Education-Job Mismatch and Gender Wage Gap: Evidence from Recent College Graduates in Korea

Data Source: Graduates Occupational Mobility Survey (GOMS)

This study utilizes cross-sectional survey data from the Graduates Occupational Mobility Survey among the cohort graduated in the year 2015 (hereafter, the 2015 GOMS). The GOMS is an annual census of all graduates who have completed 2/3-year courses in colleges or higher education institutions in Korea. The GOMS sample is composed of 4% of all Korean graduates, which corresponds to 18,000 individuals. The GOMS is conducted annually, and takes places three months to one year after graduation, typically in September. This national graduate survey focuses on education situation and labor market status (including employment, unemployment, and economic inactivity) of graduates from higher education institutions (hereafter, HEIs), one year after graduation. The GOMS dataset covers a wide range of graduates’ characteristics, such as their labor market activities, personal and socio-economic characteristics, and information on their educational experience (past and current). The survey findings are used as a vital tool to gather information for various HEIs and as guidelines for educational and employment authorities and researchers.

In order to get a more homogenous sample, the sample used in this empirical analysis is restricted to full-time salaried graduate workers aged 20 to 34 who have provided valid information on the variables of interest, such as wages. Thus, self-employed (i.e., own-account) and unpaid family workers at the time of the survey are excluded. These exclusions, after deleting some missing data, result in a final sample size of 10,116 observations. Of the total respondents in the sample, 5,193 (51.3%) are males and 4,923 (48.7%) are females. Such homogeneity may limit any bias associated with a lack of effective controls for informal human capital such as that obtained through labor market experience and on-the-job training, thus allowing this study to more effectively isolate ability-related impacts (i.e., ability bias).

Definitions and summary mean statistics for the variables used for empirical analysis are provided in Table 1 and Table 2, respectively. As can be observed from Table 2, the raw geometric mean of the log of hourly wage is higher for men, with women’s wages being on average approximately 89% of those earned by their male counterparts. It suggests that young female college graduates working full-time received, on average, nearly 11% lower hourly wages than their male peers. It can also be highlighted that men and women are basically alike in terms of productive characteristics such as age and acquired level of education (i.e., there is no major difference). Gender differences in most other characteristics such marital status, region, union membership, and type of employer are also generally small. Some differences between men and women are found in occupational characteristics and industrial sectors. Industry variables in particular are those showing the largest gender difference: men dominate the workforce in manufacturing and mining and women constitute the majority in health care, social services, recreational and cultural services, and education and research sectors.

Table 1

Definition of Variables

Table 2

Summary Mean Statistics

Definition of Educational Mismatch

There are three common approaches to the definition of educational mismatch in the literature: (i) worker self-assessment, (ii) job analysis, and (iii) realized matches. The worker self-assessment approach is based on the worker’s subjective measure of educational mismatch, while the job analysis and realized matches approaches use objective assessments of educational mismatch. The validity and choice of measurement methods generally depend on the availability and limitations of the data. In this paper, the worker self-assessment approach is the only available method used to define educational mismatch. As a subject indicator, unfortunately, the worker self-assessment approach may lead to biases arising from respondents due to outdated or inaccurate information. Nevertheless, it is worth noting that each of the approaches discussed above has its merits and drawbacks. Moreover, as noted by Hartog (2000), overall, the worker self-assessment approach is considered the best available method for educational mismatch. Indeed, key earlier studies in the educational mismatch literature adopting the worker self-assessment approach include Duncan and Hoffman (1981), Green and Zhu (2010), Rumberger (1987), and Sicherman (1991), among others.

The main purpose of this research is to investigate the relation between educational mismatch and the gender wage gap in the Korean graduate labor market. As noted earlier, there are two types of educational mismatch for college graduates: vertical mismatch and horizontal mismatch. Fortunately, the 2015 GOMS dataset employed here contains yearly information on both vertical and horizontal mismatches as the basis. In order to gather these two sets of information, the following two survey questions are used. The first question asks respondents directly about the appropriate level of formal education required to be qualified for their current job. The response to this question is used to assess the extent to which formal education (mis)matches the jobs, and is referred to as vertical (mis)matching status. Specifically, respondents are asked directly: “What is the level of education required by your current job, compared to your education level?” They can choose between five possible responses: (i) Very Low, (ii) Low, (iii) Appropriate, (iv) High, and (v) Very High. Individual workers claiming the first two choices (Very Low, Low) are defined as over-educated, Appropriate as adequately-educated, whereas High and Very High are defined as under-educated. Since the paper focuses on college graduates, the case of under-education is not considered, as this group generally has the highest level of education. In this sense, vertical mismatch as used here has the same meaning as over-education, and thus over-education and adequately-matched in education are used to construct the degree of vertical (mis)matching status in this empirical analysis.

