Content uploaded by Seamus Mcguinness
Author content
All content in this area was uploaded by Seamus Mcguinness on Jul 03, 2018
Content may be subject to copyright.
OVEREDUCATION IN THE LABOUR
MARKET
Se
´amus McGuinness
Melbourne Institute, Melbourne
Abstract. This paper presents a review of the literature on overeducation. The
paper assesses the consistency of overeducation within the context of a number of
theoretical frameworks including Human Capital Theory (HCT) and Assignment
Theory. The analysis goes on to discuss the various measurement controversies
associated with the study of overeducation in order to provide an assessment of the
extent to which the impacts of the phenomenon represent an economic reality as
opposed to a statistical artefact. After reviewing the literature, it is concluded that
the impacts of overeducation are likely to be non-trivial and that the phenomenon
may potentially be costly to individuals and firms, as well as the economy more
generally. The existence of overeducation also raises some doubts with respect to
the validity of some of the central assumptions and predictions of HCT that are
unlikely to be fully explained by gaps in the standard wage equation framework.
Keywords. Overeducation; Returns to schooling
1. Introduction
Overeducation describes the extent to which an individual possesses a level of
education in excess of that which is required for their particular job.
1
The
phenomenon was first brought to the attention of researchers by Richard
Freeman in his 1976 study of the US graduate labour market.
2
However, interest
has mushroomed since the late 1980s as economists attempt to assess the implica-
tions of the continued rapid expansion in educational participation rates that has
become a key feature of labour market policy in most developed economies. For
instance, within the UK, it is the Government’s current stated intention that half
of all under 30s should benefit from some form of higher education. Such policies
implicitly assume that there is either unmet demand for graduate labour or
employers hiring graduates will upgrade their production techniques in order to
take advantage of a more educated labour force. However, should demand prove
insufficient or unresponsive to changes in relative supply, then workers may be
forced to take jobs for which they are overeducated.
Overeducation is potentially costly to the economy, the firm and the individual.
At a macroeconomic level, national welfare is potentially lower than would be the
case if the skills of all overeducated workers were fully utilized within the
0950-0804/06/03 0387–32 JOURNAL OF ECONOMIC SURVEYS Vol. 20, No. 3
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK
and 350 Main Street, Malden, MA 02148, USA.
economy; in addition, it may be that tax revenues are also being wasted on
equipping individuals with non-productive education. At a firm level, there is
some evidence to suggest that overeducation is associated with lower productivity.
Using firm level production and individual level employee data from companies
in the US Bell corporation, Tsang (1987) estimated that a 1-year reduction in
surplus schooling in these companies alone would increase output by more than
8% translating to an additional gain of almost $5billion. Tsang et al. (1991)
investigating the impact of overeducation on a number of productivity corollaries
(job satisfaction and intention to quit) also found that overeducated male workers
were significantly less satisfied with their jobs and more likely to quit relative to
male workers who were well matched. Higher rates of turnover amongst over-
educated workers was also reported by Alba-Ramirez (1993) and Sloane et al.
(1999), suggesting that firms hiring such workers are more likely to lose invest-
ments in training, recruitment and screening.
As individuals, overeducated workers, by virtue of the fact that a proportion of
their educational investment is unproductive, are likely to earn a lower return on
their investment relative to similarly educated individuals whose jobs match their
education. Overeducated workers may also incur non-transitory costs associated
with lower levels of job satisfaction (again see Tsang et al., 1991; Battu et al.,
2000). It is also possible that previously well-matched workers in the economy will
be ‘bumped down’ in the labour market and, perhaps out of it entirely, as over-
educated workers move into lower level occupations thus raising the mean educa-
tional level within these occupations rendering some previously adequately
educated individuals undereducated. Battu and Sloane (2000) found some bumping
down at the higher educational categories; however, there was nothing to suggest
that individuals at the lower end of the educational spectrum had been forced out
of the labour market. Finally, in relation to individual level impacts, whilst there
is little evidence to support Frank’s theory (1978) of differential overqualification
that married females in smaller labour markets are likely to be overeducated
(McGoldrick and Robst, 1996; Sloane et al., 1999; Buchel and van Ham, 2002;
Buchel and Battu, 2003), there is some UK evidence to suggest that ethnic
minorities are likely to be more severely affected. Battu and Sloane (2002) report
that, relative to whites, individuals with an Indian background were statistically
more likely to be overeducated.
However, not all economists believe that overeducation is either permanent in
nature or associated with high costs. The unwillingness of many researchers to
accept the notion is perhaps due to the fact that to do so raises some very serious
questions with respect to the validity of some of the assumptions and predictions
associated with the conventional (neoclassical) view of the labour market.
Consequently, a good deal of the debate within the literature has focused on
the extent to which the existence of overeducation represents a real challenge to
Human Capital Theory (HCT) and, related to this, the extent to which over-
education is merely a statistical artefact generated by either inadequate measure-
ment techniques or a lack of sufficient controls within the standard wage equation
framework.
388 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
This paper examines the overeducation literature in order to provide an
overview of the theoretical and empirical debates that surround the notion of
overeducation and the extent to which these stand up in light of existing evidence.
The paper is structured as follows: Section 2 provides a summary of the main
theoretical frameworks within which researchers have attempted to explain the
overeducation phenomenon. Section 3 looks at the empirical and measurement
issues in an attempt to assess the extent to which poorly defined empirical
approaches may be obscuring or overstating the extent of overeducation in the
economy and its associated costs. Section 4 reviews the international evidence on
both the incidence of overeducation and the wage costs and assesses the extent to
which studies have lent support to the different theoretical frameworks. A sum-
mary and conclusion are presented in Section 5.
2. Theoretical Frameworks
There is no accepted unified theory of overeducation, although some authors
have attempted to conceptualize and explain the problem within the framework
of semi-formal economic models (Freeman 1976; McMillen et al., 1999). However,
a significant portion of the literature on overeducation considers how the
phenomenon sits within the context of existing views of the labour market, and
quite a few studies have attempted to empirically test which theoretical perspec-
tive is most in keeping with the observed facts (Duncan and Hoffman, 1981;
Rumberger, 1987; Hartog and Oosterbrook, 1988; Groot, 1996; Sloane et al.,
1999; Battu et al., 2000; Dolton and Vignoles, 2000). In fact, Sloane (2003) argues
that the major contribution of the overeducation literature has been to widen the
debate on the importance of job characteristics in determining wages thus broad-
ening the human capital framework. The following section gives a broad overview
of the three labour market perspectives and makes an assessment of each views
consistency given the presence of overeducation in the labour market.
2.1. HCT
Becker’s (1964) monograph ‘Human Capital’ is the seminal work with respect to
this view of the labour market which represents the dominant framework adopted
for explaining the distribution of earnings within developed economies. In addi-
tion to this, as Heckman et al. (2003) points out, Mincer’s earnings model (1974)
also constitutes a cornerstone of economics as it provided the empirical frame-
work within which the central predictions of HCT can be assessed and measured.
In relation to HCT, the proposition that firms are willing to fully utilize the skills
of their workforce by adapting their production processes in response to any
changes in the relative supply of labour derives explicitly from Becker’s repeatedly
stated assumption that workers will always be paid their marginal product. Wages
will always therefore equate to the individual worker’s marginal product, which in
turn will be determined by the level of human capital that they have accumulated
through either formal education or on-the-job training. Thus overeducation,
OVEREDUCATION 389
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
which is associated with worker under-utilization and wage rates below the
marginal product, would appear entirely inconsistent with this view of the labour
market. However, as will become apparent, some economists have continued to
argue that HCT remains fully consistent despite the existence of overeducation
(see Section 4).
The overeducation phenomenon does not necessarily overturn HCT as it is
entirely plausible that workers will be overeducated in the short run, whilst firms
adjust their production processes in order to fully utilize the individuals’ human
capital or alternatively for as long as it takes workers to find a more appropriate
match through job search. HCT can therefore be rationalized by allowing for the
existence of short-run disequilibria. Of course, should overeducation prove to be
a non-transitory phenomenon that persists in the long-run, this explanation will
not ‘save’ HCT.
The empirical framework adopted for testing HCT may also provide an
explanation for overeducation which might make it entirely consistent with the
neoclassical view. The standard approach was developed by Mincer (1974) and is
based on an earnings regression centred around years of schooling (see Section 3
below), and therefore, less formal measures of human capital such as on-the-job
training, which Becker argues is directly substitutable with schooling, are ignored.
Thus, individuals with more schooling may be compensating for a lack of work-
related human capital, and the apparent lower earnings of these ‘overeducated’
may be attributable to an omitted variables problem, i.e. a lack of controls for less
formal measures of human capital accumulation. The problem is illustrated
graphically in Figure 1 which shows that different combinations of experience
and education produce workers of similar productivity by an isoquant. Thus,
relative to the average level of qualification Q*, some workers appear
Q*
Informal human ca
p
ital
Formal human capital
Figure 1. Human Capital Trade-Off.
390 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
overeducated (and underpaid in a model without experience controls), whilst
some appear undereducated when in fact they all possess the same amounts of
human capital and earn a wage consistent with it.
