Dynamics of returns to vocational education in China: 2010–2017

Abstract

In this paper, we use the Chinese General Social Survey data (2010–2017) to analyze the dynamics of returns to different education qualifications. We find increasing returns for all types of education, with returns to vocational and academic education increasing roughly at the same speed. We additionally compare the returns to vocational education with returns to academic education. Compared to those who only complete compulsory education, upper secondary graduates earn about 20% more, vocational college graduates earn 50%, and academic university graduates earn 75% more. At tertiary level, academic education pays better than vocational education, although the difference shrinks over the years. At upper secondary level, the evidence is indeterminate, depending on different econometric techniques (i.e., OLS, IV, Lewbel method, or PSM). These findings add to the limited quantitative evidence on returns to vocational education. The dynamics emerged from the findings echo the discussion on labor market mismatch and overeducation in China, which has important policy implications.

Introduction

The vocational education and training sector has been at the heart of recent economic reforms all over the world (Chappell 2003; Tran 2021). As the economy develops, more high value-added industries will substitute low value-added industries (Loyalka et al. 2016). Consequently, the demand for high-skilled workers grows. The positive relationship between economic development and demand for skilled workers has been observed in many developing countries in recent years. To better prepare future graduates for the job market, a country’s education system has to evolve accordingly. Vocational education thus plays a vital role in a country’s education system (Tran 2021; Patrinos et al. 2021).

As one of the world’s largest and most vibrant economic entities, China is no exception in upgrading its vocational education and training system. In China, vocational education and vocational training are different. Vocational education is school-based and earns a credential whereas vocational training is work-based and may or may not earn a credential. Our focus is on the former, which is a kind of education qualifications. ¹

China’s rapid economic growth sees a dramatic increase in the demand of high-skilled workers (Heckman and Yi 2014). As a response, the State Council has enacted a series of policies (e.g., Decisions on Accelerating Development of Modern Vocational Education in 2014) aiming to improve the social status of skilled workers. However, as a result of a long-standing tradition of degrading vocational graduates, lower secondary graduates are reluctant to pursue a vocational credential unless their academic performance is too poor for an academic degree (Hansen and Woronov 2013). Figure 1 presents the share of individuals with different education qualifications in China over the period 2010–2017. From Fig. 1a, we see that the largest proportion of individuals only completed lower secondary education; around 14% completed academic upper secondary education while only around 9% completed vocational upper secondary education; from 2010–2013 to 2015–2017, there is a shift from lower education qualifications to higher education qualifications, reflecting an overall improvement in the quality of labor supply. Figure 1b shows that the story is similar when we zoom into the young cohort. Notably though, in 2015–2017, the share of academic university graduates surpasses the share of lower secondary graduates. The difference shows a trend that more well-educated employees are needed for China’s booming economy.

Fig. 1: Share of graduates by education qualifications.

a All working age people, b Young cohort (age 24–35).

Despite the upsurge in the supply of academic university graduates, employers in China still face a supply shortage of skilled labors. This is for two reasons. On the one hand, curricula of academic universities are theory-oriented and do not really prepare students for skill-intensive jobs. On the other hand, academic university graduates prefer “white-collar” jobs and are reluctant to do “blue-collar” jobs. Consequently, the supply and demand of labor constantly mismatch. To alleviate the shortage of supply of skilled workers, the Chinese government has been actively promoting vocational education since the 1990s. Yet, skilled workers are still under-supplied in the labor market (Fudan University, Tsinghua University 2016). Two reasons likely have contributed to this perpetuating supply shortage. One reason is that the culture of degrading vocational graduates persists, discouraging parents from sending their child to vocational institutes in the first place. The other reason is that the returns to vocational education in the labor market is not high enough to attract sufficient students. In this paper, we speak to the second reason by looking at returns to different education qualifications, especially vocational education ones, from 2011 to 2017 in China. In education-related literature, dynamics means the changes of returns to education over a time period (Arias and McMahon 2001; Söderbom et al. 2006; Crespo Cuaresma and Raggl 2016). Specifically, we aim to answer the following research questions: What are the dynamics of returns to vocational education in China? What implications do the dynamics have on education policy and practice? On the one hand, we see high-quality vocational workers are in short supply. On the other hand, we see high-quality vocational workers are in great demand. Thus, understanding the changes of returns to vocational education in China has important and urgent policy and practical implications.

The research makes contribution to the dynamics of returns to vocational education in China – a rarely addressed issue in the following aspects. Along with this investigation, we provide an updated and more holistic picture of the returns to different education qualifications in China during 2010–2017, focusing on vocational education. Second, to cope with the notorious endogeneity issue in estimating returns to education (Card 1999; Woodridge 2010; Castellano et al. 2014), we adopt rigorous econometric techniques (i.e., conventional IV, heteroskedasticity-based Lewbel (Lewbel, 2012) IV) to compare the returns to vocational education and academic education, both at the upper secondary level and at the tertiary level. In doing this, we not only come up with a novel instrument (i.e., the proportion of university graduates relative to the entire population per year but also replicate the enrollment ratio IV as is used in Dai and Martins (2020). Although we are able to replicate Dai and Martins (2020) using the same IV and find similar “wage premium” enjoyed by vocational graduates as opposed to academic graduates using the same instrument, the “premium” disappeared when we try with the novel instrument or various econometric techniques as robustness checks. The fact that some findings are not robust to alternative econometric specifications serves as a cautionary reminder to exercise greater diligence in the selection of estimation techniques before drawing definitive conclusions. Additionally, we find evidence of increasing gender inequality in subsample analysis. Last but not least, we tie the discussion of the dynamics in the returns to different education qualifications with the ongoing vocational education reforms in China, providing international readers with an enriched picture of vocational education in China.