The other question asks respondents whether their jobs are highly, partly, or not related to their major field of study. The response to this question is used to assess the extent to which the major field of study matches the jobs, and is referred to as horizontal (mis)matching status. Horizontal mismatch is identified using the following question: “Thinking about the relationship between your work and your education, to what extent is your work related to your college major?” There are five choices: (i) Not Related, (ii) Somewhat Related, (iii) Related, (iv) Closely Related, and (v) Very Closely Related. The first two answers (Not Related, Somewhat Related) are classified as horizontal mismatch and the other three (Related, Closely Related, Very Closely Related) as horizontal match. Specifically, an individual who claims to be working in a job somewhat related to his/her field of study in college and working in a job not related to his/her major field are identified as being horizontally mismatched. This study considers those college graduates who are working in jobs somewhat related or not related to their major field of study as presumably using less of the knowledge, skills, and training leaned in that field of study.

There are two types of vertical (mis)matching status as well as two types of horizontal (mis)matching status. Therefore, the combination between vertical mis(match) and horizontal (mis)match leads to the four possible forms of (mis)matching categories below:

• (a) Vertically mismatched only: The individual worker is horizontally matched, but vertically mismatched.
• (b) Horizontally mismatched only: The individual worker is vertically matched, but horizontally mismatched.
• (c) Both vertically and horizontally mismatched: The individual worker is mismatched in education both vertically and horizontally.
• (d) Well-matched: The individual worker is adequately matched in education both vertically and horizontally.

Estimation Model

To examine the effects of educational mismatch on wages, two modified specifications of the semi-logarithmic Mincer equation (1974) are often used in the literature: The ORU (Over-, Required and Under-education) specification proposed by Duncan and Hoffman (1981) and the dummy variable specification proposed by Verdugo and Verdugo (1989). This paper considers a specification with three dummies for educational mismatch, which is a special case of the Verdugo and Verdugo approach, and includes dummies both for being over-educated and being under-educated. Due to the limitation of data availability on years of education, this work is unable to perform the ORU approach. For estimating the wage effects of three forms of educational mismatch, the estimation equation is specified as:

where i indexes individuals, ln W_i denotes the natural logarithm of the hourly wages of the individual worker i, and X is a vector of n control variables including personal and employment characteristics. VM, HM, and BOTH are dummy variables which take the value 1 if an individual worker is vertically mismatched only, horizontally mismatched only, and both vertically and horizontally mismatched, respectively, and 0 if an individual worker is adequately matched in the workplace. According to the Eq. (1), vertically mismatched only, horizontally mismatched only, and both vertically and horizontally mismatched workers are thus compared to workers who are well-matched. The coefficients of γ₁, γ₂, and γ₃ represent the average wage effects of three forms of educational mismatch in comparison to their well-matched counterparts, respectively. μ_i is a conventional mean zero disturbance.

The Blinder-Oaxaca Decomposition Approach

The second part of the paper analyzes whether the educational mismatch affects the gender wage gap. To this end, this study employs a generalized Oaxaca-Blinder decomposition model introduced by Oaxaca and Ransom (1994). In order to gain a better insight into the composition of the gender wage gap, one may consider methods of decomposing wage differentials by gender into contributing factors to analyze the contributions of different variables to the gender wage gap. Oaxaca (1973) and Blinder (1973) were the first to suggest a methodology for assessing wage differentials between sub-groups of the population. Within this framework, wage differentials are decomposed into a component that can be explained by group differences in productivity and a component (or a residual) unexplained by these observable productive differences, often interpreted as discrimination effects. Regarding the gender wage gap, there are differences in the groups’ average human capital levels, and discrimination in the marketplace measured by different rates of return on human capital is also generally prevalent.

In this empirical analysis, males (m) and females (f) identify the two separate groups, the natural logarithm of the hourly wages is the dependent variable (W), and several variables indicating personal background and employment characteristics are predictors. The question is how much of the mean outcome difference:

where the upper bar indicates the mean of the variables, denotes the expected value of the dependent variable, is accounted for by group differences in the predictors. A standard decomposition of Blinder (1973) and Oaxaca (1973) is

This standard decomposition adopts the male wage structure as the nondiscriminatory norm, although an alternative choice for a nondiscriminatory wage standard can also be the female wage structure. The first term on the right-hand side (hereafter, RHS) of Eq. (3) represents the component of the wage differential that is explained by group differences in the mean of the explanatory variables or endowments and is considered ‘the explained part’ of the wage gap. The second term on the RHS of Eq. (3) stands for the part of the wage gap that is interpreted as a measure for discrimination and is considered ‘the unexplained part’ of the wage gap, since it refers to differences in coefficients, or returns on productive characteristics of male and female workers. This component thus implies unequal treatment of productive characteristics in the labor market.

