Returns to Schooling in Urban China, 2001-2010: Evidence from Three Waves of the China Urban Labor Survey

This study provides estimates of the returns to schooling in urban China for migrants and nonmigrants using three waves of the China Urban Labor Survey (CULS), corresponding to 2001, 2005 and 2010. We find that the returns to schooling increased about 2-3 per cent between 2001 and 2010. The two-stage least squares (TSLS) estimates, using spouse’s education as an instrumental variable, are slightly higher than the ordinary least squares (OLS) estimates, although TSLS estimates using an internal instrument constructed from the heteroskedasticity in the data are similar to the OLS estimates. We find that returns to schooling are higher for non-migrants than migrants and higher for males than females over the decade. * We thank Paul Miller and Qingguo Zhai for helpful comments on earlier versions of this paper. Wenshu Gao: Institute of Population and Labor Economics, Chinese Academy of Social Sciences Russell Smyth Department of Economics, Monash University, Australia Telephone: +(613) 9905 1560 E-mail: [email protected] © 2012 Wenshu Gao and Russell Smyth All rights reserved. No part of this paper may be reproduced in any form, or stored in a retrieval system, without the prior written permission of the author. Introduction Since the introduction of economic reforms in the late 1970s, China‟s economy has transformed from a socialist planned economy into a predominantly market based system. Over this period China has experienced phenomenal economic growth, such that by 2010 China was the world‟s second largest economy (World Bank, 2012). During the 1980s and first half of the 1990s wage differentials were slow to expand. However, since the mid-1990s the deepening of economic reform in urban areas has enhanced flexibility and diminished administrative control in the workplace. As a result, wages have become increasingly more responsive to productivity and wage differentials have increased. The educational level of the labor force may explain the overall rate of economic growth (Griliches, 1970). It has been proposed that one important reason for China‟s rapid economic growth is that its labor force is relatively well educated (see eg. Fang et al. 2012). The education of China‟s labor force (aged 15 plus) has increased over the course of market reforms, from an average of 4.7 years in 1980 to 8.2 years in 2010. 1 The rising education of the labor force, combined with enhanced urban labor market flexibility, suggests that returns to schooling in China may have also increased over the period of the market reforms. Studies have documented expanding wage differentials and rapid increases in income inequality in urban China in the 1990s, which have somewhat plateaued since the 2000s (see eg. Meng et al. 2012). Changes in returns to schooling may be an important contributing factor in explaining changes in urban income inequality in 1 See http://barrolee.com (accessed on October 5, 2012). China through the 2000s (Meng et al., 2012; Wang, 2012a). Moreover, returns to schooling can provide information about the efficiency of resource allocation, the incentives for human capital accumulation and the distributional consequences of differences in human capital (Zhang et al., 2005). Heckman (2003) suggested that historically China has invested much more in physical capital than human capital because of differences in the returns to human and physical capital. The purpose of this paper is to estimate the private returns to education using three waves of the China Urban Labor Survey (CULS) for 2001, 2005 and 2010. We contribute to the literature in three ways. First, we use a representative survey, administered in a consistent manner, to examine changes in returns to schooling over the first decade of the twenty-first century. Most studies of the returns to schooling in urban areas use a single year of data. Differences in the peculiarities of datasets and differences in the methodologies and specifications make it difficult to compare results across time. There are some studies, which have examined trends in returns to schooling using a consistent dataset, but these mainly focus on the 1980s and 1990s (see eg Appleton et al., 2005; Qiu & Hudson, 2010; Zhang et al., 2005). Updating these studies for the period 2000-2010 is important for at least two reasons. One reason is that the Chinese labor market has experienced much change over this period. Since 2000, there has been considerable growth in the non-state sector, rural-urban migration has increased and there has been an expansion in the higher education sector. Each of these factors contributes to the labor market dynamics and can be expected to affect returns to schooling in urban China compared with earlier time periods. A second reason is that the jury is still out on how returns to schooling vary during economic transition. Previous studies for China suggest that returns to schooling increased markedly through the mid-1990s (see eg. Zhang et al., 2005), but studies for transition economies in Central Eastern Europe have found the rising trend in returns to schooling to be quite weak (see eg. Flabbi et al., 2008). While rising education levels and labor market reforms suggests returns to schooling should increase, other factors may offset the expected increases in returns to schooling. Experience and skills acquired under central planning may be less useful in the market reform period (Kang & Peng, 2012). This is reflected in the large-scale redundancies from China‟s state sector, which occurred in the aftermath of the Fifteenth Party Congress in 1997. In 1997-1998, it is estimated that around 12 million workers in the state-owned sector lost their jobs (Meng, 2000). Dong and Xu (2008) show that downsizing over this