This paper is structured as follows. In Section “Literature review”, we explain the research context of the education system in China, as well as summarizing the related literature. In Section “Theoretical framework”, we provide a theoretical framework as a roadmap guiding the empirical analysis in this paper. In Section “Methodology”, we describe the data and discuss the methodology. Section “Findings and discussions” presents the results and discussion. We conclude in Section “Conclusion”.

Literature review

Vocational education has attracted much attention in recent decades, especially after the massification of higher education in developed countries and recently in some developing countries including China (Meng 2012; Mok and Marginson 2021). In this section, we first provide contextual explanation of the current education system in China and its potential implications on the changes of returns to vocational education. Then we summarize the literature studying the returns to education in China and position our paper in this specific strand of literature.

The education system in China

The current education system in China mainly consists of two stages. The first stage is compulsory and lasts 9 years - 6-year primary education plus 3-year lower secondary education.² The second stage is competitive and thus noncompulsory, including upper secondary level education and those above. Figure 2 illustrates the major progression paths in China’s education system. As can be seen, a student normally sits in the High-school Entrance Exam (HSEE) after completing the compulsory education (Davey et al. 2007). Depending on his/her performance in the exam, he/she may enter an academic upper secondary school or a vocational upper secondary school. Some scholars have argued the HSEE to be “a form of class sorting” in modern China as the majority of those who fail the HSEE are “destined to a life in the working class” (Woronov 2016). The upper secondary education normally lasts three years, after which students compete to enter tertiary level education or directly enter the job market. The path to tertiary education is distinct for academic students and vocational students, which is different from the education system in some other countries (Brunello and Rocco 2017). An academic upper secondary graduate has to attend the College Entrance Examination to enter tertiary education. Depending on his/her performance, he/she may end up in an academic university or a vocational college. Meanwhile, a vocational upper secondary graduate only needs to attend a transitional exam, which is normally carried out by individual vocational colleges, to progress to a vocational college. Meanwhile, a vocational upper secondary graduate may still choose to take the National College Entrance Examination if he or she prefers not to take the transitional exam. Although the latter path to tertiary education seems easier, it is less attractive to students for two reasons. First, the latter path eliminates the possibility of directly entering an academic university which is more highly valued in the current job market. Second, the latter path makes it hard to continue pursuing postgraduate degrees. Until 2019, a vocational college graduate cannot directly attend the Graduate Entrance Exam, a necessary path for general undergraduates to progress to postgraduate education. Clearly, vocational education was designed as a suboptimal option compared to academic education.

Fig. 2

Education system in China.

Meanwhile, as the Chinese economy expand and grow, the Chinese government and China’s industries demand specialized, technically adept workers, meaning that vocational graduates face favorable policies (Schulte 2013). In 1996, the Chinese government enacted the Vocational Education Law of the People’s Republic of China, setting out official standards for vocational education in China. Since then, vocational education has received increased attention and resources from the government (Li et al. 2021). In 2005, the Chinese government announced Decision of the State Council on Vigorously Developing Vocational Education, which stimulated a rapid growth period of vocational education. For example, from 2005 to 2017 the national financial funds for vocational education reached 2.7 trillion yuan. As of 2017, the total investment in vocational education was about 434 billion yuan, nearly five times that of 93.9 billion yuan in 2005, with an average annual growth rate of 14%. In 2019, the State Council released the National Vocational Education Reform Implementation Plan, which further clarified the types and statuses of vocational education, and at the same time carried out pilot vocational education at the undergraduate level. In 2021, Opinions on Promoting the High-quality Development of Modern Vocational Education was initiated by the central government. In 2022, the Vocational Education Law of the People’s Republic of China was amended to formally recognize vocational education at the undergraduate and postgraduate level.

However, the effort to improve the situation of vocational education is counteracted by two main forces. First, China has a long tradition of discriminating vocational graduates (Schulte 2013). Even parents in rural areas are reluctant to send a child to a vocational institute unless getting into an academic upper secondary school is impossible (Hansen and Woronov 2013). As a result of the tradition, good students try to avoid vocational education if they can, which further lowers the quality of vocational graduates and perpetuates the vicious cycle (Woronov 2016). Second, the massification of higher education since 1999 escalates the supply of university/college graduates, both academic ones and vocational ones (Dai and Martins 2020). Both kinds of university graduates then have to compete in the same job market. Yet, until September 2019, vocational college graduates can only obtain diplomas or occupational qualification certificates but not degrees (Xiong 2011; Li et al. 2022), being disadvantaged even further. As the increasing rates of university graduates are much higher than those of positions requiring skilled workers, job market competition has soared, and academic graduates have had to compete with vocational graduates.

Returns to (vocational) education in China

In the existing education literature in China (and neighboring countries such as Japan and Korea), the focus has been largely on academic stream education but not vocational stream education (Hansen and Woronov 2013). More specifically, only a handful of studies quantitatively explore the returns to vocational education, especially vocational upper secondary education, in developing countries including China. The lack of relevant evidence is partly due to data unavailability (Guo and Wang 2020; Woronov 2016).