There are two main measurement issues raised in the standard decomposition approach. The first issue lies in the choice of the non-discriminatory wage structure. A number of researchers, including Cotton (1988), Neumark (1988), and Oaxaca and Ransom (1994), have raised this issue and have demonstrated that the results of the wage decompositions are sensitive to the choice of the non-discriminatory norm. Often researchers select the wage structures for the group of workers believed to be dominant in the labor market. Much prior research in the decomposition literature adopts the male wage structure (m) as the non-discriminatory benchmark. Another issue that has been raised with the standard decomposition approach lies in the fact that the standard decomposition technique estimates only relative differences. In the case of discrimination estimates, one does not know how much of the discriminatory (unexplained) wage gap arises from favoritism toward one group of workers and how much arises from pure discrimination against the other group.

In order to resolve this identification issue, following Oaxaca and Ransom (1994), the nondiscriminatory wage structure is estimated from a pooled sample (both males and females) in this empirical analysis. This more general decomposition method allows the discrimination component to be further disaggregated into two parts: overpayment (favoritism) and underpayment (pure discrimination) as shown below. One implication of the standard decomposition procedure is that all of the discriminatory differentials are ascribed to underpayment of the subordinate group rather than to overpayment of the dominant group (Neuman & Oaxaca, 2004).

Let β^* be such a nondiscriminatory coefficient factor. It is based on the assumption that discrimination can be negative (against one group) or positive (in favor of other group). The outcome differences can be decomposed into explained (Q) and unexplained (U) components of the wage gap:

where the first component Q,

is the part of the wage differential explained by group differences in the magnitudes of the determinants of the wages (the quantity effect from productivity differences between analyzed groups), and the second component U,

is the unexplained part that describes average discrimination in favor of the group in the relatively better situation ($$ {\overline{X}}_m^{\text{'}}\left({\beta }_m-{\beta }^{*}\right)$$), and average discrimination against the group in the relatively worse situation ($$ {\overline{X}}_f\left({\beta }^{*}-{\beta }_f\right)$$). This is usually attributed to discrimination, but it also captures all potential effects of differences in unobserved variables.

Incidence of Educational Mismatch

In order to answer RQ1(Does the educational mismatch phenomenon affect males and females to different extents?), Table 3 provides the incidence of educational mismatch among Korean college graduates one year after graduation for both men and women, where the educational mismatch variable is derived from a combination of vertical and horizontal mismatches. As can be observed in panel A, the overall share of college graduate workers mismatched in education is non-negligible in the youth labor market. The proportions of being educationally mismatched reach approximately 35% for both genders (34.6% in the case of males and 35.4% in the case of the female sample). This suggests that at least three out of ten young Korean college graduates are mismatched in the early stage of their careers (i.e., one year after graduation).

Table 3

The Incidence of Educational Mismatch (col. proportions)

When college graduate workers mismatched in education are disaggregated into three categories of educational mismatch in panel B, the results show that approximately 10% of college graduates are mismatched vertically, but matched horizontally (vertically mismatched only)—9.5% (resp. 11.3%) in the male (resp. female) sample. It appears that these proportions are generally comparable to or higher than those obtained in other earlier studies on educational mismatch of college graduates. Using data from the Higher Education as a Generator of Strategic Competences (HEGESCO) and the Flexible Professional in the Knowledge and Society (REFLEX) surveys on college graduates in 17 European countries and Japan, for instance, Verhaest, Sellami, & van der Velden (2017) conclude that on average, countries had an identical early incidence of being “vertically mismatched only”— each at 8%. Based on data from the Household, Income and Labour Dynamics in Australia (HILDA) survey (2001–2012), Mavromaras, McGuinness, O’Leary, Sloan, and Wei (2013) show that approximately 13% of Australian graduate workers are “vertically mismatched only.”

Similarly, the results report that approximately 15% of recent college graduate workers consider that their current jobs are irrelevant to their studied field of discipline (i.e., horizontally mismatched only). These findings are also consistent with those of Verhaest et al. (2017), showing that the incidence of being mismatched in education is somewhat higher for the case of being “horizontally mismatched only” than for the case of being “vertically mismatched only.” Not surprisingly, the proportions of both vertically and horizontally mismatched appear smaller than those of the two types of educational mismatch discussed above. This relatively low proportion of both vertically and horizontally mismatched workers may suggest that there is a major lack of skilled or qualified graduate workers in the Korean labor market, although these results come from different measurement approaches.

The Wage Effects of Educational Mismatch

In order to answer RQ2 (Does educational mismatch play a large role in the wage determination process? If so, does the effect of educational mismatch on wage returns differ for males and females?), Table 4 reports empirical evidence on the effect of educational mismatch on the natural logarithm of the hourly wages (LNHRW) for the whole sample as well as for men and women separately. According to the results in column (1), on the whole, the estimated coefficients for educational mismatch show that being educationally mismatched has a significant negative effect on wages, indicating that young college graduate workers mismatched in education earn lower wages than their well-matched peers with the same educational attainment level who work in jobs that fully utilize their educational qualifications. These findings mean that over-education is significantly associated with lower wages (i.e., wage penalties).