period had adverse implications for labor market adjustment and workers‟ earnings into the 2000s. Given the uncertainties that exist about returns to schooling during economic transition, it is important to further examine the issue using recent data for the world‟s largest transition economy. Our second contribution is that we employ a twofold strategy to address the endogeneity of years of schooling. We use spouse‟s education as a conventional instrumental variable (IV) as well as a novel identification strategy, proposed by Lewbel (2012), which utilizes a heteroscedastic covariance restriction to construct an internal IV. For a long time, ordinary least squares (OLS) were used to estimate returns to schooling in China, which reflected the lack of suitable IVs in commonly used datasets. Thus, as recently as 2005, Li et al (2005) stated: “Despite the rapid accumulation of evidence on the returns to education in China, no study has yet established causality”. More recently, a series of studies have used various conventional IVs to identify the causal effect of education on earnings in urban China (see eg. Heckman & Li, 2004; Fleisher et al. 2004; Li & Luo, 2004; Chen & Hamori, 2009; Fang et al. 2012; Kang & Peng, 2012; Mishra & Smyth, 2013a). Common IVs have been parents’ education, family background characteristics and spouse’s education. However, each of these IVs has been criticized for not satisfying the exclusion restriction. Given these concerns, using the Lewbel (2012) method, which does not rely on satisfying the exclusion restriction in addition to a conventional IV, provides the results with added robustness. If the results using these identification strategies are similar, this should increase confidence in the findings. The third contribution is that we explicitly consider returns to schooling for both migrants and non-migrant samples. 2 An advantage of CULS is that over each of its three waves it contains data on migrants and non-migrants living in urban areas. There are very few datasets with representative data on education and wages of rural-urban migrants in China (Liu & Zhang, 2012). Datasets employed in previous studies that have examined trends in returns to schooling over time (eg. the urban household 2 We use the terms „migrant‟ and „non-migrant‟ as opposed to „migrant‟ and „urban‟ because some migrants in the survey have acquired an urban household registration and some non-migrants, living on the urban fringes, have a rural household registration. surveys employed by Zhang et al., 2005) do not contain data on rural-urban migrants. 3 Yet, excluding rural-urban migrants excludes an important segment of the urban labor market and this is increasingly the case since 2000. In 2011, there were approximately 160 million rural-urban migrants in the urban areas of China, constituting 44 per cent of the urban labor force 4 . In some developed coastal areas, such as Fujian and Shanghai, this figure is likely to be in excess of 50 per cent (Gagnon et al., 2011). Returns to Education in Urban China Returns to schooling in the Chinese urban labor market in the 1980s and 1990s were extremely low. For example, Byron and Manaloto (1990), Johnson and Chow, (1997) and Liu (1998) all reported a rate of return in the range 3-4 per cent. Li (2003) controlled for heterogeneity in working hours and found a higher rate of return of 5.4 per cent. Other studies have found that returns to schooling in the mid-1990s in China were in the range 5-6 per cent (see eg. Bishop & Chiou, 2004). These estimates compared with the world average of 10.1 per cent (Psacharopoulos, 1994). Returns to schooling have increased since the mid-1990s. Ge and Yang (2011) found that the rate of return to one additional year of schooling in urban China, estimated using OLS, increased from 3.6 per cent in 1988 to 11.4 per cent in 2007. The figure from Ge and Yang (2011) for 2007 is consistent with OLS estimates for urban China 3 The urban household surveys have contained data on rural-urban migrants since 2002, but the data is not representative of this group – see the discussion in Meng et al. (2012). 4 National Bureau of Statistics of China (NBSC), The national economic and social development statistics bulletin of China in 2011. February 22, 2012. http://www.stats.gov.cn/tjgb/ndtjgb/qgndtjgb/t20120222_402786440.htm. (accessed on October 5, 2012). for 2005 reported in Qian and Smyth (2008) of 12-13 per cent. Other studies report slightly lower estimates using OLS. Qiu and Hudson (2010) found that returns to schooling in urban China at the beginning of the 2000s was 7 per cent. Mishra and Smyth (2013b) report that returns to schooling in Shanghai in 2007 were 6.9-7.4 per cent. Similarly, Chen and Hamori (2009) and Ren and Miller (2012) found returns to education in urban China in 2004 to 2006 were in the range 7-8 per cent. The reason for the low rates of return to education in the pre-reform period and initial stages of the post-reform period is that China‟s long-term allocation of labor resulted in relatively equal distribution of income. From the late 1950s to the late 1970s, under the planned allocation of labor, the allocation of workers to employers was totally done through the labor bureaus and wages were determined based on seniority. The state-owned sector dominated output and there was virtually no labor market. There were various wage reforms through the 1980s and 1990s designed to lift productivity. In 1986, the State Council introduced a system of labor contracts, which was the catalyst for the emergence of more flexible labor markets. In 1994 the introduction of a new Labor Law was a catalyst for a much tighter connection between accomplishment and reward to emerge. This occurred through more profitable firms paying their employees higher salaries in the form of bonuses. The