Higher Education Expansion and Return to Education in China: Evidence from CGSS2005 and CGSS2013

We conducted an empirical study to estimate the private internal rate of return to years of schooling (IRR) in China during the period after the implementation of higher education expansion policy using data from the Chinese General Social Survey data conducted in 2006 and 2014 (CGSS2005, CGSS2013). The major conclusions are as follows: first, from 2005 to 2013, IRR decreased from 8.6% to 7.8% for the whole sample, IRR decreased from 8.3% to 7.4% for men, and IRR decreased from 9.0% to 8.2% for women. Second, IRR values among various education category groups are different. IRR is greater for the high-level education group than that for the middle and low-level education groups in both 2005 and 2013. Third, to consider the impact of the higher education expansion policy on IRR, the IRR of the university graduates decreased from 15.4% (2005) to 11.2% (2013), whereas the IRR of the graduate school graduates rose from 10.1% (2005) to 19.0% (2013). The effect of the policy on IRR differs between the university and graduate school graduates. Fourth, the IRR is higher for women than for men. There is a gender disparity for IRR; IRR is different by ownership types, registration system types, industrial and regional groups in both 2005 and 2013.

The contributions of this study are as follows. First, we analyze the changes in return to education from 2005 to 2013, using the latest survey data (CGSS2013) which provides new evidence. 2005 to 2013 is relevant as it is the period following the implementation of the higher education expansion policy. We also discuss the impact of the policy on IRR. Second, IRR is calculated by various education category groups. As 2005 to 2013 is the period following the implementation of the higher education expansion policy, to compare the change of IRR for high-level education graduates, we discuss the impact of the higher education expansion policy on wages. Third, the gender disparity of IRR is estimated to control the other factors (e.g. individual characteristics, employment status, occupation, industry, regions) in order to investigate the differentials of IRR by gender. Lastly, IRR is estimated by other groups: public sector and private sectors, industrial sectors (primary, secondary and tertiary industries) and regional groups (Eastern, Central, and Western Regions).

Models
First, to calculate IRR, this study uses the wage function based on OLS which is usually used in the previous studies (Note 6). It is expressed in equation (1). (1) In equation (1), is the dependent variable (logarithmic value of wage); denotes individuals; is schooling years; denotes experience years; are the other factors which affect the individual wage (e.g. gender, occupations, employment status, industry sectors, and regions); indicates constant; and is error item.
Then, to estimate IRR by various education category groups, the Psacharopoulos (1981) model is utilized. The model is expressed in equation (2) and (3).
In equation (2), ~indicates the different education levels. In the study, they are a set of education dummy variables (no schooling, primary school, junior high school, senior high school, college and vocational school, university, and graduate school).
is similar with that in equation (1). is estimated coefficients. Based on the set of coefficient of ~, the IRR by education category groups can be calculated by equation (3): ISSN 1923-4023 E-ISSN 1923 IRR of primary school: IRR of junior high school: IRR of senior high school: IRR of college: IRR of university: IRR of graduate school: ( Lastly, to investigate the gender disparity of IRR, two kinds of methods are used. First, to consider the individual characteristics disparity by gender, subsamples female workers and male workers are used separately. This method is usually used in the published studies (Li and Ding, 2003;Zhang, et al., 2005;. However, the results from this method cannot investigate if, when the other factors (e.g., the occupation, industry sector or employment status) are consistent, the IRR differs by gender. To address the problem, the second model is used: it is expressed in equation (4).
In equation (4), indicates the gender dummy variable (equal to 1 if male, equal to 0 if female), expresses the interaction item of and . is estimated coefficients. When is statistically significant, it indicates that even though the individual characteristics are similar, IRR differs by gender. When is negative statistically significant, it indicates that IRR is greater for women than that for men.
We also use a set of subsamples to estimate the IRR and compare the disparities of IRR among these groups: i. public sector (government organization, state-owned enterprises) and private sector groups; ii. Eastern, Central, and Western Region groups; and iii. the rural resident and urban resident groups.