Among existing studies on this topic, the evidence is mixed. Using the 2003–2015 Chinese General Social Survey data on all workers, Dai and Martins (2020) compare returns to upper secondary vocational education versus returns to upper secondary academic education in China. The authors find that, although secondary vocational graduates do not enjoy wage premia over secondary vocational graduates on average in OLS regressions, secondary vocational education does result in more than 30% wage premia for individuals with average earnings potential when estimated by instrumental variable quantile regressions. Using CGSS survey data from 2008, Guo and Wang (2020) take propensity score approach and find that vocational upper secondary education attracts significantly higher private returns, especially for academically low-performing students, compared to academic upper secondary education. The overall premium of attending vocational education is around 8–8.4%. Loyalka et al. (2016) carried out a two-wave longitudinal study involving 10,071 first-year computer-majored students among 259 upper secondary schools in two provinces in China. They find that attending vocational upper secondary school negatively affects a student’s general skill by 0.30–0.44 SDs (as measured by math score) and does not significantly improve a student’s specific skills (as measured by computer skills). If anything, attending a vocational high school seems to detract a student from accumulating human capital. In the same vein, the authors also find that students in vocational schools are significantly more likely to drop out compared to students in academic schools, with a particularly high drop-out rate among low-income and low-ability students. A similar paper with a two-wave longitudinal study involving more than 12,000 students in one Chinese province has similar findings Yi et al. (2018). Using data of twins in urban China, Li et al. (2012) find that returns to upper secondary vocational education is around 22%, whereas return to academic upper secondary education is not statistically different from return of compulsory education. Their finding is in drastic contrast with the most recent findings from Dai and Martins (2020).

The mixed evidence in these studies can be attributed to several reasons. First, different datasets are utilized in various studies. For example, the data in Yi et al. (2018) are collected within one province in China whereas the data in Li et al. (2012) are based on urban twins, making their conclusions less generalizable. Wang and Wang (2023) only include post-secondary graduates from the CFPS dataset in their analysis, leaving out the individuals with secondary or below level of education. The data in Guo and Wang (2020) are the 2008 wave CGSS, which are collected more than 15 years ago and may not represent the current situation. By contrast, our study utilizes the more recent waves of the CGSS data and covers the years from 2010 to 2017. We also include individuals with all levels of education. In this sense, our study provides a more updated and holistic picture of the returns to education, especially vocational education, in China. Additionally, given the relatively long span of the data, we are able to gauge the dynamics of the educational returns over the years. Even if we assume the results based on the nationally representative samples (e.g., CGSS, CFPS) are largely comparable, the different estimated returns to education signal a potential shift in the supply and demand of vocational graduates due to the massive expansion and promotion of vocational education in the recent decades. All the above factors call upon further investigation regarding returns to vocational education, a gap that our research aims to fill in.

Theoretical framework

This paper is motivated by supply and demand theory from labor economics, with analysis guided by determinants of returns to education from the economics of education and discussions enriched by vocationalism-related theories. As was pointed out in several seminal works (Tinbergen 1975; Psacharopoulos 1989), the changes in returns to education are results of the race between technology and education. Specifically, technology development shifts the demand curve for educated labor to the right, although educational expansion shifts the supply curve of educated labor to the same direction as well, leaving the equilibrium returns to education undetermined and “an empirical matter” (Psacharopoulos 1989). In empirical studies, the returns to education are found to relate to a list of factors (more details are provided later in this section). Vocational education exhibits its uniqueness by virtue of its pronounced result-orientation nature and its proclivity to closely mirror economic fluctuations. In subsequent paragraphs, we discuss each of the three strands in greater details, linking to China’s specific context.

In labor economics, a competitive labor market will reach its equilibrium wage and equilibrium employment when labor supply equals labor demand (Borjas and Van Ours 2010). However, due to endless shocks in a modern industrialized economy, the labor market is unlikely to reach a stable equilibrium. Rather, wages and employment tend to be reaching new equilibria. As can be seen from Fig. 3, labor supply and labor demand jointly affect the returns to education while supply and demand are both constantly affected by other factors. Labor supply is affected by the educational background of the labor force, whether the education system is vocationalism in its policies and practices, the education policies implemented by the government, and the technological advancement in the economy. For example, although the demand of skilled workers has upsurged in China after the Reform and Opening-up policy in 1978, the returns to education may not have risen accordingly in recent decades due to the supply boost of highly educated labor force by the higher education massification. Labor demand is more affected by the latter two factors, i.e., policy intervention of the government and technological advancement, because the educational background of individual workers and the vocationalism practice of educational institutes is unlikely to directly affect the demand side.

Fig. 3

Theoretical framework.

Returns to investment in education can be affected by various factors. Early works in the economics of education focus on the relationship between education investment and labor outcomes. Controlling for individual demographics is of vital importance in estimating the effects of education (McCaffrey et al. 2004; Castellano et al. 2014). Such factors include gender, social economic status, marital status, union membership etc. (Angrist and Krueger 1992; Woodridge 2010; Castellano et al. 2014). In addition to these generic demographics, returns to education are also affected by heterogeneous factors that vary across contexts. For example, hukou is an influencing factor which is unique to China. Thus, we need to take hukou into account when estimating the returns to education.