expansion in the higher education sector, coupled with the increased cost of education, has focused attention on the size of the college premium (Li et al., 2012, Wang, 2012b; Zhong, 2011). Recent studies find a sizeable college premium, reflecting the traditional shortage of college graduates in urban China. The college premium has been found to be correlated with the quality of the college (Zhong, 2011). The findings suggest that graduates from elite colleges earn a premium over other college graduates, even controlling for student ability, major, college location, individual characteristics and family background characteristics (Li et al., 2012). Since the mid-2000s, some studies have used an IV approach (Chen & Hamori, 2009; Fang et al., 2012; Fleisher et al., 2004; Heckman & Li, 2004; Kang & Peng, 2012; Li & Luo, 2004; Mishra & Smyth, 2012, 2013a; Wang, 2012b; Zhang et al., 2007). Most of these studies use conventional IVs for own education. Mishra and Smyth (2012) and Wang (2012b) exploit heteroskedasticity in the data as an identification strategy. Mishra and Smyth (2013a) use a combination of conventional and internal IVs. Fang et al., (2012) uses the introduction of the Compulsory Education Law in 1986 as a natural experiment to address endogeneity of education. Giles et al. (2008) and Meng and Gregory (2007) use the differential impact of the Cultural Revolution across time and geographical location as an IV for schooling. Studies using an IV approach have generally found a higher rate of return than studies that have used OLS. In a meta-analysis of estimates of returns to schooling in China, Liu and Zhang (2012) found that estimates using IVs were 3.2-3.6 per cent higher than OLS estimates. There is considerable evidence of a growing gender wage gap in urban China for both the urban population as a whole and for rural-urban migrants. Moreover, existing studies suggest that a significant portion of the wage gap cannot be explained by differences in human capital and other observed characteristics (see eg. Gustafsson & Li, 2000; Zhang et al., 2008; Magnani & Zhu, 2012). Traditionally most studies have found that returns to schooling are higher for females than males. Deolalikar (1993) argued that males have a comparative advantage in physical strength so that schooling becomes relatively more important to females, whose comparative advantage is in skill-intensive jobs. Li (2003) suggested that the higher returns to schooling for females reflect the relative dearth of highly educated females in urban China. Several recent studies have found that returns for schooling are higher for males than females since 2000 (Chen & Hamori, 2009; Qian & Smyth, 2008; Kang & Peng, 2012; Ren & Miller, 2012). Some studies have found that OLS estimates are biased down for males and biased upwards for females (Chen & Hamori, 2009; Kang & Peng, 2012). Data We use household data from the three waves of the CULS (2001, 2005 and 2010). CULS was designed by an international research team from China, the United States, the United Kingdom and Australia. It was administered by the Institute of Population and Labor Economics at the Chinese Academy of Social Sciences, working in collaboration with local National Bureau of Statistics Survey teams. In each wave, CULS was administered to samples of migrant and non-migrant households The 2001 survey was administered in five provincial capital cities (Shanghai, Wuhan, Shenyang, Fuzhou and Xi‟an). For the non-migrant household sample, proportional population sampling was used to sample an average of 10 households in each of 70 neighbourhood clusters, where the clusters were selected using the 2000 Chinese Population census. For the migrant sample, 60 communities were selected using the 2000 Chinese Population census. On average 10 non-migrant households were interviewed in each neighbourhood cluster and 10 migrants were interviewed in community in each city. Thus, in total, in each city, all individuals aged 16 and above in 700 non-migrant households and 600 individual migrants were surveyed. The 2005 survey was administered in the same five provincial capital cities, but, in addition, seven municipal cities (Wuxi, Yichang, Benxi, Zhuhai, Shenzhen, Baoji and Daqing) were added. Using the same sampling approach as the 2001 survey, 500 non-migrant households and 500 migrant households were investigated in each of the five provincial cities, and 400 migrant households were investigated in each of the seven municipal cities. In each household, all family members who were aged 16 or above and who were no longer in school were individually interviewed. Following the same sampling method as the 2001 and 2005 surveys, the 2010 survey was carried out in the same five provincial capital cities covered in the earlier surveys, plus Guangzhou. The seven municipal cities included in the 2005 survey were not included in the 2010 survey. In the 2010 survey, the survey was administered to 700 non-migrant households and 600 migrant households in each of the six cities. To ensure the results are comparable across time, this study only uses data from the five provincial capital cities surveyed in all three years; that is, Shanghai, Wuhan, Shenyang, Fuzhou and Xi‟an. Given our focus is on returns to schooling we restrict the sample to urban non-agricultural workers aged 16-64 years who are in fulltime employment, „on-job‟ (zaigang) and who have complete information on education and earnings. A common approach in the literature is to exclude self-employed from the sample. In the main set of results below we include the self-employed, given the rise in the prevalence of self-employment in urban China throughout the 2000s, reflecting the influx of rural-urban migrants into Chinese cities. A high proportion of rural-urban migrants are self-employed in Chinese