Data
This study employs two periods of the Chinese General Social Survey (CGSS) survey data.
Ge was conducted in 2006 and the information for wages and jobs for 2005 which is the prior period of new workers graduated from the college, university or graduate school of university after the policy implementations; because CGSS2013 is the most recent survey, it will give us the most up to date information about the issue. The samples are composed of 10,372 (CGSS2005), and 11,438 (CGSS2013) individuals in 26 provinces and municipal cities, which covers nearly the whole of China. The CGSS includes respective information about individual characteristics (e.g., education, experience year, gender, and marital status), job information (e.g., employment status, wage, occupation, industry, and work place).
The self-employed, retired workers, and the unemployed are excluded because the subject of the study is employees. A retirement system and employee basic pension system have been implemented in the state-owned sector in China (Note 8). To reduce the effect of these systems on the analysis result, only those between the ages of 16 and 60 are included.
The dependent variable for the wage function is the logarithm of the annual wage (Note 9). The wage is defined as the total earnings from work (called "the total wage"). We use the CPI (consumption price index) in 2005 and 2013 to adjust the nominal wage. The explaining variables are the variables likely to affect the wage, such as years of schooling or education category dummy variables (no schooling, primary school, junior high school, senior high school/vocational school, college, university, and graduate school) (Note 10), experience years (Note 11), male dummy variable (equal to 1 if male, equal to 0 if female), communist party member (equal to 1 if communist party member, equal to 0 if not), regular worker dummy variable (equal to 1 if regular worker, equal to 0 if not), the married dummy variable (equal to 1 if married, equal to 0 if not), Han dummy variable (equal to 1 if Han majority, equal to 0 if minority), ownership (government organization, state-owned enterprises, private enterprises) dummy variables (Note 12), occupation (manager, technician, clerk, manual worker, the other) dummy variables, industry (primary, secondary and service and tertiary industries) dummy variables, rural registration dummy variable (equal to 1 if a worker with rural registration, equal to 0 if a worker with urban registration), and region (Western, Central, and Eastern Region) dummy variables.
The statistical description of variables is shown in Table 1  Date source: Calculated based on CGSS2005 and CGSS2013. Table 2 summarizes the mean value of annual wage and wage gaps between the various education category groups. For wage gaps between various educational category groups, the reference group is the wage mean value of senior high school (the middle-level education group). The main findings are shown below.
First, compared with workers graduated from senior high school, the wages are lower for no-schooling workers, and workers graduated from primary school and junior school (the low-level education group). Wages are higher for workers graduated from college, university, and the graduate school of university (the high-level education group) for both male and female workers in both 2005 and 2013. The results show a wage gap between various education al category groups in both 2005 and 2013.
Second, the wage gap between the various educational category groups differs by gender. For example, the wage gap between the middle-level education group (senior high school) and low-level education group (no schooling, primary school, junior high school, and senior high school) is greater for women than for men in both 2005 and 2013. The education wage gap between the middle-level education group and high-level education groups (college, university, and graduate school of university) is overall greater for men than that for women in both 2005 and 2013.
Lastly, the wage gap changes from 2005 to 2013 differ by gender. For example, for the female group, the wage gap between senior high school and university decreased (from 2.36 in 2005 to 1.16 in 2013), whereas the wage gap between senior high school and the graduate school of university increased (from 2.75 in 2005 to 3.11 in 2013). However, for the male group, the wage gap between senior high school and university increased (from 1.57 in 2005 to 1.63 in 2013), whereas the wage gap between senior high school and the graduate school of university decreased (from 3.55 in 2005 to 3.21 in 2013). These results may be caused by the labor supply and demand for high-level education workers differing by gender.  Table 3 summarizes wage function results based on equation (1). The coefficient of years of schooling is estimated IRR. Estimation (1) uses the total sample (male + female), Estimation (2) uses subsamples: male or female. The main findings of Estimation (1) are as follows.