Vocationalism is particularly suitable when discussing the dynamics of returns to education to China, because the changes in returns to education in China largely resonate the country’s tremendous economic development and industrial upgrading. Vocationalism means that the design and operation of the educational system reacts to the operation of the economic system (Bills et al. 2016; Xiong 2011; Xiong 2013). More specifically, vocationalism involves an array of practices and policies aiming to maximize the occupational value of schooling, by meeting different skill requirements of employers. It encompasses, but is not limited to, vocational education. In the context of this paper, all different levels of education speak to the demand of the employers in some way. For the older generations who were born when education was not compulsory, they may enter the job market with very limited (or even no) education. At that time, China was rather underdeveloped with recovering agriculture and early-stage industry, both of which were labor-intensive and did not require much professional knowledge of the labor force. By contrast, those who were born after the one-child policy may not enter the job market until their late 20 or 30 s due to overeducation and heavy investment into the single child by the entire family (Li et al. 2008). The overeducation situation was exaggerated by the higher education expansion around 2000, adding uncertainties to the returns to education. At this stage, China’s industry has seen huge growth and is becoming a leading power in many sectors worldwide, attracting talents from all over the globe. Consequently, educational requirement of the labor force has drastically increased while the returns to education are becoming more dispersed.

In sum, the dynamics of returns to education, especially vocational education, in China is a multi-disciplinary matter that does not induce a definitive answer. Theories across various disciplines are considered and discussed to inform us on the various forces that contribute to the dynamics. Figure 3 provides a holistic theoretical framework which depicts the interplay among the different theories that eventually affect the returns to education. The holistic theoretical framework in this section can be used as a guide for reading the empirical evidence.

Methodology

Data description

This paper utilizes quantitative data from the Chinese General Social Survey (CGSS), one of the nationally representative surveys in China. It is a repeated cross-sectional survey project that was launched by the Renmin University of China in 2003. The survey adopts a multi-stage stratified sampling method, covering households from all 31 mainland provinces.³ Starting in 2010, a modified sampling method is adopted to collect a more representative sample nationwide. In this paper, we analyze six annual survey datasets collected from 2010 to 2017 (except for 2014 and 2016, due to data unavailability). The 2017 data was released in 2020 and is the latest dataset available. After dropping observations without annual wage entries, we are left with a total of 27,545 observations. Table 1 summarizes the main variables of interest. The average hourly wage across all six years is 15.391 yuan, consistent with existing literature (Asadullah and Xiao 2020; Dai and Martins 2020).

Table 1 Summary statistics of waged workers, 2010–2017.

In the summary statistics, the years of education is constructed by mapping the Chinese education system onto the International Standard Classification of Education⁴. Note that because it typically takes three years to complete vocational education at tertiary level and four years to complete academic education at tertiary level in China, the years corresponding to vocational education are one year shorter than those of academic education. Following the convention in the literature (Kang et al. 2021), vocational tertiary education is referred to as vocational college and academic tertiary education is referred to as academic university throughout this paper. In our econometric analysis, we treat the education qualifications as dummies. The actual years of education are irrelevant when we focus on the two types of upper secondary education, both of which typically last three years. The years of education is controlled when we investigate other education qualifications.

The average years of work experience for a waged worker are 29.08. Female composes about 41.2% of the sample. Ethnically, only less than 9% are from a minority group. 60.3% of the sample have an agricultural hukou. 15.0% of the sample are union members, whereas 11.6% are communist party members. The average years of education are 5.72 for father and 3.97 for mothers. 82.6% of the sampled are married and 10.5% remain single. As for education qualifications, more than 56.3% of the sample only attended compulsory or less education. 8.0% finished vocational upper secondary education, whereas 13.1% finished academic upper secondary education. Roughly the same proportion - around 10.7% - of vocational college graduates and academic university graduates are observed in the sample. Only 1.3 out of 100 completed postgraduate and above level of education. In terms of occupation sector, more than 33.9% of the sampled subjects are in the agricultural sector. More than 38.3% work in privately owned enterprises. 20.2% work in state owned enterprises, followed by collectively owned enterprises, foreign funded enterprises, and Hong Kong, Macau or Taiwan funded enterprises. Location-wise, 37.6% are from the eastern provinces, 26.5% are from the western ones, 23.1% are from the central, and 12.8% are from the northeast provinces. ⁵

Figure 4 shows the relationship between change in employment rate and the share of individuals with certain credential. Each dot corresponds to an industrial sector categorized following ISCO2008 two-digit codes.⁶ In all four figures, the vertical axis represents the sector-wise change in employment from 2010 to 2017. The horizontal axis captures the share of tertiary graduates (in figures a and c) or upper secondary graduates (in figures b and d). Figure 4a, b are for all the working age subjects, whereas c and d are for the young cohort aged 24–35. Comparing the figures horizontally, we see that the majority of sectors cluster around the left end in a and the distribution spreads more evenly in b. In other words, in most of the sectors, upper-secondary credential holders are relatively common whereas university degree holders are rare (mostly occupying less than 10% of the population in a sector). Comparing the figures vertically, the share of tertiary graduates is less clustered for the young cohort as opposed to the entire working age group, indicating that workers with tertiary degrees are more common in most sectors; the share of upper secondary graduates are more clustered on the right end for the young cohort as opposed to the entire working age group, indicating upper-secondary credential holders also become more common. Additionally, the figures manifest one constantly expanding sector (i.e., 52 - Sales Workers) and one constantly shrinking sector (i.e., 1 - rural agriculture). The expansion of the sales sector echoes the massive growth of Chinese market, especially the housing market, in the past decade. The compression of the rural agriculture sector corroborates the well-known rural-urban migration in China.