urban areas, typically averaging 50-60 per cent (Wang et al., 2010). In each wave of CULS, the self-employed constitute in excess of one quarter of the sample, with the proportion of self-employed migrants much higher than non-migrants. Arabsheibani and Mussurov (2007) adopt a similar approach in their study of the returns to schooling in Kazakhstan, where the level of self-employment is also high. Similarly, Mishra and Smyth (2013a) in a study of returns to schooling for a sample of ethnic Koreans from China‟s northeast, follow this strategy given that a high proportion of this minority group are self-employed in China‟s urban labor market. However, in robustness checks we exclude the self-employed and the returns to schooling are similar. The numbers of respondents for which we had valid data were 6197 in 2001, 6490 in 2005 and 8956 in 2010. Table 1 provides descriptive statistics for the full sample and for migrant and non-migrant respondents in each of the three years studied. In 2001, approximately 50 per cent of the sample worked in private enterprises or individual businesses. This figure increased to approximately 70 per cent in 2010. The proportion of migrants working in private enterprises and individual businesses is considerably higher than non-migrants in each of the three years. In 2001 28 per cent of the sample worked in collective or state-owned enterprises, but this figure fell to 20 per cent in 2010. Reflecting China‟s demographic trends, the average age of the full sample increased from 35 years in 2001 to 37.5 years in 2010. The average age of non-migrants was generally five to six years older than migrants in each of the three years. Average years of schooling for the full sample increased from 10.39 years in 2001 to 11.46 years in 2010. In each of the three years, the average years of schooling is higher for non-migrants than migrants. Table 2 shows the relationship between educational attainment and real wage per hour for the full sample and for the migrant and non-migrant subsamples. The real wage rate increased by 75 per cent for the full sample, 67 per cent for non-migrants and 100 per cent for migrants between 2001 and 2010. Wages for non-migrant respondents exceeds that for migrant respondents in each of the three years. For the full sample as well as each subsample there is a positive relationship between educational attainment and wages in each year. -------------------Insert Tables 1 & 2 --------------------Empirical Specification and Methodology We employ a Mincer earnings function in which the log of hourly earnings (measured in RMB) is regressed on years of schooling, post-school experience, post-school experience squared and a series of control variables. The specific control variables that we employ are gender, marital status, household registration (hukou) and ownership, sector and city dummy variables. A problem with the OLS estimates of the earnings function is that presence of measurement error and omission of an individual‟s ability may bias estimates of returns to schooling. Thus, in addition to OLS, we also present two-stage least squares (TSLS) estimates in which we instrument for education. The problem with IV estimation is finding a valid IV. CULS has data on three potential candidates for IVs that have been employed in previous studies to instrument for education; namely, spouse‟s education, father‟s education and mother‟s education. Of these three potential IVs, we used spouse‟s education to instrument for education. Mother‟s education and father‟s education were not significant in the first stage regression and thus were not valid IVs in our case. Spouse‟s education has been used as an instrument for education in several previous studies (see eg. Arabsheibani & Mussurov, 2007; Di Pietro & Pedace, 2008; Lall & Sakellariou, 2010; Trostel et al., 2002), including studies for urban China (Chen & Hamori, 2009; Mishra & Smyth, 2013a). For spouse‟s education to be a valid IV, it must be correlated with schooling, but not with the residual in the earnings function. The rationale for using spouse‟s education as an IV is predicated on the notion of assortative matching, which suggests married couples share common experiences and interests and that many of their characteristics, including years of schooling, will be positively correlated. A number of studies document empirical support for the existence of assortative mating in marriage (see eg. Pencavel, 1998; Weiss, 1999). However, one may not be convinced that spouse‟s education satisfies the exclusion restriction. The problem is that intuitively it is conceivable that the education of one‟s spouse will have direct effects on one‟s wage. 5 Kang and Peng (2012) argue that spouse‟s education is correlated with family background and therefore not really exogenous. Family background may have a direct influence on the income of an individual. For example, family background might assist one to find a better paying job through connections or nepotism or serve as a proxy for children‟s productivity, which is not captured by other observables such as education. Chen and Feng (2011) provide evidence consistent with both ideas in urban China. Given these concerns about the validity of spouse‟s education as an IV, we used a different identification strategy as a robust check that does not rely on the assumption that spouse education is uncorrelated with the error term in the wage equation. Specifically, we employ the methodology proposed by Lewbel (2012), which is useful for applications where other sources of identification, such as instrumental variables, are either not available or are potentially weak. A pre-condition for identification using Lewbel‟s (2012) method is the existence of heteroskedasticity in the data. The estimation problem in the current study can be summarised as: