C2) 0.177 2.766* −0.035 −0.028* −0.007 0.000 0.009* 0.056* 0.005*
Private college (C3) 0.345 8.026* −0.071* −0.053* −0.013* 0.000 0.019* 0.107* 0.010*
Score 70–80 (S2) 0.177 1.662* −0.037* −0.027* −0.006 0.000 0.010* 0.055* 0.005*
Score 80–90 (S3) 0.427 4.040* −0.094* −0.063* −0.013* 0.002 0.027* 0.127* 0.013*
Score above 90 (S4) 0.565 3.593* −0.088* −0.089* −0.036* −0.007 0.009 0.199* 0.012*
Forgot score (S5) 0.054 0.455 −0.011* −0.008* −0.002 0.000 0.003 0.017* 0.002
Business (M1) −0.104 −1.438 0.023* 0.016* 0.003 0.000 −0.007* −0.031* −0.003*
Social science (M2) 0.384 3.686* −0.067* −0.061* −0.021* −0.003 0.013* 0.130* 0.010*
Literature (M3) 0.334 4.314* −0.065* −0.052* −0.015* −0.001 0.016* 0.107* 0.010*
Science (M4) 0.152 1.692* −0.031* −0.024* −0.006 0.000 0.008* 0.048* 0.005*
Medicine (M5) 0.470 5.526* −0.081* −0.074* −0.026* −0.004 0.014* 0.159* 0.012*
Engineering (M6) 0.071 0.768 −0.015* −0.011* −0.003 0.000 0.004* 0.022* 0.002
Other (M7) 0.118 1.335 −0.024* −0.018* −0.005 0.000 0.006* 0.037* 0.004
Junior high school (EDF2) 0.009 0.159 −0.002 −0.001 0.000 0.000 0.001 0.003 0.000
Senior high school (EDF3) 0.037 0.783 −0.008* −0.006* −0.001 0.000 0.002 0.011 0.001
Junior college (EDF4) 0.041 0.715 −0.009* −0.006* −0.001 0.000 0.002 0.013 0.001
College (EDF5) 0.232 3.579* −0.045* −0.036* −0.010 −0.001 0.011* 0.075* 0.007*
Graduate school (EDF6) 0.220 1.775* −0.042* −0.035* −0.010 −0.001 0.010* 0.071* 0.006*
© 2014 The Authors

Missing (EDF7) 0.167 1.520 −0.033* −0.026* −0.007 0.000 0.008* 0.053* 0.005*
Mu( 1) 0.697 37.724*
Mu( 2) 1.085 54.915*
Mu( 3) 1.241 60.710*
Mu( 5) 1.854 78.306*
X2 879.292
Number of observations 3923

195
*Denotes significance at the 10% level.
ASIAN ECONOMIC JOURNAL 196

colleges, respectively. That is, the graduates of the highly-ranked colleges are less
likely to have jobs unrelated to their majors given that they are overeducated.9
A similar rule is also applied to MS1. Given the better matches (i.e. adequate
education), the marginal probabilities are positive and increasing for MS6
(adequate and highly related) as the graduates move from the lower-ranked to the
more highly ranked colleges. To be more specific, the marginal probabilities of
falling in the best match (MS6) are 0.355, 0.056 and 0.107 for public colleges,
public technology colleges and private colleges, respectively. That is, the gradu-
ates of the highly-ranked colleges are more likely to have jobs highly related to
their majors given that they are adequately educated. To sum up, graduates from
the lower-ranked colleges also have higher propensities to work in jobs that are
not or are partly related to their majors and suffer from the horizontal mismatch
problem. Graduates of highly-ranked colleges attain better matches and avoid
worse matches than graduates of the lower-ranked colleges.
In addition, more involvement in extracurricular activities is found to increase
the propensity for better matches, although it does not significantly increase
earnings, as shown in Table 3. The higher the college grades, the better the
educational matches the graduates can achieve. The regression results in the
‘ordered probit’ column in Table 6 indicate that the graduates majoring in medi-
cine, social science, science and literature can attain significantly better matches
and avoid worse matches than those with the base major. The coefficients of
business and engineering are not significantly different from zero. However,
according to the estimated marginal effects, the graduates majoring in engineer-
ing can also attain significantly better matches and avoid worse matches than the
reference group. Business is the only major that does worse than the base major
in terms of educational mismatch. The marginal probabilities for the business
major are significantly negative for the adequate education and undereducation
groups (MS4 to MS6) and the marginal probabilities are significantly positive for
most overeducation groups (MS0 and MS1). That is, college graduates majoring in
business are more likely to be overeducated and less likely to be adequately-
educated or undereducated. However, according to Table 3, the earnings of
graduates with a business major are significantly higher than of those with the
base major. In sum, the academic characteristics affect the level of educational
mismatches significantly.
Finally, the father’s education of more than junior high school lowers
the probability of vertical and horizontal educational mismatches among their
offspring. If the father has completed college or graduate school, it increases the
probability of an adequate education and horizontal educational match.