Results of Returns to Schooling
First, the coefficients of years of school are 0.086 (male 0.083, female 0.090) for 2005, and 0.078 (male 0.074, female 0.082) for 2013. It is shown that from 2005 to 2013, IRR decreased from 8.6% to 7.8% for the total sample, from 8.3% to 7.4% for men, from 9.0% to 8.2% for women. These results are similar to Heckman and Li (2004), Zhang et al. (2007), Giles, Park and Wang (2008), Chen and Hamori (2009), Ge and Yang (2011), Kang and Peng (2012), Liu and Zhang (2012), Ren and Miller (2012). As shown in Appendix Table 1, the IRR values in these studies are in the range from 1.4 (Byron and Manaloto, 1990) to 44.0  in urban China. The estimated IRR in the study are among the values in previous studies.
Second, the IRR decreases from 2005 to 2013 for both men and women. The results may be caused by the labor supply of high-level education workers increasing greatly with the implementation of the higher education expansion policy. In the other wards it is indicated that there may exist an over education problem for the high-level education group in China after the implementation of the higher education expansion policy. ISSN 1923-4023 E-ISSN 1923 from 2005 to 2013. These results are consistent with Gustafsson, and Li (2000), and Li and Ma (2015). (2) Wages are higher for the more experienced year group in 2013. (3) To compare with minority group, the wage is 15.8% higher for the Han majority group in 2005, whereas the Han majority dummy variable is not statistically significant in 2013. A wage gap occurs between minority and majority groups in 2005, whereas the wage gap became smaller in 2013. (4) The wage is 39.2% (2005), 17.0 % (2013) higher for a regular worker than for an irregular worker. Although there is a wage gap between the regular worker and the irregular worker groups in both 2005 and 2013, the wage gap became smaller from 2005 to 2013. This might be because the economic transition, the impact of market-mechanism on wage determination in the public sector became greater, thus the wage gap between the public sector and the private sector decreased. (5) The influence of marital status on wage is not statistically significant in 2005, whereas, in 2013, the coefficient of the married dummy variable is 0.157 for male, and -0.092 for female, and they are statistically significant at a 1~10% level. It is indicated that the discrimination by maternal status for women increased from 2005 to 2013. (6) 8) The wage is higher for urban residents than that for rural resident in 2005, while the wage gap is greater for rural resident than that for urban residents at a 10% statistical level in 2013. It is shown that the wage gap by the registration system became smaller from 2005 to 2013. (9) There are regional wage gaps in both 2005 and 2013. To compare the economic developed region (Eastern Region), the wage is lower for the economic developing region (Central region and West region). (2) shows the IRR differs by gender, and the influences of individual characteristic and sector dummy variables on wage also differ by gender. We will discuss more details on gender disparity of IRR in the following (section 4.3). Note: 1. *,**,*** denote statistical significant in 10%,5%,1% level.

Values in brackets are estimated standard deviations.
Source: Calculated based on CGSS2005 and CGSS2013.

Values in brackets estimated standard deviation.
Data source: Calculated based on CGSS2005 and CGSS2013.
The IRR for the various education category groups is calculated based on equation (3), the results are summarized in Table 5. The main findings are as follows.
the other education category groups in 2005. The IRR of graduate school (19.03%) is higher than for the other education groups in 2013. These results are consistent with human capital theory, the higher the education level the higher the wage. (2) The IRR of university decreased from 15.35% in 2005 to 11.15% in 2013. The results might be because the labor supply of university graduates increased with the implementation of higher education expansion policy. Although it is thought the increase of high-level education workers may be due to the decrease of IRR for the high-level education group, when the labor demand for high-level education workers increased along with the explosion of technological innovation, the wage for the high-level education group may increase, which may cause the rise of the IRR for the high-level education group. The results in the study indicate that the influence of the labor supply side factor on the university group wage is greater than labor demand side factor. (3) However, the IRR of graduate school rose from 10. 1% (2005) to 19.0% (2013). It is indicated that the higher education expansion policy positively affects the wage for the highest level education group.
Second, to compare the return to schooling between the low, middle and high-level education groups, in 2005, the IRR is higher for the low-level and high-level education groups, and lower for the middle-level education group: these results are similar to Lou's (2009). However, in 2013, to compare with low-level education group, the IRR are higher for middle-level and high-level education groups. These results may be because after the world financial crisis in 2007-2008 the Chinese government implemented the industry upgrade policy to change industry from a low-technological level to higher level; and the technology innovations increased with economic growth. Thus the labor demand might have become greater for the middle-level education group (e.g. the junior and senior high school) than that for low-level education group in 2013.
Lastly, the IRR differs by gender for each educational category group, but the situations differ by period. For example, to look at the high-level education groups, (1) the IRR of graduate school is higher for men (14.8%) than for women (5.6%) in 2005; whereas it is higher for women (21.1%) than for men (17.4%) in 2013. It is clear that the increase of IRR of graduate school is greater for women than men. (2) Even though the IRR of university is higher for women (17.9%) than men (13.5%) in 2005, the IRR is the similar for women and men, both of them are 11.2% in 2013.  Table 5.

