Hukou -based labour market discrimination and ownership structure in urban China

Abstract

This paper utilises a 2008 survey on the wave of rural–urban migration in China to provide new empirical evidence on labour market discrimination against rural-to-urban migrants with rural hukou in two ownership sectors within urban China: state-owned enterprises and private firms. By employing a double-selectivity approach, this study shows that the high-wage state-owned enterprise sector is more discriminatory than the private sector against rural hukou holders in urban China. Furthermore, the results from quantile regressions and the wage decomposition inform us that the hukou-based wage discrimination is mainly against middle- and high-wage workers, especially in state-owned enterprises. Moreover, in terms of the magnitude of discrimination, female workers suffer more discrimination than male workers in state-owned enterprises.

Keywords

China labour market discrimination ownership structure rural-to-urban migration

Introduction

Urban China is much wealthier than rural China. According to the most recent data published by the National Bureau of Statistics (NBS), the average real annual income per capita is 24,564 yuan in urban areas and only 7916 yuan in rural areas.¹ In response to the higher real earnings workers can receive in urban China compared to rural China, a great deal of rural-to-urban migration has taken place and continues to take place. The most recent report from the NBS estimates that 262.61 million workers migrated from rural to urban areas in China in 2012.²

However, Chinese factories in a few coastal cities have been facing a labour shortage since 2003 because they cannot hire as many workers as they need. The labour shortage has continued to spread to other cities since 2003 (Chan, 2010). In contrast, we still observe hundreds of millions of people staying in rural areas and earning very low wages. Some scholars call the low-wage workers in the rural areas the rural labour surplus (Fields and Song, 2013; Knight et al., 2011). Why do these remaining rural workers choose to work in the agriculture sector and earn very little rather than migrating to cities for better-paying jobs?

Previous studies have provided several reasons for the co-existence of the urban labour shortage and rural labour surplus, such as credit market imperfection with costly migration and the high cost of living faced by rural-to-urban migrants in cities due to the household registration system (hukou) (Chau et al., 2014; Knight et al., 2011). There is another possible reason for this phenomenon that previous studies have mostly ignored: rural-to-urban migrants face labour market discrimination in cities due to their hukou status, especially in the urban high-wage sector at companies such as state-owned enterprises (SOEs).

A hukou is a record in the system of household registration required by law in China. The current hukou system requires that each Chinese citizen be assigned either rural or urban hukou in a given location. People inherit at birth the hukou status from their parents. Furthermore, one’s hukou status would remain unchanged no matter where the individual physically moved, unless he or she goes through a formal procedure of hukou conversion (Fields and Song, 2013). However, the present hukou conversion policy in big cities – the destination of the majority of migrants – is almost totally geared towards the super-rich and the highly educated and is beyond the reach for the majority of rural migrants (Song, 2014).

This paper attempts to provide new empirical evidence on labour market discrimination against rural hukou holders in Chinese cities, so as to offer an explanation for the simultaneous urban labour shortage and rural labour surplus. Specifically, this paper examines the extent of labour market discrimination against rural hukou holders in two ownership sectors: SOEs and private firms.

To my knowledge, this paper is the first one of its kind to distinguish the hukou-based discrimination in different ownership sectors, and there is an important reason for this differentiation. Based on existing literature, the Chinese government still has much influence over the wages in SOEs, which enables SOEs to provide higher wages and benefits as well as better job security compared to private firms, while the wages in private firms are largely determined by market forces (Chen et al., 2005; Dong and Xu, 2009; Song and Li, 2010). Thus, if the discrimination against rural hukou holders mainly occurs in the private sector where the market competition is severe, then this discrimination may be short-lived and will not last long (Becker, 1971; Dulleck et al., 2012). Moreover, since the private sector pays relatively lower wages, the discrimination in this sector would not discourage many rural workers from migrating to cities. On the contrary, if the discrimination is more serious in the high-wage SOE sector, it would have existed for a long time and become a more pressing issue since the SOE sector is commonly considered part of the government, and political power would help the SOE sector afford the discrimination. Furthermore, the discrimination in the SOE sector would lower rural workers’ expected payoffs associated with their migration to cities and thus limit the magnitude of rural–urban migration. Therefore, studying the discrimination in two ownership sectors separately can help us to analyse whether or not the discrimination would last long and impede the urbanisation process, and what policies should be adopted to further promote urbanisation.

In addition to testing discrimination in different ownership sectors, another important contribution of this paper is to investigate the heterogeneity in the extent of discrimination for different income groups along the wage distribution by using quantile regressions and wage decomposition techniques. This paper is the first attempt to explore the differences in the extent of wage discrimination against rural hukou for different quantiles of wage distribution.

Finally, a methodological contribution of this study is to adopt a double-selectivity approach to account for selectivity both in sector allocation and hukou status determination. This paper is the first one to apply the double-selection framework to study labour market discrimination in China.

The first selection issue comes from the possibility that SOEs selectively hire people with certain characteristics. Another selection problem arises because a person’s hukou status is not randomly assigned. Although people inherit at birth the hukou status from their parents, they may be able to change it upon meeting the criteria specified by local governments of their migration destinations. Small cities usually admit local urban hukou to people who have salaried jobs or own houses in their administrative areas, or possess some types of occupational skills that fit the requirements stipulated by local governments (Chan and Buckingham, 2008). Big cities set more strict criteria to grant local hukou. For instance, in Guangzhou, along with several other big cities, the government converts various criteria, such as schooling, working skills, contributions to social insurance, volunteering and blood donation into credit points, which accumulate to help people to obtain local urban hukou (Cai, 2011).

The structure of this paper is as follows. The next section reviews the previous literature and elaborates on the contributions of this paper; following that I describe the methodology of studying wage discrimination in two ownership sectors; the penultimate section presents the data description and reports the econometric results; and the final section develops a discussion of my results and then concludes.