Finally, it may also be the case that overeducated workers are in some way less
able relative to their adequately matched counterparts; therefore, lower wages are
merely a reflection of lower ability/productivity. Should the empirical framework
prove inadequate for picking up any such skill differences that might exist, this
will also introduce bias into the estimated wage effects of overeducation.
Therefore, to summarize, whilst some authors argue that overeducation suggests
that HCT is not consistent with the observed facts [Dolton and Vignoles (2000)],
this is only true should overeducation prove to be a long-run phenomenon and/or
persist when controls are included for work-based human capital investments
and/or worker skill heterogeneity.
2.2. The Job Competition Model
Some economists question the ease with which firms can adjust their production
techniques to facilitate changing factor input prices (Duncan and Hoffman, 1981;
Hartog and Oosterbeek, 1988) especially where firms have technology geared for
team working which brings together groups of heterogeneous workers. Also,
institutional arrangements (national pay agreements etc.) may be far more rigid
than suggested by HCT. If firms cannot adapt quickly (or at all) then individual’s
productivity and hence their earnings will also depend on their jobs. Indeed, one
particular framework, the Job Competition Model, suggests that job character-
istics may be the only factor determining earnings. The Job Competition Model,
which is based on Lester C. Thurow’s (1975) book entitled ‘Generating Inequality’,
has attracted considerable attention within the overeducation literature.
Thurow’s model characterizes a market within which individuals compete for
job opportunities based on their relative training costs, as opposed to competition
based on the wages individuals are willing to accept given their human capital.
However, Thurow does not propose that his model adequately describes all
labour market behaviour; instead he argues that the job and wage competition
models can, and probably do, co-exist as market-clearing mechanisms. The
central element of the Job Competition Model is based around the observation
(Thurow cites US surveys) that the majority of workplace skills are acquired
through on-the-job training as opposed to formal education. Thus, the labour
market is not a bidding market for selling existing skills but a training market
where training slots must be allocated to different workers. How these training
slots are distributed across individuals depends on factors determining where
individuals are located within a particular job queue and the distribution of
jobs (training opportunities) in the economy. Once individuals reach the top of
the queue and are allocated a job, their wage will be pre-determined by the
characteristics of the job in question. Thus, the marginal product resides in the
job rather than the individual’s characteristics.
OVEREDUCATION 391
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
The Job Competition Model emphasizes the importance of a person’s relative
position. Thurow postulates that were an individual to observe his neighbour
participating in education, then under the HCT framework that individual would
be less likely to participate in education as supply would be higher and the return
less. However, under the Job Competition Model, the same individual would now
be more likely to participate as education is a defensive necessity, necessary to
protect their place in the queue. The larger the numbers of educated persons in
the economy the more imperative for individuals to invest in education. The Job
Competition Model therefore provides a clear explanation for educational over-
investment and thus overeducation. In many ways the model is very similar to the
signalling framework in that individual investments are motivated by preserving
ones position, although in the Spence (1973) model there is a limit to the amount
of education an individual will invest in (based on the balance between earnings
and the cost of education). However, it is difficult to determine how a ceiling is
reached within the Job Competition framework with respect to educational
participation. Nevertheless, it provides a theoretical framework with which over-
education is entirely consistent. The implications of the Thurow model are that
wages will be wholly dependent upon required education and that the returns to
education that are over and above that required by the job (surplus education)
will be zero.
2.3. Assignment Models
A middle ground between what could be argued as the two previous extremes is
found within the assignment literature. Despite some differences, assignment
models all specify the jobs or sectors available to workers, the relevant differences
amongst workers, the technology relating job and worker characteristics to out-
put and the mechanisms that assign workers to jobs. Within this framework, the
earnings function is no longer a directly observable relationship but instead it is
the equilibrium outcome to the solution of the assignment problem. Sattinger
(1993) points out that relative wages have been changing over time with earnings
becoming more unequal. He argues that these changes are hard to explain within
the standard neoclassical framework in which the productivity and earnings are
exclusively linked to education and experience and thus independent of the
availability and/or quality of jobs in the economy.
Whilst there are various assignment interpretations of the labour market that
are of interest in their own right, their general predictions with respect to the
allocation of workers to jobs and their subsequent earnings are more important
within the context of overeducation. Assignment models differ significantly from
the Job Competition interpretation in that they stress that choice of job or sector
creates an intermediate step between an individual’s characteristics and their
earnings, i.e. the job allocation process is not merely a lottery. Income maximiza-
tion guides workers to choose particular jobs over others. Thus, higher wages for
workers with some characteristics play an allocative role in the economy rather
than simply being rewards for the possession of particular characteristics.
392 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Workers found in a particular sector (or job) are not randomly distributed but are
there based on the choices made to maximize their income or utility. Thus, the
central and crucial prediction arising from the assignment literature is that in
order to adequately explain changes in the distribution of earnings, we must give
some consideration to both individual and job characteristics. Thus, overeduca-
tion is entirely consistent with an Assignment Interpretation suggesting
that marginal product and thereby earnings will depend to some extent on both
the individual and the job; these models also imply that there is no reason to
expect that wage rates will be wholly related to acquired schooling or other
individual attributes (Human Capital Model), neither should we expect that
wage rates will be wholly related to the nature of the job (Job Competition
Model).
2.4. The Influences of Gender and Space
Some explanations for overeducation have been put forward that, at first glance,
appear to be largely unrelated to any major theoretical framework, in particular,
a number of recent studies have examined the evidence in support of spatially
based explanations for overeducation. However, that is not to say that spatial and
theoretical considerations are necessarily independent. In the same way that
Thurow hypothesized that different frameworks might co-exist within different
markets, it is also plausible to suggest that variations in labour market character-
istics at either national or regional level may also determine, or be determined by,
the framework that prevails. Nevertheless, it must be pointed out that, to date,
most of the research has focused on the importance of various spatial character-
istics with very little attempt made to conceptualize the results within the context
of the principal theoretical approaches. However, given the weight of literature on
the issue, it certainly cannot be ignored.
Frank (1978), was the first economist to examine the link between overeduca-
tion and spatial factors; however, this was related to gender-segmented labour
markets in dual earner households. In this context, the husbands’ job seeking
takes place in a global market, whereas the wives are largely restricted to a
regional labour market exposing her to a higher relative risk of overeducation.
This hypothesized inverse relationship between gender-based overeducation prob-
abilities and labour market size is termed ‘differential overqualification’.
McGoldrick and Robst (1996) sought to further develop the relationship between
space and overeducation by testing the proposition that geographic restrictions
affect the labour market outcomes of all workers (as opposed to married women
only). In a similar vein, Buchel and van Ham (2002) argue that employment
opportunities will be determined at regional level as most people will tend to look
for work on the local (regional) labour market due to limited spatial flexibility.
They argue that job availability is determined by two factors, firstly, the location
of their residence in relation to the spatial configuration of employment oppor-
tunities and secondly, the size of the labour market that can be searched from
OVEREDUCATION 393
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
their residence. The size depends on the spatial flexibility (commuting tolerance)
of the job seeker in terms of their commuting and migration tolerances.
3. Empirical Measurement and Methodological Issues
As was noted, there are a number of empirical considerations relating to the
measurement of both the incidence and wage effects of overeducation that are of
crucial importance in assessing the true significance of the phenomenon with
respect to both the economy and education policy more generally. The central
empirical debates centre on the precision and consistency of the various measure-
ment approaches and the lack of sufficient controls within the wage equation
framework.
3.1. The General Empirical Framework
The typical specification adopted is based around the standard Mincer relation-
ship (Mincer, 1974) which itself has its origins in Becker (1964). The typical
human capital model assumes that an individual will participate in schooling up
until the point where the present value of the S
th
year of schooling just equals the
cost, i.e.:
XTs
t¼1¼WsWs1
ð1þrsÞt¼Ws1þcsð1Þ
if Tis sufficiently large and c
s
sufficiently small we can rearrange the equation to
get:
rsffiWsWs1
Ws1
ð2Þ
Thus, the internal rate of return to schooling r
s
can be approximated by log
W
s
log W
s1
implying that we can estimate the return to schooling by seeing
how log wages vary with schooling. This is the basic premise of the Mincer
regression, which is generally written, in the following form:
log w¼x1þ2Sþ3Ex þ4Ex2þ"ið3Þ
where xis a vector of personal characteristics correlated with earnings, Sis years
schooling and Ex is experience. The general educational mismatch specification
varies slightly from this in that years schooling is decomposed into required,
surplus and deficit education so that the model is now written as:
log w¼x1þ2Srþ3Soþ4Suþ5Ex þ6Ex2þ"ið4Þ
where S
r
is years of required schooling (the methods of defining this required
level for the job is discussed below), S
o
is years of surplus schooling above the
required level (overeducation) and S
u
is years of deficit schooling below the
required level (undereducation). Many overeducation studies will estimate both
394 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
equations 3 and 4 to allow comparisons between the returns to acquired, required,
surplus and deficit schooling.