Robustness Checks: Estimation of IRR by Groups
(1) IRR by gender Table 6 summarizes the results of IRR by gender. Estimation (1) uses male and female groups separately based on equation (1). Estimation (2) uses the total samples including both male and female groups based on equation (4). The interaction of male dummy variable and school year is utilized in Estimation (2) which shows the gender disparity of IRR based on the assumption that the other factors (e.g. human capital) are similar. The main findings are as follows.
Second, the results in Estimation (2) show that when other factors are consistent, the IRR is 1.8% (2005), 1.6% (2013) lower for men than that for women.

Values in brackets estimated standard deviation.
3. Experience years, male, party member, the married, Han, regular worker, occupation (Manager, Technician, Clerk, Manual), industry (Primary, Secondary, Service industries), work place sector (government organization, SOEs), registration (the rural), regions (Eastern, Central Regions) dummy variables are also estimated, these results are not shown in Table 5.
Data source: Calculated based on CGSS2005 and CGSS2013.
Why is the return to education greater for women than men in China? There are three reasons.
First, based on Becker (1957), when the employer, colleague, or customer prefer a male to a female worker, even though the years of school are similar, the wage may differ by gender, which causes the gender disparity of IRR (discrimination hypothesis). Discrimination against female workers is widespread, and the gender wage gap expanded during the economic transition period in China (Gustafsson and Li, 2000;Maurer-Fazio and Hughes, 2002;Xing, et al., 2014;and Li and Ma, 2015).
Second, Zhang, et al. (2005) suggest it may be caused by self-selection. Based on the home productivity model, Becker (1985) pointed out that because men can earn a higher wage in the market than women, and women take more responsibility for child care and parent care than men, labor participation is lower for women than men. The scarcity of female workers based on self-selection may cause the IRR to be greater for women than for men (the self-selection hypothesis).
Third, when the proportion of high-level education workers are less in the female group than for the male group, a high-level education female labor supply shortage may occur, therefore to compare with the low-level education group, the probability of the high-level education group getting better jobs is greater for women than men, which may cause the IRR to be greater for women than men. In China, particularly in the rural region, because of the influence of Confucianism, boys might receive more preferential treatment than girls, therefore the boy might enjoy more intra-household resources (Liu, 2008). When the investment in education in a household is higher for a boy than for a girl, there will be fewer female workers with a high-level education entering the labor supply. The scarcity of female higher-level education workers may be because of the gender disparity of IRR (scarcity of female higher-level education worker hypothesis). The IRR values may be explained as follows. When the influence is greater for the self-selection hypothesis and the scarcity of female worker hypothesis than for the discrimination hypothesis, the IRR will be greater for women than for men. ISSN 1923-4023 E-ISSN 1923 (

2) IRR by ownership types, registration systems, industry and region category groups
To consider the samples heterogeneities by ownership types, registration systems, industrial groups and regional groups, the IRR are calculated for each group, and the results are summarized in Table 7. Estimation (1) analyses the total sample. Estimation (2), (3), (4), and (5) are analysed using subsamples separately, public sector and private sector in Estimation (2); urban resident group and rural resident group in Estimation (3); manufacturing industry and service industry in Estimation (4); and Western, Central and Eastern Regions in Estimation (5). The main results are as follows.
First, the IRR for government organizations is greater (9.5% in 2005, 9.7% in 2013) than for state-owned enterprises (SOEs) and the private sector in both 2005 and 2013; in 2005, IRR of SOEs (8.2 %) is smaller than for the private sector (8.5%), whereas in 2013, the IRR for SOEs (8.5 %) is greater than private sector (7.2%); the IRR for the public sector (government organization, and SOEs) rose from 2005 to 2013, whereas the IRR for the private sector decreased in the period (Estimation 2). These may be because the progress of SOEs reform, and the influence of market mechanisms on wages became greater in the public sector from 2005 to 2013.
Second, the IRR of urban residents (8.8% in 2005, 9.5% in 2013) is greater than that for rural residents (6.8% in 2005, 5.4% in 2013)  Third, in 2005 the IRR is smaller for secondary industry (7.8 %) than tertiary industry (8.6%), whereas in 2013 the IRR is greater for secondary industry (8.2 %) than for tertiary industry (7.3%). The IRR increased from 2005 to 2013 for secondary industry, whereas the IRR decreased for tertiary industry in the period (Estimation 4). This may be because the proportion of irregular workers is greater for the tertiary industry sector than for the secondary industry sector. More empirical studies of the wage gap between regular workers and irregular workers in China are needed.
Fourth, to compare the Eastern Region and the Western Region, the IRR of the Central Region is lowest in both 2005 and 2013 (6.7% in 2005, 3.5 % in 2013). Moreover, the IRR of the Eastern Region increased from 9.1% to 9.7%, whereas the IRR of the Central Region decreased from 6.7% to 3.5%, and the IRR of the Western Region decreased from 9.2% in 2005 to 5.5% in 2013. The results can be explained by the fact that the economic development level is highest in the Eastern Region, and the labor demand for the high-level education group is higher for the Eastern Region.
These results indicated that generally, IRR values differ between various groups, and the changes of IRR from 2005 to 2013 also differ by various groups. To look at the results for women and men separately, the disparity of IRR by various groups are similar with the results utilized the total samples. The IRR for all groups is greater for women than for men in both 2005 and 2013. The results confirm the gender disparity of IRR in China when considering the heterogeneities by groups. Note: 1. *,**,*** denote statistical significant in 10%,5%,1% level.