Literature on hukou-based wage discrimination

Arrow (1973) defines wage discrimination as the valuation in the market place of personal characteristics of the worker that are unrelated to worker productivity. Most of the previous studies have used the Blinder-Oaxaca decomposition technique to conclude that wage discrimination against people with rural hukou exists in urban China (Meng, 2012; Song, 2014). The basic idea of the Blinder-Oaxaca decomposition is that differences in average wages between two groups can be decomposed to differences in characteristics (sometimes called endowment) and differences in returns to these characteristics (also called coefficients). The latter is conventionally used as an estimate of discrimination (Oaxaca, 2007).

For example, Lee (2012) used the dataset from China’s Urban Labor Survey in 2005 and showed that the average hourly wage for local urban hukou workers was 7.75 yuan and only 4.4 yuan for rural migrants. Around 28% of this difference cannot be explained by observable characteristics. Gravemeyer et al. (2011) compared the wage differential between workers with urban and rural hukou in Shenzhen city, and found that 52.9% of the average wage difference between the two groups cannot be explained by observed characteristics related to productivity. Deng (2007) used the China Household Income Project (CHIP) data in 2002 and found that 60% of the income differential between rural migrants and urban residents originates from unexplained factors. Wang (2005) conducted econometric analysis using part of the census data in 2000 and claimed that 57% of the wage differential between rural migrants and urban residents can be explained by human capital variables after controlling occupation and industry, but the remaining 43% of the wage differential remained unexplained.

To my knowledge, only one previous study explored the heterogeneous effects of hukou status on wages in different economic sectors. Gagnon et al. (2011) utilised part of the census data in 2005 and analysed the wage differential between urban and rural hukou holders in three sectors: the formal sector, the self-employment sector, and the ‘no contract’ sector. The ‘no contract’ sector was defined as jobs without a written labour contract and the formal sector was defined as salary jobs with a written labour contract. Using this division, they discovered that the wage differential between urban and rural hukou holders was much larger in the formal sector than in the other two sectors.

Although the study above compared the wage discrimination in different sectors, it did not consider the division between SOEs and private firms. A considerable share of private firms has labour contracts with their employees and thus belongs to the formal sector as defined in Gagnon et al. (2011). Therefore, we could conclude that no previous study has provided quantitative evidence on labour market discrimination against rural hukou holders in separate ownership sectors.

The present paper aims to fill in this gap to explore the extent of wage discrimination against rural hukou in SOEs and private firms separately by adopting a double-selectivity approach. Furthermore, I will combine quantile regressions and decomposition techniques to investigate the heterogeneity in terms of the extent of wage discrimination for different income groups.

Methodology and model specification

This section adopts the regression-based decomposition technique to examine the extent of wage discrimination against rural hukou holders in two ownership sectors. We differentiate between four groups of workers by two dimensions: ownership sector (SOE or private) and hukou status (urban or rural). Accordingly, allowance for different wage structures for each group of workers requires an estimation of four separate wage regressions.

In addition, since workers with each hukou status and in each ownership sector are not randomly drawn from the population, two selection equations that determine sector allocation and hukou status, respectively, need to be examined to consistently estimate four wage equations. In summary, we have a system of six equations in total, with two selection equations and four wage equations as follows. Let $S_{i}^{*}$ and $U_{i}^{*}$ be two latent unobserved continuous variables to determine sector allocation and hukou status, respectively, for person i. Moreover, let $W_{j}^{i}$ (i = U or R, j = S or P) denote the wages for urban and rural hukou holders in the SOE sector and private firms, respectively, with the superscript standing for the hukou status, and subscript representing the ownership sector people work in.

\begin{matrix} S_{i}^{*} = X_{1 i}' δ_{1 i} + u_{1 i}, SOE sector \\ determination \end{matrix}

(1)

U_{i}^{*} = X_{2 i}' δ_{2 i} + u_{2 i}, Urban hukou determination

(2)

\ln W_{S}^{U} = X_{S}^{U} γ_{S}^{U} + v_{S}^{U}

(3)

\ln W_{S}^{R} = X_{S}^{R} γ_{S}^{R} + v_{S}^{R}

(4)

\ln W_{P}^{U} = X_{P}^{U} γ_{P}^{U} + v_{P}^{U}

(5)

\ln W_{P}^{R} = X_{P}^{R} γ_{P}^{R} + v_{P}^{R}

(6)

The error terms in the six equations above are assumed to have the following variance-covariance matrix.

\sum = [\begin{matrix} σ_{u 1} & ρ & a_{11} & a_{12} & a_{13} & a_{14} \\ ρ & σ_{u 2} & a_{21} & a_{22} & a_{23} & a_{24} \\ a_{11} & a_{21} & σ_{S}^{U} & 0 & 0 & 0 \\ a_{12} & a_{22} & 0 & σ_{S}^{R} & 0 & 0 \\ a_{13} & a_{23} & 0 & 0 & σ_{P}^{U} & 0 \\ a_{14} & a_{24} & 0 & 0 & 0 & σ_{P}^{R} \end{matrix}]

(7)

No selectivity correction

In the absence of selection issues, we can employ OLS estimation to consistently estimate the four wage regressions from equations 3–6. This simple specification assumes that all of the off-diagonal terms in equation 7 are zero. Given the OLS results, the average log wage differential is then decomposed into the endowment effect, which can be explained by productivity-related characteristics, and the unexplained coefficient effect, which is the estimate for wage discrimination. Specifically, the decomposition of the wage differential between urban and rural hukou holders in a given sector j (SOE or private) can be written as follows:

\begin{matrix} \bar{\ln W_{j}^{U}} - \bar{\ln W_{j}^{R}} = \bar{X_{j}^{U}}' {\hat{γ}}_{j}^{U} - \bar{X_{j}^{R}}' {\hat{γ}}_{j}^{R} \\ = (\bar{X_{j}^{U}} - \bar{X_{j}^{R}})' γ_{j}^{*} + [\bar{X_{j}^{U}}' ({\hat{γ}}_{j}^{U} - γ_{j}^{*}) \\ + \bar{X_{j}^{R}}' (γ_{j}^{*} - {\hat{γ}}_{j}^{R})] \end{matrix}