3.2. The Use of Dummy Variables
However, in some instances, the presence of overeducation is indicated through
the presence of dummy variable on the basis that a decomposition of acquired
schooling into required, deficit and surplus elements is not possible. In such
instances, the educational mismatch model in equation 4 is modified slightly to
become:
log w¼x1þ2Doþ3Duþ4Ex þ5Ex2þ"ið5Þ
Where D
o
and D
u
are dummy variables indicating if the individual is over-
educated or undereducated, obviously in the case of an exact match both vari-
ables will equal zero. The benchmark with which individuals are being compared
will vary depending on whether specification 4 or 5 is being estimated. When
overeducation is measured in terms of years (equation 4) then the overeducated
are being compared to people doing the same job who are not overeducated and
who have less education. In the majority of studies that utilize equation 4, the
overeducated have been found to earn a return to surplus education which is
generally positive but less than the return to required education. When dummy
variables are used (equation 5), the overeducated are being compared to people
with the same education who are adequately matched, and the coefficient on
overeducation is generally negative suggesting they earn less than their compar-
ably educated counterparts who are appropriately matched. Cohn and Kahn
(1995) demonstrate the potential dangers of misinterpretation by replicating the
results of Verdugo and Verdugo (1989) who interpreted a negative coefficient on
the overeducation dummy variables to indicate a negative return to overeduca-
tion. Using the Verdugo and Verdugo data, Cohn and Kahn estimate overeduca-
tion models under both specifications (4 and 5). They demonstrate that a negative
estimate on the overeducation dummy does not necessarily imply a negative
return to years of overeducation. Cohn and Kahn (1995) in fact found that
overeducated workers earn less than persons with similar levels of schooling who
are not overeducated (equation 5) but report that the return to overeducation is
positive relative to workers in similar jobs who are well matched (equation 4) thus
demonstrating that Verdugo and Verdugo had misinterpreted their results.
3.3. Subjective and Objective Measures of Overeducation
There are four basic approaches to measuring required education for a job and
hence overeducation, two subjective measures and two objective measures.
Realistically speaking, the choice is typically restricted by data availability; never-
theless, there is a growing literature centred on assessing the levels of consistency
and potential biases associated with the various approaches.
OVEREDUCATION 395
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Overeducation can be assessed subjectively by asking the respondent to give
information on the minimum requirements of the job and then comparing this
with the individual’s acquired education or by simply asking the respondent
whether or not they are overeducated. Overeducation can also be assessed
objectively by using information provided by professional job analysts (such as
in the Standard Occupational Classification System in the UK or the Dictionary
of Occupational Titles in the US) to determine an individuals required education
on the basis of their job title and again comparing this with their actual level
of education. A second objective measure of overeducation is obtainable by
calculating the mean education level for a range of occupations with an individual
defined as being overeducated if they were more than one standard deviation
above their occupation’s mean education level.
Subjective measures of overeducation have been criticized on a number of
grounds, firstly overeducated workers may be less likely to respond to question-
naires due to higher levels of job apathy which may lead to an underestimation
of the incidence of overeducation. Secondly, workers in smaller and/or less-
structured organizations may lack sufficient benchmarks against which to assess
their job requirements, a factor, which will again lead to measurement error.
Finally, even where benchmarks are available, respondents may be applying
differing criteria when assessing their job requirements, i.e. the actual level of
education required to do specific tasks or the formal educational requirements
necessary to get the job. However, on this last point, Green et al. (1999) using an
alumni dataset from Newcastle University in the UK found that in the vast
majority of cases, the assessment of the education levels needed to do the job
tended to match those needed to get the job suggesting a broad level of consis-
tency between both subjective approaches.
The occupational dictionary-based objective measure is open to the criticism
that occupations may contain a number of skill levels, so that in fact people with
the same job titles may be doing very different jobs, for instance, the tasks
undertaken by managers are likely to vary widely. Also, rising education levels
in the economy imply that employers will allocate workers differently. For
example, Mason (1996) reports that managers are now employing university
graduates in mid-clerical positions, posts traditionally held be persons educated
to O and A level standard (predominately high school graduates). Thus, the
educational requirements of various occupations will evolve with changes in
relative supply, a factor not always readily incorporated into occupational
classification systems which tend to be relatively static in nature. The standard
deviation-based objective measure of overeducation has also been criticized because
of the arbitrary nature of the choice of cut off points; also, if a particular
occupation contains a high proportion of overeducated workers, this will raise
the occupational average and corresponding cut-off point thus underestimating
the true level of overeducation. The assumption of symmetry that the standard
deviation method implies seems unrealistic. In addition, the use of the occupational
mean has also been criticized with a number of studies (Kiker et al., 1997; Mendes
de Oliveria et al., 2000) arguing for the superiority of a modal-based measure.
396 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
3.4. Is There Consistency Across the Various Definitions?
A number of studies have sought to assess the correlation between the various
measures of overeducation and/or the extent to which they generate varying
estimates of the incidence of overeducation and/or the returns to overeducation.
Taking these studies chronologically, McGoldrick and Robst (1996) found that
the incidence of male overeducation varied according to the definition employed,
over 50% were overeducated under the objective occupational dictionary-based
measure, 30% under the subjective measure and just 16% under the objective
standard deviation approach. Battu et al. (2000) take the analysis a step further
by examining the level of correlation between two subjective and an objective-
based measures derived from two separate panels of data (1985 and 1990). The
subjective measures are based on respondents’ opinions if (1) a degree was a
requirement in the job specification and (2) the level of satisfaction with the match
between the individuals, work and qualifications, whilst the objective measure is
based around an occupational dictionary approach. The authors found that the
various measures tended to identify different people as being overeducated, for
instance, the correlation for overeducated males under the two subjective measures
was 21 and 31% respectively, whilst the correlation between the objective
and subjective measures ranged between 20 and 33% for males. The rates of
correlation were somewhat higher for females; however, in no case did they
exceed 50%. Despite poor correlation, the authors report that the various
approaches generate similar results with respect to the effect of overeducation
on earnings suggesting that the wage equation estimates may be picking up
factors other than overeducation such as unobserved job and/or personal char-
acteristics. The implication of this is that the more poorly specified the initial
model the potentially more distorted the overeducation influence may be, thus the
problem associated with the lack of a consistent definitional approach is poten-
tially compounded by a tendency amongst researchers to examine the issue within
the context of models omitting sufficient controls for job and/or individual
characteristics. This point is illustrated in Table 1 which estimates the overeduca-
tion wage penalty under a number of specifications using data from a cohort of
Northern Ireland University graduates collected in 1999 (see McGuinness 2003b
for a full description of the dataset used). Specification 1 is a basic Mincer model,
3
specification 2 is expanded to include faculty controls whilst the final specification
includes personal and job characteristics including A level point score which, to
some extent at least, will measure ability heterogeneity amongst the graduates. It is
obvious from the results that the more poorly specified the model the more upwardly
biased the overeducation penalty will be. The results demonstrates the importance of
including job characteristics and some form of ability heterogeneity control when
attempting to estimate the relationship between overeducation and wages.
Groot and van den Brink (2000a) also measure the correlations between two
objective (standard deviation and occupational dictionary measures) and one
subjective-based measures (overeducated yes/no response). They report that the
incidence of male overeducation varied from 8.7 to 12.3%; however, in this case,
OVEREDUCATION 397
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
the subjective-based measure yielded the lowest estimate. In relation to the extent
to which the definitions identified the same individuals as being overeducated, the
results were again less than convincing. Groot and van den Brink (2000a) also
sought to assess the extent to which the alternative definitions generated varying
outcomes within the wage equation. To some extent, their results contradict the
findings of Battu et al. (2000) who report that the various definitions generate
broadly similar estimates within the wage equation framework. However, Groot
and van den Brink’s findings are somewhat less convincing due to the counter-
intuitive nature of some of their results.
4
Whilst the previous studies indicate the extent to which the various definitions
tend to identify different people as being overeducated and generate different
estimates of the incidence of and returns to overeducation within the context of
the same dataset, they provide no indication of which measure is closest to the
true incidence or the extent to which any particular approach generates biased
estimates. To a large extent, the level of correlation is likely to vary according
to the dataset being used and the institutional/economic arrangements of the
country in question. The issue of empirical bias associated with the various
definitions was more effectively addressed by Groot and van den Brink (2000b)
who conducted a cross-country meta-analysis of 25 studies, utilizing the various
subjective and objective methodologies. The authors found in a preliminary
Table 1. Graduate Overeducation Estimates for Northern Ireland Males.