9 It appears that adding the father’s education to the model causes multicollinearity. The standard
errors of the marginal coefficients of public colleges and public technology colleges are 0.267 and
0.231, respectively. By dropping the father’s education from the model, the standard errors of the
marginal coefficients of the public colleges and public technology colleges decline to 0.009 and 0.005,
respectively, while the magnitudes of both coefficients do not substantially change and are significant
at the 1-percent level.

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES 197

V. Estimation of the Direct and Indirect Effects on the Entry Wage

The direct wage effects of the academic characteristics are straightforward, and
are simply the coefficients of the wage Equation 1. The indirect wage effects of
the academic characteristics are, however, not so straightforward. The coeffi-
cients in Equation 2 are not the marginal effects of their corresponding variables.
Table 6 presents the marginal probability of each explanatory variable, at its mean
value, falling in each group of educational mismatch. Therefore, the indirect wage
effects are computed using the formula below. In equation 3, [Pi(X = 1) − Pi(X =
0)] is the ‘marginal’ probability of falling in a type i educational mismatch (MSi),
X denotes any of the academic variables and θi is the coefficient of a type i
educational mismatch (MSi) in Equation 1. Note that X is not a continuous
variable, and the ‘marginal’ probability means that the probability changes from
X = 0 to X = 1. In a sense, the computations of the indirect wage effects are
‘expected’ or average effects:
6
Indirect wage effect = ∑ {[ Pi ( X = 1) − Pi ( X = 0)] mean ⋅ θ i }, (3)
i =1

where X = academic variables.

Table 7 presents the estimated direct, indirect and total wage effects for all
academic variables (i.e. type of colleges, college grade and major). The total wage
effects are the sum of the direct and indirect effects. College types are significant
both for the indirect and total wage effects. The significant indirect wage effects
suggest that the wage premium for the educational matches of the college gradu-
ates from public, private and public technology colleges are significantly greater

Table 7 Estimations of direct and indirect effects of academic characteristics on the wage

Direct effect t ratio Indirect effect t ratio Total effect t ratio

Public college (C1) 0.179 14.07* 0.046 8.70* 0.225 17.06*

Pub. tech. college (C2) 0.080 5.81* 0.008 9.15* 0.088 6.31*
Private college (C3) 0.102 11.02* 0.016 9.16* 0.118 6.52*
Score 70–80 (S2) 0.016 0.74 0.008 9.17* 0.025 1.84*
Score 80–90 (S3) 0.034 1.53 0.020 9.16* 0.054 4.00*
Score above 90 (S4) 0.053 1.64 0.026 8.78* 0.079 3.86*
Forgot score (S5) 0.018 0.73 0.003 9.18* 0.021 4.86*
Business (M1) 0.047 3.03* −0.005 −2.27* 0.042 1.47
Social science (M2) 0.019 0.88 0.018 18.96* 0.037 0.91
Literature (M3) 0.068 4.07* 0.016 12.29* 0.083 4.36*
Science (M4) 0.063 3.33* 0.007 11.47* 0.071 4.85*
Medicine (M5) 0.244 12.91* 0.022 13.27* 0.266 16.59*
Engineering (M6) 0.044 2.21* 0.003 4.75* 0.047 3.60*
Other (M7) −0.020 −1.09 0.006 18.13* −0.015 −0.99

*Denotes significance at the 10% level. The effects are measured in terms of the wage logarithm.

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
ASIAN ECONOMIC JOURNAL 198

than those from private technology colleges. When compared with the worst
college grade, all the direct wage effects of the college grade dummy variables are
insignificant at the 10-percent level, but their indirect wage effects are significant
at the 1-percent level. In particular, the effects on the wage through the indirect
way of good academic performances (S3 and S4) are substantial. The total wage
effects are significant for college grades. It is interesting to visualize that diligent
study does not strongly raise one’s wage directly, but strongly increases the
likelihood of being adequately matched and then increases the wage indirectly.
When compared with college graduates majoring in the base major, all indirect
wage effects are significantly positive, except for those majoring in business for
which the indirect wage effect is significantly negative.
The results in relation to the business major are consistent with Green and
McIntosh (2007). In Green and McIntosh (2007), the business and management
major is the reference base, and the coefficients of all other majors in the over-
qualified (overeducation) regression are negative, while in their wage regression,
the wage for the business major is higher than that for most majors. Their
explanation is that the higher propensity of being overeducated for college gradu-
ates with a business major is probably because their jobs do not demand a college
degree. By analyzing the samples from Taiwan, we also discover that the gradu-
ates majoring in business are most likely to face an unfavorable mismatch (see
Table 6), but their earnings are better than the base reference (as indicated in
Table 3). This part of the findings leads to a similar result to that found for the UK
by Green and McIntosh (2007). Without taking into consideration the indirect
effects of the educational mismatches, graduates with a business major appear to
have a high wage. However, when these indirect effects are considered, a business
major does not ensure a higher wage than the base reference. This demonstrates
the value of our empirical approach. From using this approach, we have an insight
into how academic characteristics comprehensively influence earnings.

VI. Conclusions
Our empirical findings are summarized as follows. An ordered probit model with
the confirmed order of educational mismatch, which combines both vertical and
horizontal educational mismatches, is used to analyze the relationship between
academic characteristics and the degree of educational mismatches. First, gradu-
ates of high-ranking colleges attain better matches and avoid worse matches than
graduates of lower-ranked colleges. Second, a horizontal educational mismatch
appears to be an extended scenario of overeducation because graduates from
lower-ranked colleges also have higher probabilities of working in jobs that are
not or are only partly related to their majors. In addition to college rankings,
college grades are a determinant of educational mismatches.
As for the entry wage effects of academic characteristics, we report the total
effects after the calculation of the wage premium for educational matches. The
ranking of college types in terms of wage performance accords with the ranking
© 2014 The Authors
Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
VERTICAL AND HORIZONTAL EDUCATIONAL MISMATCHES 199

of educational matches. A better college grade does not significantly raise a

person’s wage directly, but increases the likelihood of being adequately matched
and then significantly increases the wage indirectly.
The wage premium of the vertical educational match is more substantial than
that of the horizontal educational match. It is found that a better horizontal
educational match augments the wage premium of an improvement in the vertical
educational match, and vice versa. Consequently, for a graduate who has just
entered the labor market and believes that he/she can find a better matched job, it
is worth waiting. As a matter of fact, in the first stage of the sample selection
model of entry earnings (the result is not shown in the study), only graduates from
public college have a significantly higher unemployment rate. It is possible that
the oversupply of college graduates lengthens the amount of time to search for a
better matched job because it is more difficult for employers to locate a suitable
candidate in a large pool of graduates. Therefore, Taiwanese officials should
encourage the establishment of job-match accelerators to facilitate the job search
of new college graduates.

References
Allen, J. and R. Van der Velden, 2001, Educational mismatches versus skill mismatches: Effects on
wages, job satisfaction, and on-the-job search. Oxford Economic Papers, 53, pp. 434–52.
Battu, H., C. R. Belfield and P. J. Sloane, 1999, Overeducation among graduates: A cohort view.
Education Economics, 7, pp. 21–38.
Di Pietro, G. and P. Urwin, 2006, Education and skills mismatch in the Italian graduate labour market.
Applied Economics, 38, pp. 79–93
Dolton, P. and A. Vignoles, 2000, The incidence and effects of overeducation in the U.K. graduate
labour market. Economics of Education Review, 19, pp. 179–98.
Green, F. and S. McIntosh, 2007, Is there a genuine under-utilization of skills amongst the over-
qualified? Applied Economics, 39, pp. 427–39.
Groot, W. and H. Maassen van den Brink, 2000, Overeducation in the labor market: A meta-analysis.
Economics of Education Review, 19, pp. 149–58.
Hartog, J., 2000, Over-education and earnings: Where are we, where should we go? Economics of
Education Review, 19, pp. 131–48.
Hung, C-Y., 2008, Overeducation and undereducation in Taiwan. Journal of Asian Economics, 19, pp.
125–37.
Lee, L., 1983, Generalized econometric models with selectivity. Econometrica, 51, pp. 507–12.
Lin, C. H. and C. H. Wang, 2005, The incidence and wage effects of overeducation: The case of
Taiwan. Journal of Economic Development, 30, pp. 31–47.
Persico, N., A. Postlewaite and D. Silverman, 2004, The effect of adolescent experience on labor
market outcomes: The case of height. Journal of Political Economy, 112, pp. 1019–53.
Robst, J., 2007, Education and job match: The relatedness of college major and work, Economics of
Education Review, 26, pp. 397–407.
Robst, J., 2008, Overeducation and college major: Expanding the definition of mismatch between
schooling and jobs. The Manchester School, 76, pp. 349–68.
Verdugo, R. and N. Verdugo, 1989, The impact of surplus schooling on earnings: Some additional
findings. Journal of Human Resources, 24, 629–43.

© 2014 The Authors

Asian Economic Journal © 2014 East Asian Economic Association and Wiley Publishing Asia Pty Ltd
Copyright of Asian Economic Journal is the property of Wiley-Blackwell and its content may
not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for
individual use.