Table 4

The Wage Effects of Educational Mismatch (col. Standard errors)

When vertical mismatch is combined with horizontal mismatch, the coefficients on all types of educational mismatch are still significantly negative. Specifically, the results suggest that young college graduate workers who are vertically mismatched only (resp. horizontally mismatched only) appear to suffer about a 4.5% (resp. 4.6%) wage penalty. Not surprisingly, the relation between both vertically and horizontally mismatched and wages has the largest magnitude, indicating that the wage penalty is highest among young college graduates who are both vertically and horizontally mismatched at the beginning of their careers. This suggests that workers who were reported as both vertically and horizontally mismatched are likely to earn wages approximately 10.5% less than their well-matched counterparts.

A separate analysis is performed disaggregating the results based upon gender. The estimated coefficients for male and female samples are presented in column (2) and column (3), respectively. Similar to the results in column (1), all findings still show statistically significant negative coefficients of educational mismatch for both genders. Interestingly, the results in columns (2)–(3) point to significant gender differences in wage penalties for being educationally mismatched. Overall results show that female college graduates have a relatively higher wage penalty than their male counterparts. The wage penalty relating to vertically mismatched only (resp. both vertically and horizontally mismatched) fell from 4.9% (resp. 11.6%) in the female sample to 4.4 percent (resp. 9.8 percent) in the male sample. These findings suggest that educationally mismatched female college graduate workers appear to suffer more severe wage losses than their male counterparts in the Korean youth labor market. As the wage effects of educational mismatch are found to differ for male and female samples, this study analyzes the gender wage gap among educationally mismatched workers as well as among well-matched workers in education.

Decomposing the Gender Wage Gap

In order to answer RQ3 (Is the gender wage gap explained by differences in productivity characteristics and/or by different returns associated with such characteristics?)– RQ4 (Does educational mismatch make a significant contribution to the overall gender wage gap in the Korean graduate labor market?), as a preliminary analysis, the log hourly wages for three types of educationally mismatched and well-matched workers by gender are presented in Table 5. The gender wage gap in the full sample amounts to 11%. When educational mismatch is taken into account, the gender wage gap is comparable or slightly higher among educationally mismatched workers (11.9% of the sample being classified as Vertically Mismatched Only, 13.0% of the sample being classified as Horizontally Mismatched Only, and 10.1% of the sample being classified as Both Vertically and Horizontally Mismatched), compared to their well-matched counterparts (10.4% of the sample being classified as Well-Matched).

Table 5

Log of Hourly Wages by Gender and Gender Wage Gap (col. percent)

To gain a better understanding of the composition of the wage gap in the sample of males and females, the gender wage gap presented in Table 5 is decomposed using the Oaxaca and Ransom-type generalized decomposition discussed in the EMPIRICAL MODEL AND METHODOLOGY section. In Table 4, all three types of educational mismatch have a negative wage effect and this negative effect is more pronounced for the sample of female college graduates. Thus, the generalized decomposition approach employed in this analysis shows that educational mismatch could be one of the main drivers of the gender wage gap in the Korean graduate labor market.

The generalized decomposition results are provided in Table 6. The results are shown for the full sample (Full Sample), for the three sub-samples of recent college graduate workers who are educationally mismatched (Vertically Mismatched Only; Horizontally Mismatched Only; Both Vertically and Horizontally Mismatched), and for those who are adequately matched (Well-Matched). Overall, the explained component (i.e., the gender wage gap in productivity characteristics or endowments) accounts for a major part of the gender wage gap, with more than 50%. Also, the explained component is positive among all three types of educationally mismatched workers as well as among well-matched workers. This indicates that average observable male labor market characteristics are better that those of females.

Table 6

Decomposing the Gender Wage Gap: Generalized Decomposition

In term of the unexplained (or coefficients) component, the results indicate that the gender wage gap in different returns to productivity characteristics is also non-negligible among three types of educationally mismatched workers. This unexplained component of the gender wage gap is usually attributed to discrimination in the literature. The fact that the unexplained wage gap is not justified in terms of labor productivity is usually known as gender wage discrimination. In addition, the unexplained component shows the hypothetical wage gain for females if their own features were remunerated like those of their male counterparts. As this term is positive for both educationally mismatched and well-matched workers, but is higher among educationally mismatched workers, this suggests that gender wage discrimination among the educationally mismatched is more important than among the well-matched. The unexplained component among educationally mismatched workers accounts for at least 27.7%, compared to 14.4% among well-matched workers.