Specification 1 Specification 2 Specification 3
Constant 11.716 (0.554)† 11.345 (0.565)† 11.039 (0.536)†
Human capital
Acquired schooling 0.100 (0.029)† 0.082 (0.030)† 0.066 (0.028)*
Experience 0.027 (0.018) 0.016 (0.018) 0.006 (0.017)
Experience squared 0.001 (0.001) 0.001 (0.001) 0.000 (0.001)
Faculty
Arts 0.173 (0.090)* 0.178 (0.080)*
Medical and Med related 0.212 (0.079)* 0.187 (0.078)*
Science 0.024 (0.081) 0.019 (0.076)
Business 0.074 (0.066) 0.087 (0.063)
Math\Eng\Tech 0.065 (0.060) 0.011 (0.056)
Combined 0.097 (0.069) 0.133 (0.064)*
Other
Work Northern Ireland 0.224 (0.036)†
Public sector 0.177 (0.043)†
Married and Children 0.121 (0.924)
A level score 0.008 (0.002)†
Overeducation 0.175 (0.45)† 0.134 (0.045)† 0.111 (0.042)†
R
2
0.063 0.122 0.278
F-Statistic 6.39† 5.42† 10.27†
* 95% significant
† 99% significant
398 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
examination of the data that the standard deviation-based measure again tended to
yield the lowest estimate of the incidence of overeducation. This finding was verified
within an OLS framework which estimated the ‘true’ incidence of overeducation to
be 26% and that the standard deviation approach tended to reduce this ‘true’
incidence by 12% points. The finding of a lower incidence under the standard
deviation approach is, of itself, not surprising, as the methodology requires educa-
tion levels to be at least one standard deviation above the mean before overeducation
is determined whilst the other approaches have no such requirement. In relation to
the wage equation meta-analysis, the study reported that the ‘true’ return to a year of
required education was 7.9%, a year of surplus education 2.6% and a year of deficit
education, 4.9%. The authors did not find any of the methodological approaches
to significantly influence estimated returns, suggesting that potential definitional
biases may be restricted to estimates of the incidence of overeducation and limited to
the objective standard deviation-based measure. A more recent meta-analysis by
Rubb (2004) also found that neither the subjective nor occupational dictionary
approaches yielded estimates of the overeducation wage effect that were significantly
different from a measure based on the mean occupational level. Thus, one might
reasonably conclude from such cross-country studies that whilst there are serious
concerns relating to the low correlation between the various measures of overeduca-
tion, with the exception of the standard deviation measure, the balance of the
evidence would suggest that, in terms of estimating the incidence and returns to
overeducation, the various approaches generate broadly consistent evidence.
5
3.5. How Adequately do Existing Measures take Heterogeneity into Account?
There are significant questions relating to the assumed relationship between
education and skills that is embodied within the various approaches to measuring
overeducation. Each approach assumes that persons acquiring the same years of
education or possessing similar credentials shall have broadly similar skills. By
not allowing for heterogeneity across the skill sets of persons with similar educa-
tional backgrounds, the various measurement approaches may be providing
inaccurate measures of the incidence and labour market effects of overeducation.
For instance, in relation to the graduate labour market, Chevalier (2003) argues
that widening access to higher education has increased the heterogeneity of
graduates through lower ability students accessing higher education and an
increase in student-staff ratios. To some extent, the unobserved heterogeneity
will relate to differences in the skills that exist amongst individuals with similar
levels of education, in other aspects it will refer to ability levels. For instance,
between 1994 and 2002, the number of new graduates entering the labour market
has increased by 23% for males and 53% for females (O’Leary and Sloane, 2005).
This raises questions with regard to the extent to which a degree is any longer a
good signal of ability as one would assume that the rise in higher education (HE)
participation levels must have led to a greater spread in the ability levels of
graduates entering the labour market. Cohort-related effects can also potentially
result in significant differences across individuals with similar levels of education.
OVEREDUCATION 399
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Green et al. (2002) highlight potential heterogeneity effects that may arise because
of grade drift, although they found little evidence to support it in a UK context.
Grade drift describes a drop in educational standards implying that the level of
human capital accumulation associated with various credentials has fallen over
time. Grade drift will be evident if, ceteris paribus, employers are found to be
increasing educational requirements for younger workers. The concept of grade
drift is related to that of heterogeneity as it once again suggests that individuals
with similar levels of education potentially have significantly different ability
levels. The possibility of grade drift and other non-time-dependent causes of
worker heterogeneity, such as highlighted by Chevalier (2003), demonstrate the
potential importance of including adequate controls in the estimation framework
that allow for variations in ability levels across apparently similar individuals.
To date, those studies that have attempted to control for heterogeneous skill
effects have done so by using models that allow for some variability in worker
characteristics by controlling for unobserved effects; these studies have concluded
that when account is taken of skill differences across individuals with similar
levels of education, the wage penalty associated with overeducation largely
disappears (Bauer, 2002; Chevalier, 2003; Frenette, 2004). However, such studies
appear to assume that all unobserved individual differences relate only to skill
and thus ignore the impact of other personal/job characteristics that may have
been omitted from their initial models. Such methodologies appear to implicitly
accept the notion that wages will be fully determined by an individual’s level of
acquired human capital. However, should it transpire that HCT is rejected in
favour of a Job Competition or an Assignment Interpretation of the labour
market implying that wage rates will be determined, either in part or in full, by
job characteristics, then it is likely that such methodologies will overstate the
impact of heterogeneous skill effects on the wage distribution. McGuinness
(2003a) sheds some light on the issue by deriving an explicit self-assessed
subjective measure of skill based on the individual’s average competency across 16
areas related to job performance. When this explicit control for skill heterogeneity
was included in the wage equation, the results indicate that the impact of the
effect was much smaller than that reported by those authors adopting the more
implicit approach. Heterogeneous skill effects were found to account for less than
5% of the variance in wages, and it was reported after accounting for such effects,
overeducated graduates still earned wages that were much lower than their
well-matched counterparts. McGuinness and Bennet (2006) using quantile regres-
sion techniques to control for unobserved skill heterogeneity reported that over-
education can impact male graduates of both low and average ability and female
graduates of all ability levels, thus demonstrating that it is not sufficient to
characterize overeducation as merely reflecting low ability levels. Finally,
evidence from a UK study by Green et al. (1999) suggests that heterogeneity-
linked biases in the estimates of the incidence of overeducation are not likely to be
substantial. The study concentrated on mismatches between required and
acquired skill levels, as opposed to years of education or credentials and thus
was immune from biases associated with the assumed uniformity in the link
400 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
between educational attainment and productivity. The authors found that the
determinants of overskilling were similar to those driving overeducation and that,
as is the case with overeducation, overskilling has a significant negative impact on
wages.
3.6. Empirical Overview
There are certainly a number of empirical issues that any researcher embarking on
a study of overeducation should be aware of including the variety of approaches
to measuring the phenomenon and the potential biases that lie within these.
However, on balance, and despite concerns relating to poor correlations between
the various approaches, there is no consistent evidence to suggest that any of the
subjective or occupational dictionary-based measurement frameworks result in a
systematic and significant underestimate of either the incidence, or wage effects,
associated with overeducation. Those authors that question the extent to which
existing measurement frameworks or modelling approaches allow for hetero-
geneous skill levels amongst persons with similar levels of education are certainly
making valid points when they assert that failure to include such controls will
tend to overestimate both the incidence and effects of overeducation.
Undoubtedly, as Sloane (2003) points out, much of the effects attributed to
overeducation will in fact relate to unobserved skill heterogeneity; nevertheless,
it would appear that perhaps too much weight is being attributed to the share of
the unexplained component that is accounted for by differences in skill within
models that attempt to control for such effects implicitly. As an alternative to
such approaches, researchers should perhaps attempt to incorporate, where
possible, some explicit measures of skill into their analysis.
4. Results
This section will examine the evidence from existing studies to determine trends in
the incidence of overeducation, the estimated effect of overeducation on wages
and which, if any, of the principal theoretical frameworks discussed earlier are
most consistent with the observed facts. It should be noted that the analysis
attempts to identify patterns from the general literature; however, it must be
borne in mind that the studies are quite heterogeneous in nature as the cut across
countries, time and labour markets, nevertheless, bearing this caveat in mind,
some insightful patterns do emerge.
4.1. The Incidence of Overeducation
The various incidence estimates from the raft of overeducation studies are
summarized in Table 2. In total, 33 studies generated 62 estimates with the
number of subjective and objective estimates relatively evenly split at 34 and 28
respectively.
6
It is not easy to disentangle the information to derive any discernible
patterns; however, some observations are possible. In line with the findings from
the Groot and van den Brink (2000b) meta-analysis, objective-based approaches
OVEREDUCATION 401
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Table 2. Reported Incidences of Overeducation.