Fig. 4: Change in employment rate vs. share of workers with university/higher secondary credential.

a All working age subjects - tertiary graduates, b All working age subjects - upper secondary graduates, c Young cohort (age 24-35) - tertiary graduates, d Young cohort (age 24–35) - upper secondary graduates.

Baseline estimation

To estimate the returns to different education qualifications across different years, we apply baseline estimation strategy in this section. Because we are also interested in comparing the returns to same-level academic and vocational education, we apply various other techniques to address endogeneity problem and identify causality in the comparison, which simultaneously serve as robustness checks of the baseline comparison. These additional techniques are explained in Section “Robustness check techniques”.

Our baseline estimation strategy is the extended Mincerian equation (Mincer 1974; Duraisamy 2002; Psacharopoulos 1994). The classical semi-logarithmic functional form is as follows.

$$Y = \alpha + \beta _iedu_i + \gamma X + \varepsilon$$

where Y is the natural logarithm of hourly working income. edu_i is a vector of dummies corresponding to different education qualifications, with the baseline group being those who have completed the 9-year compulsory education at most. X is a list of controlled covariates including years of work experience, experience-squared, gender, marital status, ethnicity, hukou status, union membership, party membership, the occupation sector, and the province/location in which one completed the survey. Specifically, province fixed effect is controlled in the OLS regression. In all later regressions, provinces are grouped into four regions to avoid over-controlling dummies. We also control for the interaction between marital status and gender, as existing literature has demonstrated marital status affects men and women’s earnings differently (Antonovics and Town 2004; Juhn and McCue 2017; Chen and Pastore 2021). We additionally control for parents’ years of education, as Card (1999) noted that family background measures such as parental education typically positively affect earnings.

When education qualifications are controlled as dummies, scholars typically utilize the following formula in calculating the yearly returns corresponding to a specific education qualification (Duraisamy 2002).

$$r_i = \frac{{\beta _i - \beta _{i - 1}}}{{t_i - t_{i - 1}}}$$

where r_i represents the yearly return rate to education qualification edu_i. β_i and β_i−1 are the returns to education qualification at the i th and i − 1th level. t and t − 1 are the years of schooling corresponding to their respective qualifications. In our case, both the vocational and the academic upper secondary education typically last three years. Both types of upper secondary education also share the same baseline group (i.e., compulsory and below level of education). Thus, comparing the statistical difference between r_i and r_i−1 is equivalent to comparing β_i and β_i−1 directly. Clearly, if one is to obtain the yearly returns, one still needs to convert β_i to r_i.

Robustness check techniques

As robustness checks of the baseline estimation, three different econometric techniques are applied: the instrumental variable method, the Lewbel method, and the propensity score matching method⁷. Each method adds insights to our baseline results.

The Instrumental Variable (IV) method is helpful since the Durbin and Wu–Hausman test weakly rejects the hypothesis that the choice of vocational upper secondary education against academic upper secondary education is exogenous. IV method is often susceptible to the choice of IVs. To provide robust IV estimates, we try several seemingly feasible IVs in this paper and compare their estimates. The first IV is the proportion of university (including colleges) graduates over the total population by year. Figure 5 plots the proportion of university graduates over the total population from year 1978 to 2019. We see that the proportion of university graduates increased slowly before 2000 and surged after that. The dramatic change is due to the higher education expansion in 1999 (Meng 2012; Meng et al. 2013; Zhong 2011). This variable is exogenous in that any individual decision cannot affect the proportions over the years. This is particularly true if we think about the year in which the expansion policy is announced: the change in proportions is mainly out of a policy shock. The variable is relevant in that the changes in the proportion of university graduates will relate to the changes in upper secondary graduates. Hence, the proportion of university graduates over the total population is a valid instrument.

Fig. 5

Proportion of university graduates relative to total population.

The second instrument is the upper secondary enrollment ratio, the number of vocational upper secondary students enrolled over the number of lower secondary graduates, by province. Enrollment ratio has been used as instrument in Dai and Martins (2020) and Chen and Pastore (2021). To construct the enrollment ratio, we first extract the number of lower secondary graduates and the number of students enrolled in vocational upper secondary schools from the Educational Statistics Yearbook of China from 1987 to 2015. Then we calculate Eratio_vus, which is the average of province-level enrollment ratios each year. A province-level enrollment raio is defined as the number of students enrolled in vocational upper secondary schools in year t in a province divided by the number of lower secondary graduates in year t in the same province. A graphical representation of the Eratio_vus can be found in Fig. 6. As a robustness check, we have also replicated Fig. 7 in Dai and Martins (2020), which can be found in Fig. 7 in the Appendix. The quota is relevant in that it can predict the probability of a student entering the vocational upper secondary education in a specific province in a given year. The quota is exogenous because there are strong reasons to believe that enrollment ratio does not affect one’s income several years later (Chen and Pastore 2021).

Fig. 6

Enrollment ratio.

Fig. 7

Enrollment ratio (replication of Fig. 7 in Dai and Martins 2020).

In the third IV specification, we apply the conventionally used parental education and compare the results across three specifications. The purpose of doing this is to benchmark our results to more conventional specifications and facilitate understanding of our results. Similarly, the four IV is a dummy which equals one if the enrollment year to upper secondary education is after 1985 and zero otherwise. This IV is identical to that in Guo and Wang (2020). The rational of this IV is that a policy shock that supposedly boosts enrollment to vocational upper secondary education is announced in 1985. The IV is included for comparison purpose as well.

In addition to the conventional IV method, we also apply a heteroskedasticity-based instrumental method introduced in Lewbel (2012); Baum and Lewbel (2019). This method can utilize the heteroskedasticity in the error term of the reduced form equation to construct instruments when no external instruments are available. It can also serve as a robustness check and validity test of external instruments (Baum and Lewbel 2019).

When comparing the returns to vocational upper secondary and academic upper secondary education, one challenge is that students from those two tracks may differ in various characteristics before they enter the upper secondary education. For example, boys may prefer vocational education while girls may prefer academic education, or rural students may prefer vocational education while urban students prefer academic education. To maximize the possibility of comparing the like with the like, we also apply propensity score matching (PSM) method. The PSM method assigns observations into the treatment group and the control group, then it estimates propensity scores for each observation based on pre-treatment characteristics, and comparisons are done between observations with closely matched propensity scores. In this paper, nearest neighbor matching is used, meaning for each treatment observation, one control observation with the closest propensity score is chosen for comparison. In our case, subjects are matched based on experience, gender, ethnicity, hukou, parental education, union membership, and party membership.

Findings and discussions

Dynamics of returns to different education qualifications

First, we find that the overall dynamic is that the returns to all types of education, both vocational and academic ones, have an increasing trend in the period from 2010 to 2017. Table 2 presents the OLS estimates of the extended Mincerian equation. Noting that we have controlled for a rich set of variables, we perform the multicollinearity test using the Variance Inflation Factor. No multicollinearity is detected. Overall, compared to the baseline group who only complete compulsory education, those who complete vocational upper secondary education earns 24.5% more, those who complete academic upper secondary education earns 19.5% more, those who complete vocational college education earns 51.4% more, those who complete academic university earns 75.3% more, and those who complete postgraduate and above level of education earns 108.7%. Clearly, the returns to education increase with the level of education qualifications. From 2010 to 2017, returns to all types of education are increasing. Yet, returns to upper secondary education increase more substantially (vocational: 25.1%⁸, academic: 77.5%) than returns to tertiary and above levels of education (vocational: 24.1%, academic: 10.0%, postgraduate and above: 16.0%).

Table 2 OLS estimates of ln hourly wage on education qualifications.

Second, we compare the dynamics of returns for different types of education. We find that, at the upper secondary level, returns to vocational education increase more slowly than returns to academic education; at the tertiary level, vocational education and academic education increase roughly at the same speed. Comparing vocational education with academic education, we see that, returns to vocational upper secondary education (24.2%) are higher than returns to academic upper secondary education (19.5%). As for whether the difference is statistically significant, we relegate the comparison to Section “Comparison of the dynamics of returns to vocational andacademic education”. Looking through the six years, returns to academic upper secondary education seem to increase faster than returns to vocational upper secondary education.

Third and to uncover further nuances, we explore the dynamics of returns in subgroups with different gender, union status, hukou status, ethnic group, and occupation. From the coefficients corresponding to marital status, we see there is almost a “wage transfer” from married females to married males. Specifically, if one is married and male then he gets a 28% increase in wages compared to a single male, whereas a married female gets a 21% decrease in wages compared to a married male. From 2010 to 2017, males seem to enjoy an increasing marriage premium from 23.2% to 30.9%, grown by more than a third. This implies that males are potentially even more highly valued in the Chinese labor market nowadays than ten years ago, a strong signal of workplace gender inequality. This finding is in line with a recent meta-analysis on gender wage gap in China (Iwasaki and Ma 2020), which states that in recent years the gender wage gap in China has grown rapidly.

Looking further down the table, we find that union membership no longer brings additional returns starting from 2015. Those with agricultural hukou tend to suffer a certain level of wage penalty. The minority ethnic groups are paid comparably to the Han ethnic group, indicating no ethnical discrimination. Party membership is related to a lower wage premium of 7% only when all the years are pooled together. Returns are generally positively related to parental education, in line with the mainstream literature. Compared to the baseline agricultural jobs, all five types of enterprises pay substantially higher wages across all the years, implying that the payment situation in agriculture has not improved much over the years. Probably, this is also why the Chinese government initiated the “rural revitalization (xiangcunzhenxing)” strategy in late 2017.

Technically, because our OLS estimation is essentially a log-linear transformation of an exponential function, we can interpret the coefficients as semi-elasticity. In other words, the coefficients measure the percentage change in the hourly wage in response to any unit change in the independent variable (Pastore and Verashchagina, 2006). Because the education qualifications are dummies, the semi-elasticity interpretation can be more precise when transformed via (e^β − 1) × 100 (Halvorsen et al. 1980). By its nature, the difference between the original estimates and the transformed coefficients gets larger with the coefficients increasing. The transformed coefficients are reported in Table 3. By and large, all the values based on the original semi-elasticity interpretation hold. However, we note the imprecision in the original estimates. As expected, the larger the coefficients, the more imprecise the semi-elastic measures are. For example, the returns to postgraduate and above level education is 108.7% in Table 2 but almost doubled to 196.6% in Table 3, underlying the importance of noting this caveat in future research.