(8)

In the equations above, $\bar{\ln W_{j}^{U}}$ and $\bar{\ln W_{j}^{R}}$ are the predicted mean (log) wages for urban and rural hukou holders in sector j (S or P), $\bar{X}'$ is the mean vector of wage determining variables, $\hat{γ}$ is a vector of estimated coefficients (wage structure), and $γ *$ is the non-discriminatory wage structure. The first term above $(\bar{X^{U}} - \bar{X^{R}})' γ *$ is called the endowment effect. The second term in the square bracket $[\bar{X^{U}}' ({\hat{γ}}^{U} - γ *) + \bar{X^{R}}' (γ * - {\hat{γ}}^{R})]$ is typically interpreted as part of the wage gap that is associated with differences in returns to the characteristics between the two groups, which serves as the estimate for discrimination (Oaxaca, 2007).

A practical concern associated with the Blinder-Oaxaca approach is called the index number problem, meaning that the decomposition is not unique, depending on the choice of the non-discriminatory wage structure, that is, the choice of $γ *$ . The original version of the Blinder-Oaxaca approach assumes that $γ *$ takes extreme values equal to the wage structure of one reference group in the study. For example, if $γ *$ = ${\hat{γ}}^{U}$ , then we assume that the wage structure for urban hukou holders would be adopted in the absence of discrimination (Liu et al., 2000). This is a very common assumption in that the wage structure of the dominant group is usually considered as the hypothetical wage structure in the absence of discrimination. However, if one believes that the current wage structure for rural hukou holders represents what the competitive wage structure would be in the absence of discrimination, then $γ *$ = ${\hat{γ}}^{R}$ . As mentioned earlier, most of the studies on the Chinese labour market use this original version of the Blinder-Oaxaca decomposition method (Gagnon et al., 2011; Gravemeyer et al., 2011; Liu et al., 2000), and employ the two extreme wage structures to approximate the wage structure in the absence of discrimination.

Cotton (1988) argued that derived weights using averages of coefficients for two groups were more accurate than coefficients for one reference group. According to this approach, the non-discriminatory wage structure in sector j is $γ_{j}^{*}$ = $f_{j}^{U}$ ${\hat{γ}}_{j}^{U}$ + $f_{j}^{R}$ ${\hat{γ}}_{j}^{R}$ , where $f_{j}^{U}$ and $f_{j}^{R}$ are the proportions of urban and rural hukou holders in sector j, respectively. Thus, Cotton’s weights are sometimes called population-weighted non-discriminatory wage structures (Liu et al., 2000).

An alternative way to decompose the wage differential is to obtain the competitive wage structure ( $γ *$ ) from a regression with the pooled sample (Oaxaca and Ransom, 1994). As they argue, if by the absence of discrimination one means the complete cessation of existing discrimination activity, then it might be reasonable to suppose that the resulting competitive wage structure would be the parameter vector estimated with the pooled sample.

However, there is no consensus in discrimination literature about which weighting scheme is unambiguously better than another (Powers et al., 2011). Thus, this paper will report the decomposition results using four alternative non-discriminatory wage structures: the wage structure of urban hukou holders, the wage structure of rural hukou holders, the population-weighted wage structure, and the wage structure from the pooled sample. As can be seen from the results shown in the next section, the main results of this paper are robust to alternative weighting schemes. Hence, the major conclusion will be based on the most commonly used weighting scheme in the discrimination literature, that is, the wage structure of high income group (urban hukou holders).

Double-selectivity approach

This section takes into account the two selection issues and assumes that the two selection processes are correlated ( $ρ$ $\neq 0$ ). In this case, I adopt the double-selectivity approach to estimate the system of equations 1–6 with the stochastic specification in equation 7. Tunali (1986) shows that if the error terms have a joint normal distribution,³ then a two-step estimation procedure can be applied to correct for the double-selectivity issue, which is very similar to Heckman’s two step method (Heckman, 1979). Specifically, the two selection equations are estimated using a bivariate probit model, and augmented selectivity correction terms are derived as additional regressors in four wage equations. Given the estimates from augmented wage regressions, I can decompose the selectivity-corrected average wage differentials between urban and rural hukou holders in SOEs and private firms, respectively, as proposed by Reimers (1983). The modified wage differential is again decomposed to endowment effect and coefficient effect, where the latter is the estimate for wage discrimination. A specific introduction of Tunali’s model is provided in Appendix 1.

Although the non-linearity of the selection equations can in principle identify the system of equations specified above, it is better to include some variables in each selection equation that are assumed to be relevant only for that selection equation and excluded from the other selection equation and wage equations (Rabe, 2011). The excluded instrument that identifies the sector determination equation is a dummy variable indicating whether the individual is recommended by the government for a particular job. Since the SOE sector has a close relationship with the government in China, the government sometimes introduces people to SOEs, but this kind of government reference is very rare in private firms. Thus, whether or not having government reference for a job is a good indicator for sector determination. Moreover, since SOEs usually have established wage schemes for workers with different characteristics, such as education and experience, it is unlikely that the government reference would directly affect workers’ wages in the SOE sector.

In addition, the variable included in the hukou determination equation but excluded from the wage equation is a continuous variable representing the total amount of money an individual received from his or her parents in the last 12 months. As introduced previously, many cities grant urban hukou to people who make an investment in the city, or purchase some property such as a house, and so on (Song, 2014). Therefore, if people receive more money from their parents in the past year, they are more likely to meet the criteria specified by the local government and thus have a larger chance to be granted an urban hukou. This variable is excluded from the wage equation because employers seldom know how much money a worker received from his or her parents in the past year, and thus cannot pay workers differently based on this unknown information.⁴

Quantile regressions and wage decomposition

In order to obtain more insight on the hukou-based discrimination in urban China, I explore the extent of wage discrimination for different quantiles along the wage distribution by combining the quantile regression and decomposition technique. That is, I wonder whether high- or low-wage rural hukou holders suffer more discrimination in urban China.