Name of study Year Definition employed Country Data collection
Incidence of
overeducation
Dolton et al. 2000 Self assessment – subjective job
requirement
UK 1986 30%
Chevalier, A 2003 (1) Objective occupational
dictionary-based measure
UK 1996 17%*
(2) Subjective – job requirement 1996 32.4%*
(3) Subjective – satisfaction 1996 16.2%*
Sloane et al. 1999 Subjective – job requirements Britain 1986–1987 30.63%
Battu et al. 2000 (1) Subjective – satisfaction UK 1996 40.4%*
1996 40.7%*
(2) Objective occupational
dictionary-based measure
1996 21.75%*
1996 22.15%*
(3) Subjective – degree requirements 1996 33.65%*
1996 38%*
Groot 1996 Objective standard deviation-based
measure
UK 1991 11%
Dolton and Siles 2003 Subjective – not clear which
one they used
UK 1998 22%
Daly et al. 2000 Subjective – required education US 1976 37.65%*
US 1985 32.65%*
Germany 1984 17.5%*
Duncan and Hoffman 1981 Subjective – job requirements US 1976 42%
Sicherman 1991 Subjective – job requirements US 1976 and 1978 40.8%
Groot and van den Brink 2000b Meta-analysis 26.2%*
Hartog and Oosterbeek 1988 Subjective – required education Holland 1982 16%
objective 1960 7%
1971 13.6%
1977 25.7%
McGoldrick and Robst 1996 (1) Subjective – job requirements US 1985 50%*
402 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
(2) Objective occupational
dictionary-based measure
1985 30.8%*
(3) Objective standard
deviation-based measure
1985 12.55%*
Verdugo and Verdugo 1989 Objective standard deviation-based
measure
US 1980 10.9%†
Cohn and Kahn 1995 (1) Objective standard
deviation-based measure
US 1985 13%
(2) Subjective measure – Sicherman
definition
1985 33%
Tsang et al. 1991 (1) Subjective – required education US 1969 35%
1973 27%
1977 32%
(2) Objective occupational
dictionary-based measure
1973 57%
Patrinos 1997 Objective standard
deviation-based measure
Greece 1977 16%
Alpin 1998 (1) Objective occupational
dictionary-based measure
UK 1995 27%
(2) Objective standard deviation
modal measure
1995 37.7%
Rumberger 1987 (1) Subjective – job requirements US 1969 35%
1973 27%
1977 32%
(2) Objective occupational
dictionary-based measure
US 1973 57%
Green et al. 1999 (1) Subjective UK 1995 27.4%
(2) Subjective UK 1997 32%
Alba- Ramirez 1993 Subjective – job requirements Spain 1985 17%
Groot and Maassen
van den Brink
2000 (1) Objective standard deviation-based
measure
Holland 1994 11.85%*
(2) Objective occupational
dictionary-based measure
1994 15.9%*
(3) Subjective 1994 11.15%*
(continued)
OVEREDUCATION 403
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Table 2. Continued.
Name of study Year Definition employed Country Data collection
Incidence of
overeducation
Bauer 2002 Objective standard deviation-based
measure
Germany 1984–1998 11.5%*
Decker et al. 2002 Objective occupational
dictionary-based measure
Holland 1992 30.6%
Buchel et al. 2002 Subjective – job requirements Germany 1998 15.8%*
Robst 1995 Subjective – required education US 1976, 1978 and 1985 44.68%
Allen et al. 2001 Subjective – required education Holland 1998 14%
Vahey 2000 Subjective – required education Canada 1982 31%*
Hersch 1995 Subjective – job requirements US 1991 21%
Kiker et al. 1997 (1) Objective standard deviation
mean measure
Portugal 1991 9.4%
(2) Objective standard deviation
modal measure
1991 25.5%
(3) Objective occupational
dictionary-based measure
1991 33.1%
Cohn and Ng 2000 Objective modal measure Hong Kong 1986 35%*
1991 34%*
Hannan et al. 1998 Objective standard deviation Ireland 1992 20%
Groot 1993 Objective standard deviation-based
measure
Holland 1983 16.1%
Burris 1983 Objective occupational
dictionary-based measure
US 1977 21.7%
McGuinness 2003a Subjective – job requirements Northern Ireland 2000 20%
McGuinness 2003b Subjective – job requirements Northern Ireland 1999 24%
*Average.
†Males only.
404 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
were found to generate lower estimates with the mean objective-based incidence
standing at 22%, some 7% points below the comparable subjective figure. In
relation to cross-country comparisons, Holland yielded the lowest incidences
under both subjective and objective measures (Hartog and Oosterbeek, 1988;
Groot and van den Brink, 2000b), whilst studies of the US labour market
generated the highest incidences irrespective of the measurement approach
adopted (Tsang et al., 1991; McGoldrick and Robst, 1996).
In order to determine whether the overeducation phenomenon may be becoming
more important over time, Figure 2 plots subjective estimates against the year in
which the relevant data were collected. There are no indications that the incidence
of overeducation has been rising over time; in fact, fitting a linear time trend to
the observations is suggestive of a slight decrease. Nevertheless, given the proble-
matic nature of the data, it would be foolish to attach too heavy a weight to the
very slight negative slope of the best fit line. This is confirmed by plotting the
objectively based incidences over time and finding a slightly positively sloped
trendline (Figure 3). Given Figures 2 and 3 and Tables 2 and 3, it is probably
reasonable to conclude, on the basis of the graphical and tabular evidence, that
the incidence of overeducation has remained relatively stable across time.
4.2. Wage Rates
Twenty-one studies provide estimates of the wage effects of overeducation on
earnings; of these, 10 estimate a model based on the decomposition of earnings
only (equation 4), seven estimate an overeducation penalty derived from a
dummy variable model (equation 5), with four studies estimating both specifica-
tions. The vast majority of the 14 studies estimating a model based on equation 4
report a pattern consistent with the positive return to attained education being
0%
10%
20%
30%
40%
50%
60%
1965 1970 1975 1980 1985 1990 1995 2000 2005
Year
Incidence
Figure 2. Subjectively Measured Incidences of Overeducation.
OVEREDUCATION 405
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
less than the positive return to required education being greater than the positive
return to overeducation being greater than the negative return to undereducation
(Table 2). The only exceptions to this pattern are studies of the Dutch labour
market by Groot (1993), Groot and van den Brink (2000a) and the UK labour
market by Groot (1996) who reported a negative return to a year of over-
education and a positive return to a year of undereducation. However, one
must conclude that the atypical nature of these results is most likely related to
the slightly unusual methodological approach adopted, i.e. the estimation of
equation 4 on the basis of an objective standard deviation approach to measuring
overeducation. In relation to the studies estimating equation 5, the penalty
associated with being overeducated is estimated to lie within a range of 8to
27% with a mean penalty of 15.3%. A final point to note is that the assumed
exogeniety of overeducation could result in a significant under-estimate of its
wage impacts; Dolton and Siles (2003) found that after controlling for endo-
geniety within the context of a treatment model, the overeducation pay penalty in
current employment rose substantially.
Generally speaking, the evidence on wages is in line with assignment theory
given that the lower returns to surplus education and the overeducation penalty
suggest that the overeducated are working below their potential but are deriving
some benefit from surplus education; these lower returns are consistent with a
scenario in which overeducated workers jobs are imposing an upper limit on the
extent to which they can utilize their skills with this productivity ceiling reflected
in lower wages. The evidence is certainly not consistent with HCT which suggests
that the returns to surplus and required education should be equal and that the
overeducated should not have any pay penalty inflicted upon them. Nor is it
consistent with Thurow’s Job Competition Model which suggests that the return
to surplus education is zero.
0%
10%
20%
30%
40%
50%
60%
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000
Year
Incidence
Figure 3. Objectively Measured Incidences of Overeducation.
406 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Table 3. Reported Wage Effects.
Name of
study Year Definition employed Country
Data
collection
Return to
acquired
education
Return to
required
education
Return to
under
education
Return to
overeducation
Return to
work
experience Penalty
Dolton and
Vignoles
2000 Self assessment –
subjective job
requirement
UK 1986 11.64%
Chevalier 2003 Objective occupational
dictionary
UK 1996 27%
Sloane et al. 1999 Subjective job
requirements
Britain 1986–87 3.4% 2.8%
Groot 1996 Objective occupational
dictionary
UK 1991 5.2% 8.8% 11.9% 2.8%
Dolton and
Siles
2003 Subjective UK 1998 18%
Daly et al. 2000 Subjective – required
education
US 1976 7.55%* 7.55%* 3.5%* 5.3%* 4.35%*
US 1985 8%* 9.35%* 2.05%* 7%* 5.4%*
Germany 1984 8.2%* 9%* 5.8%* 5.75%* 3.5%*
Duncan and
Hoffman
1981 Subjective job
requirements
US 1976 7.35%* 7.7%* 2.8%* 4.05%*
Sicherman 1991 Subjective job
requirements
US 1976 and
1978
4.8% 1.7% 3.9%
Groot and
van den Brink
1997 Meta-analysis 7.9% 4.9% 2.6%
Hartog and
Oosterbeek
1988 Subjective – required
education
Holland 1982 6.1% 7.1% 2.5% 5.7%
Verdugo and
Verdugo
1989 Objective standard
deviation based measure
US 1980 13%
Cohn and
Kahn
1995 Objective standard
deviation
US 1985 8.4% 4.4% 5.9% 11.1%
(continued)
OVEREDUCATION 407
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Table 3. Continued.
Name of
study Year Definition employed Country
Data
collection
Return to
acquired
education
Return to
required
education
Return to
under
education
Return to
overeducation
Return to
work
experience Penalty
Subjective – sicherman
definition
US 1995 8.2% 3.9% 5.9% 11.1%
Patrinos 1997 Objective standard
deviation
Greece 1977 9.4%
Rumberger 1987 Subjective job
requirements
US 1973 7.6%* 4.4%*
Objective
occupational dictionary
US 1973 8.8%* 4.45%*
Alba- Ramirez 1993 Subjective job
requirements
Spain 1985 7.4% 9.2% 6% 4%
Groot and
van den Brink
2000a Objective standard
deviation Based
Holland 1994 5.95%* 3%* 1.9%* 11.95%*
Objective
occupational Dictionary
Holland 1994 5.4%* 10.8%* 2%* 10.9%*
Subjective Holland 1994 7.4%*
Bauer 2002 Objective
standard deviation
Germany 1984–1998 11.6%* 10.75%* 7.1%* 12.85%*
Kiker et al. 1997 Objective standard
deviation based
Portugal 1991 8%
Objective standard
deviation modal
Portugal 1991 7.6% 5.6% 4.8%
Objective
occupational dictionary
Portugal 1991 10.1% 4.8% 3.3%
Cohn and Ng 2000 Objective modal
measure
Hong Kong 1986 10%* 5%* 4.5%* 17%*
Hong Kong 1991 14%* 4.5%* 4.5%* 26%*
Groot 1993 Objective standard
deviation
Holland 1983 5.5% 2.6% 7.4%
McGuinness 2003a Subjective – job
requirements
Northern
Ireland
2000 34%
*Average.
408 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
4.3. Which Theoretical Framework is Most Consistent with Overeducation?
A considerable amount of effort has been expended by researchers who have
attempted to explain the overeducation phenomenon within the context of the
various views of the labour market outlined earlier and/or to use the phenomenon
itself in order to develop explicit tests of these various frameworks. The following
section will evaluate the evidence with respect to the various theoretical viewpoints.
As stated earlier, most of the evidence emanating from studies estimating wage
equations based on the decomposition of educational years acquired (equation 4)
have found that the returns to surplus education, whilst positive and significant,
tend to be lower than the returns to required education. Most researchers have
interpreted this as evidence against HCT which implies that the returns to surplus
and required education should be equal. However, Rumberger (1987) found that
in some occupations, there was no reward to surplus schooling suggesting that the
Job Competition Model may adequately explain behaviours within particular job
markets, a point consistent with Thurow’s initial analysis. Nevertheless, a number
of studies have attempted to develop more formal tests on the basis of equation 4,
specifically:
2¼3¼4ð6Þ
and,
3¼4¼0ð7Þ
Equation 6 represents a formal test of the HCT, whilst equation 7 suggests that
the marginal product is fully embodied within the requirements of the job and
thus is a test of the Job Competition Model. Failure to accept either hypothesis
implies that equation 8 will hold lending support to the assignment inter-
pretations of the labour market.
2„3„4ð8Þ
The majority of studies have tended to reject the hypothesis that the coefficients
are equal (Hartog and Oosterbeek, 1988; Alba-Ramirez, 1993; Groot, 1996; Kiker
et al., 1997; Sloane et al., 1999) either in the context of equation 6 or equation 7,
suggesting that models based around assignment views of the labour market most
accurately reflect the reality that is embodied within the data. However, Hartog
and Oosterbeek (1988) did report that for Dutch females, the assignment
specification whilst superior to the Job Competition Model was inferior to the
specification derived from HCT, which they rationalize within the context of
differences in labour supply behaviour. Hartog and Oosterbeek state that both
for participation and hours worked, female labour supply has substantially higher
wage elasticities than male supply. This explains why females have to be paid the
returns to their actual education even in jobs for which they are overeducated
(any reduction on this and relatively large proportions of them would not work).
The model also suggests that female elasticities fall with education level thus
allowing for full wage correction in jobs where females are undereducated;
OVEREDUCATION 409
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
nevertheless, as the authors point out, these results require further research.
However, the finding whilst not universal is unlikely to be unique to the Dutch
labour market given that Vahey (2000) reports no lower return to surplus educa-
tion for overeducated Canadian females despite finding them for some
overeducated Canadian males.
Some authors have adopted slightly differing approaches to assess the
adequacy of the various frameworks. Battu et al. (2000) examined whether, as
HCT suggests, firms upgrade the tasks they give to their overeducated workers,
their hypothesis being that there should be convergence in graduates’ jobs over
time, thus making overeducation a temporary phenomenon. Convergence was
tested for across time in the context of (1) a narrowing of the earnings gap, (2)
improved job satisfaction amongst the overeducated, (3) job characteristics and
(4) promotional prospects. They found that there is as much divergence as there is
convergence in the nature of the work for the overeducated compared with those
who are in matched employment. Therefore, there was no evidence to suggest that
overeducated graduates have been able to upgrade their jobs, thus suggesting
again that overeducation may be a long-term problem. Finally, McGuinness
(2003a) rejected the hypothesis that coefficients on both human capital and job
description variables were jointly zero within a hedonic wage equation model
demonstrating that both human capital and job characteristics are important
factors in determining wage rates lending further support to the assignment
interpretation.
However, all this does not spell the end for HCT and, as pointed out earlier, a
number of researchers have argued that it is still fully consistent with varying
rates of educational return under the following three circumstances. Firstly, if
overeducated workers possess lower amounts of informal human capital then
lower rates of return are simply a reflection of their true worth to the firm. This
substitution-based hypothesis argues that overeducated workers are merely
compensating for a lack of experience, implying that the lower returns to surplus
education reflect of an omitted variables problem (experience, on-the-job training
etc.) as opposed to any real under-utilization. Secondly, if overeducated workers
are less able than their well-matched counterparts, then again it is likely that they
are in fact being paid their marginal product suggesting that HCT remains
consistent. This interpretation is again consistent with that of an omitted
variables problem in that it suggests that the measures of human capital
contained within the standard Mincer equation do not sufficiently allow for the
existence of heterogeneous skill sets amongst individuals with similar levels of
education, and as we have seen, researchers have been making attempts to control
for such influences within the empirical framework. Finally, HCT will remain
consistent should overeducation proves to be a temporary phenomenon which
declines with occupational mobility and exists only for as long as it takes workers
to find an appropriate labour market match.
In support of the substitution hypothesis, a raft of studies (Duncan and
Hoffman, 1981; Sicherman, 1991; Sloane et al., 1996; Kiker et al., 1997; Sloane
et al., 1999; Cohn and Ng, 2000; Daly et al., 2000) all found that overeducated
410 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
workers tended to have lower levels of experience, tenure and/or training.
However, Groot (1993) found nothing to support the hypothesis that overeduca-
tion and enterprise training (proxies on-the-job training) were substitutes, whilst
Duncan and Hoffman (1981) found evidence to suggest that general labour
market experience is not treated by employers as a substitute for formal educa-
tion, and in addition, Alba-Ramirez (1993) still reported lower returns to over-
education after controlling for on-the-job training. Finally, the substitution
hypothesis seems somewhat at odds with graduate studies that report significant
differences in both the incidence and impacts of overeducation despite the fact
that the graduates in question entered the labour market at the same point in time
and thus possessed similar levels of experience (Dolton and Vignoles, 2000;
McGuinness, 2003a; McGuinness 2003b).
It is also important to note that none of the studies reporting an overeducation/
experience trade-off attempted to control for any cohort effects, whereby older
workers are less likely to be overeducated due to the demand conditions prevalent
on labour market entry. Groot (1996) is one of the few authors to highlight the
cohort-based explanation by considering the possibility that the observed inverse
relationship between overeducation and experience may be due to cohort effects
rather than the usually argued substitutable nature of these alternative options for
human capital investment Groot points out that if the growth in the education
level of the labour force is greater than the growth of the educational quality of
jobs, higher educated workers may displace lower educated workers and over-
education will increase. For the younger, higher educated cohorts, there are
relatively less high-qualified jobs than for the older workers. Thus, the experience
relationship may simply reflect these older (and now coincidentally more experi-
enced) workers entering the labour market during a period where there was a
higher availability of high-skilled jobs. Therefore, it can be argued that existing
studies examining the link between overeducation and informal human capital
accumulation are potentially weakened by a failure to account for cohort
influences.
In terms of the evidence that overeducated workers are in some way less able
than their well-matched counterparts, Groot (1996) reports that the negative wage
effects of overeducation increase with time (for men) suggesting that over time as
employers find out more about the productive capabilities of these workers they
tend to pay them less, suggesting that overeducated men are unproductive. Sloane
et al. (1999) argue that overeducated workers are less able on the basis that they
are less likely to be promoted. Green et al. (1999) reported evidence that the
overeducated tended to have lower mathematical abilities on the basis of which
they argue that the overeducated are likely to have less innate ability. However,
they also found that the overeducated were more likely to have better prose and
documentation skills which, as the authors later concede, suggest that the differ-
ences they detected between the adequately matched and overeducated were more
readily related to variations in the nature of the education undertaken. For
instance, it may well be that, as is the case with graduates, students electing to
take more Arts and Social Science GCSE’s/A levels (as opposed to more
OVEREDUCATION 411
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
vocational subjects) may be exposing themselves to a higher likelihood of over-
education.
7
However, there is also some evidence to suggest that overeducated
workers are more able relative to their undereducated counterparts; Mendes de
Oliveria et al. (2000) found that employers tend to value and prize overeducation
and at the same time penalize undereducation; with prolonged tenure, overedu-
cated workers were granted an ascending path of their relative earnings, while
undereducated workers saw their relative position eroded.
Finally, the extent to which overeducation is consistent with models of occupa-
tional mobility has also been examined within the literature, and as with the
previous two propositions, the evidence is mixed. Mobility-type frameworks
argue that overeducation is merely a transitory phenomenon that will gradually
disappear as workers improve their labour market position. In support of this,
Sicherman (1991) reported that overeducated workers have higher rates of firm
and occupational mobility, and they are more likely to move to a higher occupa-
tion. Alba-Ramirez (1993) found that the proportion of overeducated workers fell
with age, hinting a defined pattern in the upgrading process of overeducated
workers throughout their working lives. Alba-Ramirez also reported that over-
educated workers have a high job turnover rate pointing to an improvement in
the match over time. However, the finding that mobility does not necessarily
equate with an improved labour market position was also reported by Sloane
et al. (1999), whilst McGuinness (2003a) concluded that a large part of the
observed mobility consists of workers moving from one state of overeducation
to another. In addition, Dolton and Siles (2003) conclude that being overeducated
in first employment tends to permanently constrain graduates to low-level occu-
pations. Finally, more recent research has indicated that the permanency of the
overeducation effect will be related to the nature of the labour market individuals
are operating in; Groeneveld and Hartog (2004) report that career development
within the internal labour market is much more substantially impeded as a result
of overeducation.
4.4. Is Space Important?
Buchel and Battu (2003) report that, after controlling for distance to work, both
married males and females living in rural locations are more likely to be over-
educated. Buchel and van Ham (2002) build on these findings to further develop
the hypothesis that the risk of overeducation is highest for workers in small
regional labour markets and see spatial mobility as a potential means of avoiding
overeducation. They found that spatial flexibility reduces overeducation with
those with a car having lower probabilities (they acknowledge potential selection
effects associated with car ownership). Overeducation probabilities were also
positively correlated with potential commuting times. In terms of regional-specific
variables, the longer the travelling time to a large conurbation the higher the
overeducation probability suggesting that access to large concentrations of
employment activities reduces overeducation. Contrary to their expectations,
regional unemployment levels did not impact on overeducation (access to a
412 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
large agglomeration did) although they were important for employment prob-
abilities (whereas access to a large agglomeration had no effect). This suggests
that in order to avoid unemployment, it is not the size of the labour market that is
important but the level of competition. For overeducation, the size of the labour
market alone is relevant. To some extent, these findings contradict those of
McGoldrick and Robst who report that, for married females, large numbers of
vacancies in larger labour markets were offset be a larger number of job searchers.
Taken together, one could potentially conclude that competition effects may be
relevant in the context of married females, whereas for ‘less tied’ workers, it is the
size of the labour market that is important in determining the likelihood of
overeducation.
Lastly, in relation to cross-country differences, by comparing US data from the
Panel Study of Income Dynamics and The German Socio-Economic Panel, Daly
et al. (2000) investigate how structural changes in the US labour market over the
1970s and 1980s effected the rewards and penalties associated with education-
employment mismatches and make comparisons with the more structured
German labour market. They cite the US structural shifts as being increases in
both the demand and supply of highly skilled/educated labour, significant
increases in the returns to schooling and the introduction of non-neutral technol-
ogies that disproportionately benefit more educated individuals. Their results
suggest that the nature of overeducation has remained relatively unchanged
over time (mid-70s and mid-80s) in these countries and therefore had not reacted
radically to changes in institutional factors. They conclude that neither the well-
documented structural shifts in the US labour market between 1976 and 1985 nor
the substantial structural differences between the US and German labour markets
in 1984/1985 (Germany has more government involvement and stronger unions
implying less flexibility) significantly affected the returns to either surplus or
deficit education. They suggest that their results confirm a universal view of
labour markets with respect to overeducation given that they found more similar-
ities between countries at one point in time than within one country across
different points in time.
4.5. Results Summary
In terms of the theoretical frameworks, it is probably reasonable to conclude that
on the balance of evidence the assignment interpretation of the labour market is
most consistent with the findings of existing studies. This conclusion is drawn on
the basis that the evidence on the relative returns to education suggest differing
returns to required, surplus and deficit education and on the grounds that the
Human Capital or Job Competition Models were largely rejected by studies
carrying out formal hypothesis tests. Clearly, there are concerns that the absence
of controls for informal human capital accumulation and worker heterogeneity
will tend to overstate the effects of overeducation and understate the role of
human capital. In relation to the substitution hypothesis, the evidence is some-
what mixed, and even disregarding this, it is unclear to what extent any observed
OVEREDUCATION 413
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
overeducation/experience trade-off is due to human capital substitution or cohort
effects. In relation to the hypothesis that overeducated workers may be in some
ways less able or have fewer skills than their well-matched counterparts, there is
certainly some evidence to support this view. Nevertheless, the apparent accep-
tance of the assignment interpretation of the labour market within the literature
implies that wages are determined within a hedonic price equation containing
both individual and job characteristics. This would suggest that the methodolo-
gies that seek to control for worker heterogeneity implicitly by assuming that all
of the unobserved differences relate only to differences in skills will tend to
overestimate this effect. Taken together, the results would tend to support the
inclusion of more explicit controls for worker heterogeneity within the empirical
framework.
8
5. Summary and Conclusions
On balance, the evidence from existing overeducation studies tends to support an
Assignment interpretation of the labour market implying that both individual and
job characteristics are important for determining wage rates in the economy. The
arguments put forward by researchers who maintain that the apparent wage
effects of overeducation arise largely from flaws in the empirical framework are
at the very least debateable. Therefore, on the basis of the evidence, there is scope
to conclude that the impacts of overeducation are likely to be non-trivial and that
the phenomenon may potentially be costly to both individuals and firms and the
economy more generally.
The study therefore raises a number of important issues for policy. The wisdom
of the current government policy in pursuing a 50% higher education participa-
tion target must surely be highly questionable. In a very recent study, Walker and
Zhu (2005), using a very rough measure of graduate overeducation, indicated
that, post-expansion, the incidence of overeducation in the UK has risen for both
male and female graduates across almost all subject groups;
9
however, the rate of
growth seemed particularly rapid for graduates in the Arts and Humanities. The
obvious implication, therefore, is that should higher education participation
continue to expand in line with current government targets then rates of over-
education will inevitably rise incurring further costs on individuals, firms and the
economy. It is also clear from the literature that, particularly at the graduate level,
the incidence of overeducation is non-random with respect to subject studied as
graduates from backgrounds such as Arts, Humanities and some Social Sciences
are much more likely to end up overeducated. This raises the question with
respect to the extent to which government should seek to re-orientate the educa-
tional system away from the Arts and Humanities towards more vocationally
orientated subjects with higher levels of job relevant skills. Nevertheless, it is
likely that such a strategy, whilst providing some benefits, is unlikely to provide a
solution to overeducation, as the evidence suggests that the problem effectively
relates to the supply of educated labour exceeding demand coupled with an
inflexible labour market, whereby employers are either unable or unwilling to
414 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
alter their production processes to fully utilize the skills of their overeducated
workers.
Notes
1. Conversely, undereducation refers to the extent to which individuals acquired education
levels fall below those required by the job.
2. Although Freeman’s work is without doubt highly influential, it is not without its
drawbacks, the first thing to point out is that Freeman employs ‘Overeducation’ merely
as a descriptive term designed to characterize the observed collapse in the market for
college workers during the 1970s based on the observed decline in college earnings.
Freeman’s work does not explicitly consider the level of mismatch that exists between
education levels and job requirements, thus the extent to which declining wages were
associated with any drop in the proportion of graduates gaining graduate level employ-
ment remained largely unknown.
3. The Mincer model performs very poorly when applied to graduate cohort data (see
Dolton and Vignoles, 2000; McGuinness, 2003a) due mainly to the fact that there is
little variation in the education and experience levels of individuals in the dataset.
4. Rather bizarrely, both objective specifications reported that the return to a year of
undereducation was positive and significant with the magnitude of the return almost
three times higher under the occupational dictionary measure; in fact, the return to
undereducation under the occupational dictionary approach was over twice the
estimated return to required education.
5. It should be noted that some researchers have adopted approaches to measuring over-
education that are slightly different to those outlined above. Alpin et al. (1998) using
UK data from the Spring 1995 Labour Force Survey (restricted to GB graduates) use
two slightly alternative definitions, the first relates to SOC classification whereby they
classify the following occupations as requiring degrees, managers and administrators
(excluding Trade Union official’s minor group 19), professional occupations and com-
puter analysts from the associate professional occupations. The second definition is a
modal measure again based on the SOC classification; they define graduate occupations
as those where the qualification most frequently held by employees is a degree or higher
(based on two-digit SOC analysis). Clearly, given that many manager and administra-
tive occupations do not require degrees, it is likely that the modal method will generate
higher graduate overeducation estimates (i.e. its deeming more occupations to be non-
graduate).
6. The data relate to the combined sample; however, in circumstances where studies
reported the incidences for males and females, only the mean has been taken.
7. This in turn relates back to the substitutability point. A number of graduate studies
have found that overeducation is non-random with respect to faculty i.e. graduates from
Arts and Social Science backgrounds are much more prone to the phenomenon Dolton
and Siles (2003), McGuinness (2003a). Thus, rather than substituting formal for informal
human capital, overeducated workers may, to some extent at least, be substituting
higher amounts of ‘non-relevant’ education for a lack of job relevant education.
8. The strong evidence in support of assignment theory also serves to emphasize that the
need to include job characteristics in the wage equation framework.
9. The exception was for females from Maths and Engineering backgrounds.
OVEREDUCATION 415
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
References
Alba-Ramı
´rez, A. (1993). Mismatch in the Spanish labor market. Journal of Human
Resources 28(2): 259–278.
Allen, J. and van der Velden, R. (2001). Educational mismatches versus skill mismatches:
effects on wages, job satisfaction and on-the-job search. Oxford Economic Papers
53(3): 434–452.
Alpin, C., Shackleton, J. and Walsh, S. (1998). Over and undereducation in the UK
graduate labour market. Studies in Higher Education 1: 17–34.
Battu, H. and Sloane, P. J. (2000). Overeducation and crowding out in Britain. In
Borghans and de Grip (eds), The Overeducated Worker (pp. 157–175). Cheltenham:
Edward Elgar.
Battu, H. and Sloane, P. J. (2002). To what extent are ethnic minorities in Britain over-
educated. International Journal of Manpower 23(3): 192–207.
Battu, H., Belfield, C. and Sloane, P. (2000). How well can we measure graduate over-
education and its effects? National Institute Economic Review 171: 82–93.
Bauer, T. (2002). Educational mismatch and wages: a panel analysis. Economics of
Education Review 21: 221–229.
Becker, G. (1964). Human Capital: A Theoretical and Empirical Analysis with Special
Reference to Education. New York: Columbia University Press.
Buchel, F. and Battu, H. (2003). The theory of differential overqualification: does it work.
Scottish Journal of Political Economy 50(1): 1–16.
Buchel, F. and van Ham, M. (2002). Overeducation, Regional Labour Markets and Spatial
Flexibility. IZA Discussion Paper No. 424.
Burris, V. (1983). The social and political consequence of overeducation. American
Sociological Review 48(4): 454–469.
Chevalier, A. (2003). Measuring overeducation. Economica 70: 509–531.
Cohn, E. and Khan, P. (1995). The wage effects of overschooling revisited. Labour
Economics 2: 67–76.
Cohn, E. and Ng, C. Y. (2000). Incidence and wage effects of underschooling and
Overschooling in Hong Kong. Economics of Education Review 19: 159–168.
Daly, M., Buchel, F. and Duncan, G. (2000). Premiums and penalties for surplus and
deficit education: evidence from the United States and Germany, Economics of
Education Review 19: 169–178.
Decker, R., de Grip, A. and Heijke, H. (2002). The effects of training and overeducation on
career mobility in a segmented labour market. International Journal of Manpower
23(2): 106–125.
Dolton, P. and Siles, M. (2003). The determinants and consequences of overeducation. In
Buchel, de Grip and Mertens (eds), Overeducation in Europe (pp. 189–217).
Cheltenham: Edward Elgar.
Dolton, P. and Vignoles, A. (2000). The incidence and effects of overeducation in the UK
graduate labour market. Economics of Education Review 19: 179–198.
Duncan, J. and Hoffman, S. (1981). The incidence and wage effects of overeducation.
Economics of Education Review 1(1): 75–86.
Frank, R. H. (1978). Why women earn less: the theory and estimation of differential
overqualification. American Economic Review 68(3): 360–373.
Freeman, R. B. (1976). The Overeducated American. NewYork: Academic Press.
Frenette, M. (2004). The overqualified Canadian graduate: the role of the academic
programme in the incidence, persistence and economic returns to overeducation.
Economics of Education Review 23: 29–45.
Green, F., McIntosh, S. and Vignoles, A. (1999). ‘‘Overeducation’’ and Skills – Clarifying
the Concepts. London: Centre for Economic Performance.
Green, F., McIntosh, S. and Vignoles, A. (2002). The utilization of education and skills:
evidence from Great Britain. The Manchester School 70(6): 792–811.
416 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Groeneveld, S. and Hartog, J. (2004). Overeducation, wages and promotions within the
firm. Labour Economics 11: 701–714.
Groot, W. (1993). Overeducation and the returns to enterprise related schooling.
Economics of Education Review 17(4): 299–309.
Groot, W. (1996). The incidence and returns to overeducation in the UK. Applied
Economics 28: 1345–1350.
Groot, W. and van den Brink, H. (1997). Allocation and the returns to overeducation in
the UK. Education Economics 50(2): 169–183.
Groot, W. and van den Brink, H. (2000a). Skill mismatches in the Dutch labor market.
International Journal of Manpower 21(8): 584–595.
Groot, W. and van den Brink, H. (2000b). Overeducation in the labour market: a meta
analysis. Economics of Education Review 19(2): 149–158.
Hannan, D. F., McCabe, B. and McCoy, S. (1998). Trading Qualifications for Jobs –
Overeducation and the Irish Youth Labour Market, Dublin ESRI.
Hartog, J. and Oosterbeek, H. (1988). Education, allocation and earnings in the
Netherlands: overschooling? Economics of Education Review 7(2): 185–194.
Heckman, J., Lochner, J. and Todd, P. (2003). Fifty Years of Mincer Regressions, IZA DP,
No. 775.
Hersch, J. (1995). Optimal ‘‘mismatch’’ and promotions. Economic Enquiry 33: 611–624.
Kiker, B., Santos, M. and Mendes de Oliveiria, M. (1997). Overeducation and under-
education: evidence for Portugal. Economics of Education Review 16(2): 111–125.
Mason, G. (1996). Graduate utilisation in British industry: the initial impact of mass higher
education. National Institute Economic Review 156: 93–103.
McGoldrick, K. and Robst, J. (1996). Gender differences in overeducation: a test of the
theory of differential overqualification. American Economic Review AEA Conference
proceedings.
McGuinness, S. (2003a). Graduate overeducation as a sheepskin effect: evidence from
Northern Ireland. Applied Economics 35: 597–608.
McGuinness, S. (2003b). University quality and labour market outcomes. Applied
Economics 35: 1943–1955.
McGuinness, S. and Bennet, J. (2006). Overeducation in the graduate labour market: a
quantile regression approach. Forthcoming, Economics of Education Review.
McMillen, D. P., Seaman, P. T. and Singell, L. D. (1999). A Hedonic Analysis of Overeducation
and Undereducation. University of Oregon Working Papers, University of Oregon.
Mendes de Oliveria, M., Santos, M. C. and Kiker, B. (2000). The role of human capital and
technological change in overeducation. Economics of Education Review 19: 199–206.
Mincer, J. (1974). Schooling, Experience and Earnings. New York: Columbia University
Press.
O’Leary, N. and Sloane, P. (2005). The Changing Wage Returns to an Undergraduate
Education. IZA DP No. 1549.
Patrinos, H. (1997). Overeducation in Greece. International Review of Education 24(2/3):
203–223.
Robst, J. (1995). College quality and overeducation. Economics of Education Review 14:
221–228.
Rubb, S. (2004). Overeducation in the labor market: a comment and re-analysis of a meta
analysis. Economics of Education Review 22: 621–629.
Rumberger, R. (1987). The impact of surplus education on productivity and earnings.
Journal of Human Resources 22(1): 24–50.
Sattinger, M. (1993). Assignment models of the distribution of earnings. Journal of
Economic Literature 31: 831–880.
Sicherman, N. (1991). ‘‘Overeducation’’ in the labour market. Journal of Labor Economics
9(2): 101–122.
OVEREDUCATION 417
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
Sloane, P. (2003). Much ado about nothing? What does the overeducation literature really
tell us. In Buchel, de Grip and Mertens (eds), Overeducation in Europe (pp. 11–49).
Cheltenham: Edward Elgar.
Sloane, P. J., Battu, H. and Seaman, P. T. (1996). Overeducation and the formal
education/experience and training trade-off. Applied Economics Letters 3: 511–515.
Sloane, P., Battu, H. and Seaman, P. (1999). Overeducation, undereducation and the
British Labour market. Applied Economics 31: 1437–1453.
Spence, M. (1973). Job market signalling. Quarterly Journal of Economics 87(3): 355–374.
Thurow, L. C. (1975). Generating Inequality. New York: Basic Books.
Tsang, M. (1987). The impact of underutilisation of education of productivity: a case study
of the U.S. Bell companies. Economics of Education Review 11: 239–234
Tsang, M., Rumberger, R. and Levin, H. (1991). The impact of surplus schooling on
worker productivity. Industrial Relations 30(2): 209–228.
Vahey, S., (2000) The great Canadian training robbery: evidence on the returns to educa-
tional mismatch. Economics of Education Review 19: 219–227.
Verdugo, R. and Verdugo, N. (1989). The impact of surplus schooling on earnings. Journal
of Human Resources 24(4): 629–643.
Walker, I. and Zhu, Y. (2005). The college wage premium, overeducation and the expansion
of higher education in the UK. IZA DP No. 1627.
418 McGUINNESS
#2006 The Author
Journal compilation #2006 Blackwell Publishing Ltd