Table 3 Returns to different education qualifications - transformed coefficients.

Comparison of the dynamics of returns to vocational and academic education

As mentioned in Section “Dynamics of returns to different education qualifications”, simple OLS regression may suffer from endogeneity issue, when estimating the returns to education. Thus, we apply various econometrics techniques as robustness check to the main findings.

Specifically, we do pairwise comparison of the returns to vocational education and those to academic education, both at the upper secondary level and at the tertiary level. Table 4 presents the coefficient difference test results of the transformed OLS coefficients, based on delta method. The coefficient difference is calculated as follows.

$$\beta _{vus}^ \ast - \beta _{aus}^ \ast = \left[ {\left( {e_{vus}^\beta - 1} \right) \times 100} \right] - \left[ {\left( {\left( {e_{aus}^\beta - 1} \right) \times 100} \right)} \right] = \left( {e_{vus}^\beta - e_{aus}^\beta } \right) \times 100$$

Table 4 Coefficient difference comparison.

From Panel A, we see that returns to vocational upper secondary education are 6.2% higher than returns to academic upper secondary education. The positive difference is in line with Dai and Martins (2020) in which a 20% wage premium is captured among vocational upper secondary graduates as opposed to academic graduates. Yet our estimated difference is much smaller in magnitude (6%) than that in Dai and Martins (2020). This is partly due to the fact that we use transformed coefficients for the education qualification dummies, which presumably generates a more precise measure of the difference. Since both vocational and academic upper secondary education typically last three years, the yearly difference is around 2.1%. Panel B shows that the difference in returns to vocational college and academic university is much more substantial, vocational college graduates earning 45.3% less than academic university graduates. As explained before, most vocational college programs last three years and academic university programs last four years. When comparing the yearly difference, different denominators need to be used. The yearly difference is around 6% for the entire sample, comparable in magnitude to the benefit of obtaining an additional vocational upper secondary credential. Because the difference is both economically and statistically strongly significant, it is unsurprising that the pattern remains stable across the years. Notably though, the yearly return difference between vocational college education and academic university education keeps shrinking from 2010 to 2017. This again matches the higher education massification over the past two decades and the labor market mismatch in China - academic university degree is losing its additional charm slowly in the labor market and vocational college credential is catching up.

Meanwhile, comparisons are done for the difference between the two types of upper secondary education when all the years are pooled together. Panel A of Table 5 presents the IV estimates using the four sets of instrument variables discussed in Section “Methodology”. Because both types of upper secondary education programs last three years, we redefine the main treatment variable as a dummy which equals zero for academic graduates and equals one for vocational graduates. The first instrument is the proportion of university graduates by year, which is novelly constructed by the authors. The second instrument is the enrollment ratio, similar to the one in Dai and Martins (2020). The third set of instruments are parental education, which is conventionally used in the returns to education literature. The fourth set of instruments is identical with the one in Guo and Wang (2020), which is included for comparison purpose as well.⁹ Odd-numbered columns provide the F statistic of the first stage. Even-numbered columns present the coefficient of interest estimated from the second stage. As a rule-of-thumb, an acceptable F statistic of the first stage should be greater than 10 (Staiger and Stock 1997). As can be seen, only our novel instrument has an F statistic larger than 22, well above the conventional threshold of 10 for a qualified instrument. Interestingly, with the novel instrument, we find that vocational graduates suffer a wage penalty of close to 70% compared to academic graduates. With the other three sets of instruments, the estimated coefficients are positive. However, the weak first stage or failing to pass the Durbin test cast doubt on the validity of these positive estimates. In sum, the IV estimation instrumented by the proportion of university graduates provides evidence that vocational graduates are actually paid less well compared to academic graduates.

Table 5 Vocational vs. academic upper secondary education.

To further check the robustness of our IV estimates, we utilize the Lewbel method and the PSM method. The definition of the treatment variable and the external instruments in the Lewbel model are the same as those in the corresponding IV specifications. Note that the Lewbel method additionally incorporates all the remaining control variables as internal instruments. Results from the Lewbel regressions are reported in Panel B of Table 5. We see that all four specifications pass the Breusch-Pagan test with strong statistical significance, validating the usage of the Lewbel method. As expected, the results instrumented by the proportion of university graduates (Model 1) remains negative and statistically significant. Results in the other three specifications are insignificant, raising further concern towards instrumenting vocational upper secondary graduates using its enrollment ratio.

Turning to the PSM estimates in Panel C of Table 5, there is a lack of statistical significance which may again result from the dramatic reduction in the number of observations. This in itself has interesting implications, suggesting that the overall samples are very different from each other. In other words, the difference in returns to education between vocational and academic education is essentially a matter of the different samples. This is in line with the idea that unfortunately people going into vocational education are the worst in terms of skills and talent. This adds to the story, is not in contrast with the story. Meanwhile, the PSM result is interesting when compared with the positive difference which was found in the OLS estimates. One possibility is that the alternative to vocational education for academically low performing students is academic upper secondary education, but they perform even worse in this case by moving to academic education. Without matching, we may have underestimated the returns to academic upper secondary education because a subgroup of academic students are the would-be low performing vocational students. With matching, we have a like for like comparison. Now that the three different econometric techniques generate three drastically different answers, we conclude that the return difference between vocational and academic upper secondary education cannot be determined as far as the current paper goes. On a related note, readers will benefit from reading this paper jointly with the paper by Dai and Martins (2020) for the following reasons. First, both studies use the same CGSS dataset, although ours additionally includes a new round. Second, the second IV specification in the current study is almost identical to the one in Dai and Martins (2020) (see Fig. 7 for a reproduction of Fig. 7 in Dai and Martins (2020)). Third, the OLS estimates in both studies show that vocational upper secondary graduates earn a wage premium over academic upper secondary graduates. Fourth, Dai and Martins (2020) find no difference via IV estimations whereas we find a negative difference using a stronger IV. Last but not least, we attempt Lewbel and PSM methods to further investigate the overall difference (albeit without obtaining a determinant answer) whereas Dai and Martins (2020) switch their focus onto local differences utilizing quantile regressions.

To sum up, from the comparisons of vocational and academic education, we have the following main findings. First, academic university education degree pays around 45% better than vocational college education credential, meaning each additional year of academic university education earns 6% more than each additional year of vocational college education. This indicates the higher level of higher education can get higher returns than higher education at a lower level since vocational college education is located at junior college (zhuanke) level. Second, from 2010 to 2017, the payment premium of academic university education shrinks constantly. After 2015, yearly returns to academic university education no longer strongly differ from those to vocational college education. It seems that, with increasingly strong policy support, higher vocational education has gained major development and has started to gain social credibility and acceptance. The indication is that vocational education is gaining higher social status. Meanwhile, although OLS estimates show that vocational upper secondary graduates enjoy a 6% wage premium over academic upper secondary graduates, the difference does not survive alternative specifications (IV, Lewbel, PSM). In subsample analysis, we also find evidence that married men enjoy increasing wage premium compared to married women during 2010–2017. This signal of enlarging gender inequality merits further investigation, which is beyond the scope of the current paper.

Conclusion

In this paper, we investigate the dynamics of returns to different education qualifications in China. We also compare the returns to vocational education and returns to academic education, both at the upper secondary level and at the tertiary level.

We have the following main findings. First, compared to the baseline group with compulsory education and below levels of education, upper secondary graduates enjoy a wage premium of around 20%, vocational college graduates enjoy around 50% premium, academic university graduates enjoy 75.3% premium, and those who complete postgraduate and above levels of education more than double their returns. Second, compared to academic upper secondary education, vocational education earns a wage premium of around 6% according to OLS estimates, although not robust to alternative specificaitons. Third, compared to academic university education, vocational college education pays 45.3% less in returns, both economically and statistically strongly significant. As mentioned before, this is evidence that higher levels of higher education pays better than lower levels of higher education. Hence, a system extending to undergraduate level may help improve returns of higher vocational education. In fact, since 2014, China has started to expand higher vocational education to undergraduate levels. Fourth, during the period from 2010 to 2017, returns on investment in education at all levels exhibited an upward trend. Notably, returns to upper secondary education witnessed a more substantial increase compared to those of tertiary and higher levels, relative to their respective baselines, signaling the after-effect of higher education massification after the millennium. Last but not least, we find that males enjoy an increasing marriage premium over the study period.

Alongside the new findings, the novel econometric techniques we adopt in this paper will hopefully enrich the toolbox for researchers quantitatively studying vocational education. First, a novel instrument (i.e., the proportion of university graduates relative the entire population over the years) is introduced to account for endogeneity in the choice of vocational education against academic education at the upper secondary level. Evidence shows that this instrument outperforms the conventional parental education instruments, the enrollment ratio instrument which was used in Dai and Martins (2020), and the policy shock dummy in Guo and Wang (2020). Second, heteroskedasticity-based Lewbel method is used to fully utilize both the externally constructed instruments and internal instruments (i.e., the full set of control variables or a subset). Third, propensity score matching method is applied to match vocational students with academic students who have similar characteristics, so that self-selection bias can be partialled out to a certain extent. Nonetheless, the paper has its limitations. One is our sample captures people across all the working age. People in different generations may face completely different social and economic conditions. Readers need to bear this caveat in mind when interpreting the results. Another is that our sample size is not so large relative to the 1.4 billion population in China.

Although we do not reach a definitive answer when comparing vocational and academic upper secondary education, it is undoubtful that vocational college graduates are paid significantly lower than academic university graduates. Meanwhile, we observe that vocational education graduates are increasingly getting better paid in recent years. On the one hand, the payment disadvantage facing vocational graduates can be explained by the specific context in China. The Chinese government’s early effort in encouraging vocational education is countered by the country’s long-standing tradition of degrading vocational graduates. This conflict puts vocational graduates into a seemingly vulnerable position compared to academic graduates. On the other hand, the increasing social credibility, and consequently returns, of vocational education demonstrates the achievement of the increasing policy support by the Chinese government. ¹⁰In the foreseeable future, the Chinese education sector will continue paying more attention to vocational education and training. Understanding how the returns to vocational education change as China transits its industry can better prepare other countries that face a similar situation.