As is known, the quantile regression model introduced by Koenker and Basset (1978) extends the notion of ordinary quantiles in the location model to a more general class of linear models in which conditional quantiles have a linear form. Several notable studies have combined the quantile regression and decomposition technique to study the public–private sector wage differentials (Melly, 2005; Mueller, 1998). I will adopt this same approach to determine the extent of wage discrimination at various points in the wage distribution against rural hukou holders.

Data description and econometric results

A 2008 wave of the rural–urban migration in China (RUMiC) survey is utilised to analyse the proposed question. The longitudinal survey on RUMiC consists of three parts: the urban household survey, the rural household survey, and the migrant household survey.⁵ The survey covers three groups of Chinese households: 5000 migrant households who worked in 15 designated cities (migrant household survey) in nine provinces or metropolitan areas; 5000 urban local incumbent households in the same cities (urban household survey); and 8000 rural households (rural household survey). Most of the migrants have rural hukou while the urban sample largely takes urban hukou. This feature gives us enough observations for both urban and rural hukou holders. For the purpose of this paper, I will only use the urban sample and migrant sample in the 2008 wave of the survey.

The nine provinces in our sample can well represent the entire country since they cover a wide area of China, both geographically and economically. Specifically, Shanghai, Guangdong, Jiangsu and Zhejing located in eastern China are relatively richer, and Sichuan and Chongqing located in western China are relatively poorer, while Anhui, Hubei and Henan located in central China are in the middle in the income distribution. The RUMiC survey of migrant workers in urban cities, to the best of our knowledge, is the only random sample of migrant workers for China. The survey records detailed individual information, such as monthly earnings, hours of work, demographic characteristics, and other job-related information. A detailed definition of each variable used in this study is provided in Appendix 2.

The sample is restricted to employees aged from 18 to 60 in urban areas. Furthermore, people working at foreign-owned companies and self-employed workers are excluded from the sample. The remaining sample consists of 5012 urban hukou holders and 3795 rural hukou holders. The total sample size is 8807. The summary statistics of the key variables in the dataset are presented in Table 1.

Table 1.

Summary statistics of key variables.

	SOE		Private
	Urban hukou	Rural hukou	Urban hukou	Rural hukou
Age	40.18 (40; 9.75)	32.08 (30; 10.79)	38.01 (38; 10.16)	30.73 (29; 10.24)
Years of schooling	12.53 (12; 3.28)	9.65 (9; 2.57)	11.88 (12; 3.23)	9.36 (9; 2.48)
Experience	21.67 (22; 11.41)	16.25 (13; 11.64)	20.12 (20; 11.79)	15.12 (13; 11.13)
Monthly earnings	2294.63 (1985; 1969.25)	1529.76 (1400; 748.40)	2018.49 (1500; 2450.10)	1500.99 (1300; 957.43)
Hourly wage	15.08 (11.25; 23.36)	7.40` (6.25; 4.28)	12.76 (8.75; 22.54)	6.72 (5.60; 4.47)
Male	0.58 (1; 0.49)	0.65 (1; 0.48)	0.55 (1; 0.50)	0.61 (1; 0.49)
Parent money	683.23 (0; 3641.76)	361.57 (0; 2150.14)	668.40 (0; 2617.13)	214.02 (0;1158.24)
Gov. reference	0.45 (0; 0.50)	0.03 (0; 0.163)	0.16 (0; 0.362)	0.003 (0; 0.056)
Firm size	1261.06 (164; 14,867.35)	61.62 (7; 454.73)	254.48 (50; 1017.34)	15.03 (5; 132.07)
Social capital	35.58 (25; 46.51)	28.72 (18; 46.86)	33.19 (20; 44.03)	31.93 (20; 44.89)
N	3198	658	1814	3137

Notes: The main entries represent the means of key variables. Figures in parentheses are median and standard deviations of the variables, respectively. A detailed definition of each variable is provided in Appendix 2.

As can be seen from Table 1, of the 5012 urban hukou holders, more than half of them (3198) work in SOEs, whereas only a small proportion of rural hukou holders work in this high wage sector. In addition, the average monthly earnings are 2294.6 (yuan) and 1529.8 in the SOE sector for urban and rural hukou holders, respectively, so the average earning differential is about 50%. On the other hand, the average differential in terms of monthly earnings is about 30% in the private firms. Similarly, the differentials in median wages between urban and rural workers are also larger in SOEs than in private firms.

Moreover, Table 1 shows that on average, urban hukou holders are older than rural hukou holders, indicating that most of the rural migrants in cities are young people aged around 30. Additionally, more than 60% of rural hukou holders are male, while the gender composition is more balanced among urban hukou holders. In terms of human capital variables, urban hukou holders receive more education than rural hukou holders, and the average years of schooling is slightly higher in SOEs than in the private sector. The dataset also contains a variable to reflect a person’s social capital measured by the number of greetings sent in the past Chinese New Year festival.⁶ As is seen in Table 1, urban hukou holders in the SOE sector possess the highest social capital, and urban hukou holders have on average more social capital than rural hukou holders in each sector. Finally, in terms of a job-related characteristic, workers with urban hukou tend to work in larger firms than those with rural hukou.

The excluded variable to identify the sector selection equation is a dummy variable indicating whether the individual is recommended by the government for a particular job. Table 1 indicates that nearly half of the urban hukou holders in the SOE sector are recommended by the government for that job, while this ratio is much smaller in the private sector. Moreover, urban hukou holders received more money from their parents in the past year than rural hukou holders in each sector, which can partly verify the use of this identification variable in the hukou determination equation.

As Giulietti et al. (2012) claim, in the RUMiC dataset, hourly wages are more prone to measurement error (being calculated by combining monthly wages and weekly hours worked), so the monthly wages are chosen as the major dependent variable. As a robustness check, the wage equations have also been estimated using hourly wages, and the results are presented in the next section.