Values in brackets estimated standard deviation.
3. The other variables-experience years, male, communist party member, regular worker, the married, race, rural registration, industry sector, occupation, ownership, and region dummy variable are estimated, the results are not expressed in Table 8.
Data source: Calculated based on CGSS2005 and 2013.

Conclusions
Using the Chinese General Social Survey data conducted in 2006 and 2014 (CGSS2005, CGSS2013), this study takes an empirical study to estimate the private internal rate of return to years of schooling (IRR) in the 2000s. The major conclusions are as follows.
First, overall, from 2005 to 2013, the IRR decreased from 8.6% to 7.8% for the total sample, from 8.3% to 7.4% for men, and from 9.0% to 8.2% for women.
Second, to compare the IRR for the various education category groups, in 2005 the IRR are higher for low-level and high-level education groups, and lower for the middle-level education groups. In 2013, the IRR are higher for the middle-level and the high-level education groups than for the low-level education groups. The IRR of the high-level education group is higher than the low, middle and high-level education groups in both 2005 and 2013.
Third, to consider the impact of the higher education expansion policy implemented since 1999 on IRR, for the high-level education group: (1) in 2005 the IRR at university level is higher for the high-level education group than the other education category groups, and in 2013 the IRR at graduate school level is the highest of all education category groups. (2) The IRR at university level decreased from 15.4% in 2005 to 11.2% in 2013, whereas the IRR at graduate school level increased from 10.1% in 2005 to 19.0% in 2013. The policy negatively affected the IRR at university level, but positively affected the IRR at graduate school level.
Fourth, the results using subsamples of men and women show that the IRR is higher for women than men in both 2005 and 2013. The results using interaction of a gender dummy variable and school year indicate that when the other factors are consistent, the IRR is 1.8% lower for men in 2005 and 1.6% lower for men in 2013. These results denote a gender disparity of IRR in both 2005 and 2013.
Lastly, the IRR values differ by ownership types, registration systems, industrial and regional groups.
The policy implications of the study can be considered as follows. First, it is indicated that the IRR decreased for the university group, whereas the IRR increased for the graduate school group from 2005 to 2013. It is thought that after the higher education extension policy was implemented the labor supply of university graduates increased dramatically. This may have contributed to the problem of over-education for university graduates which may have contributed to the decrease of IRR at university level. However, the higher education extension policy may promote technological innovation and raise the IRR of graduate school graduates. More strict analysis needs to be done to evaluate the effect of the higher education extension policy on the labor market (e.g. higher-level education worker labor supply and demand, wage gap between high-level education group and middle-level, low-level education group) (Note 14) in the future. Second, the IRR is higher for women than men when the factors which affect wage level are controlled. It has been suggested that supporting more girls to receive high-level education may reduce the gender wage gap (Dougherty, 2005;Liu, 2008;. Indeed, the admission rate to middle-level and high-level education is lower for girls than for boys in rural regions. In order to build an equal society, the education policy may be amended by the government to increase public education investment in rural regions and to reduce the gender education gap.