Results without selectivity correction

In the first step, I estimate four wage equations for urban and rural hukou holders separately in each ownership sector without dealing with the selection issues. The estimation results are reported in Table 2. The dependent variables in all wage equations are the natural logarithm of monthly earnings measured in nominal Chinese currency (yuan) in 2007. The independent variables include potential labour market experience and its square, completed years of schooling, social capital, seniority, health status, weight, height, school performance, gender, marriage status, a dummy variable indicating job training, firm size, and a set of dummies including occupation, industry, and province dummies to control for differences in the cost of living in different regions.

Table 2.

Log wage regressions without selectivity correction.

	SOE		Private
	Urban hukou	Rural hukou	Urban hukou	Rural hukou
Years of schooling	0.025*** (0.005)	0.022*** (0.008)	0.028*** (0.007)	0.036*** (0.004)
Experience	−0.003 (0.005)	0.018*** (0.006)	0.009 (0.006)	0.021*** (0.003)
Exp. sq.	−0.0001 (0.0001)	−0.0005*** (0.0001)	−0.0005*** (0.0001)	−0.0005*** (0.0001)
Male	0.247*** (0.038)	0.091* (0.047)	0.343*** (0.052)	0.092*** (0.023)
Social capital	0.0016*** (0.0003)	0.0002 (0.0003)	0.001*** (0.0004)	−0.0001 (0.0002)
Firm tenure	0.013*** (0.001)	0.015*** (0.004)	0.009*** (0.002)	0.019*** (0.002)
Constant	8.425*** (0.532)	5.582*** (0.581)	6.748** (0.716)	5.704*** (0.271)
N	2679	639	1455	3040
Adjusted R-sq	0.2155	0.2536	0.2241	0.1725

Notes: *** = significant at 1%; ** = significant at 5%; * = significant at 10%.

Figures in parentheses are standard errors. The dependent variables in all wage equations are the natural logarithm of monthly earnings in 2007. All regressions also control for number of children, health status, weight, height, school performance, gender, marriage status, a dummy variable indicating job training, firm size, and a set of dummies including occupation, industry, and province dummies.

Table 2 presents the coefficients of the major human capital variables. As the table shows, the return to schooling is around 3% and is very similar across all regressions. The magnitude of the return to schooling found here is consistent with recent studies (Demurger et al., 2012; Song, 2012). In addition, it turns out that the coefficients on other human capital variables are all statistically significant for rural hukou holders and mostly larger than the coefficients for urban hukou holders in both sectors, whereas the coefficients on social capital display the opposite pattern. This may indicate that the wage structure is more market-oriented for rural hukou holders, while social networking plays a larger role for urban hukou workers in determining wages. Comparing the SOEs with private firms, we can see that the wage structures are comparable for rural hukou holders in these two sectors. For urban hukou holders, the returns to major human capital variables are higher in private firms than in SOEs, including years of schooling and work experience. However, it is noteworthy that the return to social capital and firm tenure is higher in SOEs than private firms for urban hukou holders.

Another finding from Table 2 is that for urban hukou holders, males with the same observed characteristics earn about 30% more than females, which suggests evidence of gender discrimination. For rural hukou holders, the gender wage premium is about 10%. We can conclude that the gender discrimination is more severe among urban hukou holders than among those with rural hukou.

Table 3 reports the decomposition of the overall wage differential into the explained and unexplained components in terms of log differentials. The raw log wage differential is higher in the SOE sector than in the private sector between urban and rural hukou holders. Note that the entire wage differential in the private sector between urban and rural hukou holders is due to people’s productivity-related characteristics, meaning that there is no discrimination in the private sector. In contrast, some proportion of wage differential exists in the SOE sector that cannot be explained by observed characteristics, although it is not that large.

Table 3.

Wage decomposition without selectivity correction.

		Urban-weighted	Rural-weighted	Pooled sample weighted	Population- weighted
SOE	Total diff.	0.294 (0.022)	0.294 (0.022)	0.294 (0.022)	0.294 (0.022)
	Explained	0.287(0.063)	0.289(0.101)	0.292(0.021)	0.287(0.055)
	Unexplained	0.008(0.065)	0.005(0.103)	0.003(0.013)	0.007(0.058)
Private	Total diff.	0.162(0.020)	0.162(0.020)	0.162(0.020)	0.162(0.020)
	Explained	0.289(0.031)	0.226(0.090)	0.170(0.016)	0.250(0.058)
	Unexplained	−0.126(0.035)	−0.064(0.092)	−0.008(0.009)	−0.087(0.060)

Notes: The main entries represent the decomposition of log wage differential, in which wage is measured by monthly earnings. Figures in parentheses are standard errors.

Results with double-selectivity correction

Table 4 reports the decomposition results with double-selectivity correction.⁷ As Table 4 shows, the wage discrimination is unambiguously more severe in the SOE sector than in private firms, regardless of the weight used for non-discriminatory wage structure. For example, if we take the wage structure of urban hukou holders as the non-discriminatory wage structure, the unexplained log wage differential is 0.384 in SOEs, but only 0.049 in private firms. If we translate the log differential into percentage points, the results suggest that for observationally-equivalent workers, urban hukou holders earn about 50% more than rural hukou holders do in the SOE sector, but only 5% more in the private sector.

Table 4.

Wage decomposition with double selectivity correction.

		Urban-weighted	Rural-weighted	Pooled sampleweighted	Population-weighted
SOE	Total diff. (adjusted)	1.096	1.096	1.096	1.096
		(0.393)	(0.393)	(0.393)	(0.393)
	Explained	0.712	-0.071	0.877	0.579
		(0.311)	(0.205)	(0.316)	(0.260)
	Unexplained	0.384	1.167	0.218	0.517
		(0.200)	(0.449)	(0.082)	(0.219)
Private	Total diff. (adjusted)	0.322	0.322	0.322	0.322
		(0.112)	(0.112)	(0.112)	(0.112)
	Explained	0.273	-0.038	0.263	0.077
		(0.118)	(0.252)	(0.057)	(0.165)
	Unexplained	0.049	0.361	0.060	0.246
		(0.222)	(0.301)	(0.059)	(0.239)

Notes: The main entries represent the decomposition of selectivity-adjusted log wage differential, in which wage is measured by monthly earnings. Figures in parentheses are standard errors.

Compared with the results in Table 3, the wage discrimination is aggravated by the selectivity adjustment, especially in the SOE sector. This is because rural hukou holders are positively selected into the SOE sector. In other words, mean wage offers are above mean observed wages among urban hukou holders and below mean observed wages among rural hukou holders.

Robustness check for wage discrimination

In addition to conducting robustness checks by using different non-discriminatory wage structures, this section will provide another robustness check by using hourly wages as the dependent variable. The results are presented in Table 5. Not surprisingly, we still observe the general picture that SOEs are more discriminatory than private firms in terms of wages. The unexplained wage differentials between urban and rural hukou holders are mostly larger in SOEs than in private firms. In few cases, the unexplained part is similar between the two sectors, but there is no case in which the discrimination is significantly larger in private firms than in the SOE sector. In summary, the difference in terms of the extent of wage discrimination between the SOE and private sectors is smaller if we use hourly wages in place of monthly earnings.

Table 5.

Hourly wage decomposition with selectivity correction.

		Urban-weighted	Rural-weighted	Pooled sample weighted	Population- weighted
SOE	Total diff. (adjusted)	1.013	1.013	1.013	1.013
		(0.483)	(0.483)	(0.483)	(0.483)
	Explained	0.469	−0.075	0.800	0.376
		(0.382)	(0.226)	(0.388)	(0.319)
	Unexplained	0.545	1.088	0.213	0.637
		(0.245)	(0.538)	(0.100)	(0.267)

Private	Total diff. (adjusted)	0.809	0.809	0.809	0.809
		(0.128)	(0.128)	(0.128)	(0.128)
	Explained	0.201	−0.258	0.580	-0.088
		(0.135)	(0.288)	(0.065)	(0.188)
	Unexplained	0.608	1.067	0.229	0.897
		(0.253)	(0.344)	(0.067)	(0.274)

Notes: The main entries represent the decomposition of selectivity-adjusted log hourly wage differential. Figures in parentheses are standard errors.

Further analysis based on quantile regressions

Table 6 reports the results with quantile regressions and the decomposition technique. These analyses were performed separately for males and females.⁸ Dealing with male workers first, the estimates show that the hukou-based discrimination mainly exists for middle-income workers, especially for the 50% and 75% quantiles in the wage distribution. Furthermore, the discrimination against workers around the median income is also more severe in SOEs than in private firms, which is consistent with our main results. Another interesting finding from Table 6 is that for the very high income group (0.9th quantile), the discrimination against rural hukou is more serious in private firms than SOEs, which needs further research and exploration.

Table 6.

Wage decomposition for different quantiles.

	0.1th	0.25th	0.5^th	0.75^th	0.9th
Males
SOE
Total diff.	0.104	0.202	0.327	0.473	0.578
Explained	0.172	0.206	0.238	0.389	0.569
Unexplained	−0.068	−0.004	0.088	0.084	0.009
Private
Total diff.	−0.058	0.085	0.237	0.385	0.516
Explained	0.119	0.176	0.229	0.309	0.400
Unexplained	−0.178	−0.090	0.007	0.076	0.115
Females
SOE
Total diff.	−0.037	0.150	0.331	0.467	0.589
Explained	0.068	0.116	0.193	0.270	0.292
Unexplained	−0.105	0.034	0.139	0.196	0.297
Private
Total diff.	−0.102	0.023	0.109	0.235	0.353
Explained	0.158	0.239	0.314	0.383	0.488
Unexplained	−0.261	−0.216	−0.205	−0.148	−0.135

Note: The main entries represent the urban-weighted decomposition of log wage differential for different quantiles, in which wage is measured by monthly earnings.

The lower part of Table 6 shows the results for female workers. Since the unexplained differentials are all negative in private firms for all quantiles, there is no discrimination against rural hukou in private firms. For females working in SOEs, the discrimination becomes more severe as we move along the wage distribution. Combining the results for males and females, we can conclude that the hukou-based wage discrimination is mainly against middle- and high-income workers, especially in SOEs. Moreover, in terms of the magnitude of discrimination, female workers suffer more discrimination than male workers suffer in SOEs.

Discussion and concluding remarks

To recapitulate, this paper demonstrates that migrants with rural hukou face wage discrimination in the high-wage SOE sector in urban China. The SOEs discriminate against rural hukou holders more than the private firms do partly because they are protected by the government and shielded from competitive market forces. Furthermore, the results from quantile regressions and the decomposition inform us that the hukou-based wage discrimination is mainly against middle- and high-wage workers, especially in SOEs. Moreover, in terms of the magnitude of discrimination, female workers suffer more discrimination than male workers suffer in SOEs.

Back to the question posed at the beginning: why do some rural hukou workers choose to work in the agricultural sector but not in the city to pursue higher wages? The other side of the coin is why some rural hukou workers work in cities despite discrimination. Undoubtedly, the reason is more than wage discrimination, but wage discrimination is indeed one important part of the explanation.

Different forces drive people in opposite directions, rendering an interior equilibrium in the sense that in the steady equilibrium, some people choose to stay in rural areas whereas other people choose migration. On the one hand, higher average wages in urban China encourage some rural hukou holders to migrate. However, not all rural workers migrate to cities because the hukou system plays two different roles in the Chinese labour market, which limits the magnitude of rural-to-urban migration. One role it plays is that the high wage SOE sector in urban China practises wage discrimination against rural hukou holders, which reduces their expected payoffs associated with migrating to cities, as found in this paper. Another role of the hukou system is that rural hukou holders have to bear a large cost when they live in the city, but they do not have the same access to various government programmes and public services as urban hukou holders (Chen and Yang, 2010; Song, 2014). These two reasons together make some rural hukou holders choose to stay in the rural area and accept lower wages.

In a recent paper, Fields and Song (2013) built a theoretical model incorporating the two roles of the hukou system above and theoretically proved an interior equilibrium. My paper serves as solid empirical evidence for this explanation from the perspective of labour market discrimination.

Since the Chinese government still has some influence over the pay in SOEs, an important policy implication from this study is to reduce the wage discrimination against rural hukou holders in SOEs, especially for middle- and high-wage female workers. The reduction in labour market discrimination against rural hukou holders in the high wage SOE sector would promote the urbanisation process and thus China’s economic growth in the long run.

Footnotes

Appendix 1: Tunali’s double selectivity model

Following the notations from equations (1)–(8), let ‘S’ be a dummy variable equal to 1 if $S_{i}^{*}$ $\geq 0$ and 0 otherwise. Similarly, define ‘U’ as a dummy variable equal to 1 if $U_{i}^{*} \geq$ 0 and 0 if $U_{i}^{*}$ < 0. S and U indicates whether an individual is working in an SOE and has an urban hukou, respectively. After estimating the bivariate probit model for the two selection equations, I can obtain the following selectivity-correction terms. Set C₁ = $X_{1 i}' δ_{1 i}$ , C₂ = $X_{2 i}' δ_{2 i}$ , $M_{1}$ = $\frac{C_{1} - ρ C_{2}}{\sqrt{1 - ρ^{2}}}$ , and $M_{2}$ = $\frac{C_{2} - ρ C_{1}}{\sqrt{1 - ρ^{2}}}$ . $Φ_{2}$ is the bivariate standard normal distribution function, $ϕ$ is the univariate standard normal density function, and $Φ$ is the cumulative standard normal distribution. The selectivity-correction terms generated by the bivariate estimation are ${\hat{λ}}_{S}^{U}$ = $\frac{ϕ (C_{1}) Φ (M_{2})}{Φ_{2} (C_{1}, C_{2}; ρ)}$ ; ${\hat{λ}}_{U}^{S}$ = $\frac{ϕ (C_{2}) Φ (M_{1})}{Φ_{2} (C_{1}, C_{2}; ρ)}$ ; ${\hat{λ}}_{S}^{R}$ = $\frac{ϕ (C_{1}) Φ (- M_{2})}{Φ_{2} (C_{1}, - C_{2}; - ρ)}$ ; ${\hat{λ}}_{R}^{S}$ = $- \frac{ϕ (C_{2}) Φ (M_{1})}{Φ_{2} (C_{1}, - C_{2}; - ρ)}$ ; ${\hat{λ}}_{P}^{U}$ = $- \frac{ϕ (C_{1}) Φ (- M_{2})}{Φ_{2} (- C_{1}, C_{2}; - ρ)}$ ; ${\hat{λ}}_{U}^{P}$ = $- \frac{ϕ (C_{2}) Φ (- M_{1})}{Φ_{2} (- C_{1}, C_{2}; - ρ)}$ ; ${\hat{λ}}_{P}^{R}$ = $- \frac{ϕ (C_{1}) Φ (- M_{2})}{Φ_{2} (- C_{1}, - C_{2}; ρ)}$ ; ${\hat{λ}}_{R}^{P}$ = $- \frac{ϕ (C_{2}) Φ (- M_{1})}{Φ_{2} (- C_{1}, - C_{2}; ρ)}$ .

In the second step, I estimate four augmented wage equations for urban and rural hukou holders in two ownership sectors as specified below.

(A1)

\ln W_{S}^{U} = X_{S}^{U} γ_{S}^{U} + α_{11} {\hat{λ}}_{S}^{U} + α_{12} {\hat{λ}}_{U}^{S} + w_{S}^{U}

(A2)

\ln W_{S}^{R} = X_{S}^{R} γ_{S}^{R} + α_{21} {\hat{λ}}_{S}^{R} + α_{22} {\hat{λ}}_{R}^{S} + w_{S}^{R}

(A3)

\ln W_{P}^{U} = X_{P}^{U} γ_{P}^{U} + α_{31} {\hat{λ}}_{P}^{U} + α_{32} {\hat{λ}}_{U}^{P} + w_{P}^{U}

(A4)

\ln W_{P}^{R} = X_{P}^{R} γ_{P}^{R} + α_{41} {\hat{λ}}_{P}^{R} + α_{42} {\hat{λ}}_{R}^{P} + w_{P}^{R}

Given the estimates obtained from the augmented wage regressions (A1–A4), I can decompose the average wage differentials between urban and rural hukou holders in SOEs and private firms, respectively. Equations (A5) and (A6) provide the formulae for the wage decomposition in the SOE sector and private sector, respectively ( $γ *$ = ${\hat{γ}}^{U}$ ).

(A5)

\begin{matrix} (\bar{\ln W_{S}^{U}} - α_{11} {\bar{λ}}_{S} - α_{12} {\bar{λ}}_{U}) - \bar{(\ln W_{S}^{R} -} α_{21} {\bar{λ}}_{S} \\ - α_{22} {\bar{λ}}_{R}) = (\bar{X_{S}^{U}} - \bar{X_{S}^{R}})' {\hat{γ}}_{S}^{U} + \bar{X_{S}^{R}}' ({\hat{γ}}_{S}^{U} - {\hat{γ}}_{S}^{R}) \end{matrix}

(A6)

\begin{matrix} (\bar{\ln W_{P}^{U}} - α_{31} {\bar{λ}}_{P} - α_{32} {\bar{λ}}_{U}) - \bar{(\ln W_{P}^{R} -} α_{41} {\bar{λ}}_{P} \\ - α_{42} {\bar{λ}}_{R}) = (\bar{X_{P}^{U}} - \bar{X_{P}^{R}})' {\hat{γ}}_{P}^{U} + \bar{X_{P}^{R}}' ({\hat{γ}}_{P}^{U} - {\hat{γ}}_{P}^{R}) \end{matrix}

Appendix 2: Variable definitions

This section explains the measure for each variable used in this paper. The independent variables enter regressions in original forms, not in log forms.

Acknowledgements

The author would like to thank Prof. Gary Fields and Prof. Francine Blau for their valuable suggestions on the earlier draft of this paper.

Funding

Project 71403281 supported by National Natural Science Foundation of China. The Project Sponsored by the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry.

Notes

References

Arrow

(1973) The theory of discrimination. In: Ashenfelter

Rees

(eds) Discrimination in Labor Markets. Princeton: Princeton University Press, pp. 3–33.

Becker

(1971) The Economics of Discrimination (2nd ed.). Chicago: University of Chicago.

Cai

(2011) Hukou system reform and unification of rural–urban social welfare. China & World Economy 19(3): 33–48.

Chan

(2010) The household registration system and migrant labor in China: Notes on a debate. Population and Development Review 36(2): 357–364.

Chan

Buckingham

(2008) Is China abolishing the hukou system? China Quarterly 195: 582–606.

Chau

Kanbur

Qin

(2014) Do public work schemes deter or encourage outmigration? Empirical evidence from China. IZA Journal of Labor & Development 3(4): 1–20.

Chen

Yang

(2010) Access to compulsory education by rural migrants’ children in urban China: A case study from nine cities. Educational Research 1(10): 512–519.

Chen

Démurger

Fournier

(2005) Earnings differentials and ownership structure in Chinese enterprises. Economic Development and Cultural Change 53(4): 933–958.

Cotton

(1988) On the decomposition of wage differentials. Review of Economics and Statistics 70(2): 236–243.

10.

Demurger

Yang

(2012) Earnings differentials between the public and private sectors in China: Exploring changes for urban local residents in the 2000s. China Economic Review 23: 138–153.

11.

Deng

(2007) Earnings differential between urban residents and rural migrants: Evidence from Oaxaca-Blinder and quantile regression decompositions (in Chinese). Chinese Journal of Population Science 2: 8–16.

12.

Dong

(2009) Labor restructuring in China: Toward a functioning labor market. Journal of Comparative Economics 37(2): 287–305.

13.

Doornik

Hansen

(2008) An omnibus test for univariate and multivariate normality. Oxford Bulletin of Economics and Statistics 70: 927–939.

14.

Dulleck

Fooken

(2012) Institutional and individual labor market discrimination based on hukou status: An artefactual field experiment. In: Proceedings of 27th Annual Congress of the European Economic Association. Malaga, Spain.

15.

Fields

Song

(2013) A theoretical model of the Chinese labor market. IZA discussion paper series, no. 7278. IZA, Bonn, Germany, March.

16.

Gagnon

Xenogiani

Xing

(2011) Are all migrants really worse off in urban labour markets: New empirical evidence from China. IZA discussion paper series, no. 6268. IZA, Bonn, Germany, December.

17.

Giulietti

Ning

Zimmermann

(2012) Self-employment of rural-to-urban migrants in China. International Journal of Manpower 33(1): 96–117.

18.

Gravemeyer

Gries

Xue

(2011) Income determination and income discrimination in Shenzhen. Urban Studies 48(7): 1457–1475.

19.

Heckman

(1979) Sample selection bias as a specification error. Econometrica 47(1): 153–161.

20.

Knight

Deng

(2011) The puzzle of migrant labor shortage and rural labor surplus in China. China Economic Review 22(4): 585–600.

21.

Koenker

Basset

(1978) Regression quantiles. Econometrica 46: 33–50.

22.

Lee

(2012) Decomposing wage differentials between migrant workers and urban workers in urban China’s labor markets. China Economic Review 23(2): 461–470.

23.

Liu

Meng

Zhang

(2000) Sectoral gender wage differentials and discrimination in the transitional Chinese economy. Journal of Population Economics 13: 331–352.

24.

Melly

(2005) Public–private sector wage differentials in Germany: Evidence from quantile regression. Empirical Economics 30(2): 505–520.

25.

Meng

(2012) Labor market outcomes and reforms in China. The Journal of Economic Perspectives 26(4): 75–101.

26.

Mueller

(1998) Public–private sector wage differentials in Canada: Evidence from quantile regressions. Economics Letters 60: 229–235.

27.

Oaxaca

(2007) The challenge of measuring labor market discrimination against women. Swedish Economic Policy Review 14: 199–231.

28.

Oaxaca

Ransom

(1994) On discrimination and the decomposition of wage differentials. Journal of Econometrics 61: 5–21.

29.

Powers

Yoshioka

Yun

(2011) Mvdcmp: Multivariate decomposition for nonlinear response models. The Stata Journal 11(4): 556–576.

30.

Rabe

(2011) Dual-earner migration: Earnings gains, employment and self-selection. Journal of Population Economics 24(2): 477–497.

31.

Reimers

(1983) Labor market discrimination against Hispanic and Black men. Review of Economics and Statistics 65: 570–579.

32.

Song

(2012) Poverty reduction in China: The contribution of popularizing primary education. China & World Economy 20(1): 105–122.

33.

Song

(2014) What should economists know about the current Chinese hukou system? China Economic Review 29: 200–212.

34.

Song

(2010) Ownership and earnings inequality in urban China. Frontiers of Economics in China 5(4): 582–603.

35.

Tunali

(1986) A general structure for models of double-selection and an application to a joint migration/earnings process with remigration. Research in Labor Economics 8(B): 235–282.

36.

Wang

(2005) Urban labor market discrimination against migrants (in Chinese). Social Sciences in China 5: 36–46.

37.

Zhang

Zhao

(2011) Social-family network and self-employment: Evidence from temporary rural–urban migrants in China. IZA discussion paper series, no. 5446. IZA, Bonn, Germany, January.