Wage differentials between urban and rural-urban migrant workers in China Li Zhang, Rhonda Vonshay Sharpe, Shi Li, William A. Darity Jr. PII: DOI: Reference:
S1043-951X(16)30132-8 doi: 10.1016/j.chieco.2016.10.004 CHIECO 987
To appear in:
China Economic Review
Received date: Revised date: Accepted date:
28 January 2014 18 October 2016 18 October 2016
Please cite this article as: Zhang, L., Sharpe, R.V., Li, S. & Darity, W.A. Jr., Wage differentials between urban and rural-urban migrant workers in China, China Economic Review (2016), doi: 10.1016/j.chieco.2016.10.004
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Wage Differentials between Urban and Rural-Urban Migrant Workers in China*
PT
Li Zhanga,**,Rhonda Vonshay Sharpeb,**, Shi Lic, William A. Darity Jr.d a
TE
D
MA
NU
SC
RI
School of Economics, Nanjing University of Finance and Economics, 3, Wenyuan Road, Xianlin College Town, Nanjing 210023, China b Women’s Institute for Science, Equity and Race, 9291 Laurel Grove Road, Mechanicsville, VA 23116 c School of Economics and Business Administration, Beijing Normal University, 19, Xinjiekouwai Dajie, Beijing 100875, China d African and African-American Studies and Economics, Sanford School of Public Policy, Research Network for Racial and Ethnic Inequality, Social Science Research Institute, Duke University, Box 90239, Durham, NC 277080239, USA ABSTRACT: Since the end of the 1980s, the number of migrants working in the urban labor market has increased dramatically. However, migrant workers are treated differently from urban workers. In this paper we examine the labor market discrimination against rural migrants from the point of view of wage differentials using CHIP-2007 data. We apply Jann pooled method to deal with index number problem and use Heckman two step model to correct selection problem when decomposing the wage gap. The decomposition results show that a significant difference in wage gains persists between the two groups as late as 2007. In 2007 migrants only earned 49% of urban workers’ income and 17 percent of the wage gap cannot be explained by observed factors. In detail, differences in educational attainment, work experience and distribution across industry, occupation, and ownership of enterprises account for most of the explained wage gap.
1.
AC CE P
Keywords: Chinese labor market, Wage differentials, Rural-urban migrants, Discrimination
Introduction Since 1950, rural-urban segregation in China has been enforced by a household registration system called “hukou”. Under the hukou system, individuals born in rural areas are assigned the “agricultural hukou” (migrant “hukou”), while those born in cities are assigned the “nonagricultural hukou” (urban “hukou”). Before the reform of 1978, about 80 percent of the Chinese population were restricted to live in rural areas and their future generations would be registered as “agricultural hukou” at birth. If rural hukou holders moved to the city, the necessities rationing system, which provided coupons for food, clothing, meat, and other basic needs to urban “hukou” residents, made it difficult, if not impossible, for migrants to survive in the city. Market-oriented reforms began in 1978, but by the mid-1990s, people could buy basic necessities in the market. Under the opening-up policy, China implemented an export-oriented growth model that generated huge demands for unskilled labor in the urban labor market. To meet the demand for unskilled labor in the urban market, the central government deregulate the hukou system, which gave local governments control over hukou policies. As a result, rural surplus labor migrated to cities and migrant *
This study has been supported by the Humanities and Social Science Research Project of Ministry of Education of China (Project No. 13YJC790212) and by the Jiangsu Government Scholarship for Overseas Studies. ** Corresponding author. E-mail address:
[email protected] (L. Zhang),
[email protected] (W. Darity),
[email protected] (S. Li),
[email protected](R. Sharpe).
1
ACCEPTED MANUSCRIPT
AC CE P
TE
D
MA
NU
SC
RI
PT
workers gradually became invaluable to the urban labor market. By the end of 2014, migrants were approximately 40% of the total urban labor force and 35% of the total nation labor force (National Bureau of Statistics of China, 2015). Despite their large share of the urban labor force, migrants encounter differential treatment in cities such as, lower wages, delayed payment or wage arrears, no written contracts, long working hours, and inadequate social security coverage (Li, 2010). Although the Lewis turning point generated upward pressure on the wages of migrant workers, a sizeable wage gap still exists between the urban and migrant working groups (Cai, 2010; Cai & Du, 2011). Previouse studies utilize the wage decomposition method to identify potential determinants of the wage gap between urban and migrants; however, the findings are inconsistent. Some studies find severe earning discrimination against migrants workers (Meng&Zhang, 2001; Wang, 2003; Deng, 2007; Liu, 2005; Xie & Yao, 2006; Wang, 2007). Others find that differences in endowments explain the wage gap (Yao et al., 2008; Lee, 2012; Xing, 2008; Démurger et al., 2009; Messinis, 2013). (See Table A.1. in the appendix). We offer three potential explanations for the inconsistency in results of the previous studies. First, migrant workers usually work more hours than urban workers to compensate for lower wages. Hence, the monthly or the yearly wage gap between migrant and urban workers may be smaller than the hourly wage gap. Therefore, studies that use the monthly or yearly earnings as the dependent variable (Démurger et al., 2009; Messinis, 2013) are more likely to undervalue the wage gap than studies that use the hourly wage as the dependent variable (Deng, 2007). Second, the endogeneity caused by the omission of some variables may bias results. For example, the omission of the distribution of workers by occupation and/or industry may overestimate the explanatory power of education (Lee, 2012; Démurger et al., 2009). Finally, and perhaps the most important, is the choice of the “non-discriminatory” wage structure – the index number problem. Studies that use the urban workers’ wage structure as the “non-discriminatory” wage structure may undervalue the discrimination effect (Xing, 2008) while studies that use the migrants’ wage structure as the “non-discriminatory” wage structure may overvalue the discrimination effect. Some studies have addressed the index number problem by averaging all estimation results (Démurger et al., 2009; Lee, 2012). This paper examines wage inequality between urban and migrant hukou holders employed in the urban labor market by addressing the three potential explanations offered above for the inconsistency of results in the current literature. Specifically, we: 1) apply Jann (2008) pooled method to mitigate the index number problem; 2) take the hourly wage as dependent variable rather than the monthly or the yearly wage; 3) control for the distribution of occupations, ownerships and industries to handle with the omission variables problem; and 4) use Heckman two steps model to correct some kind of the selection bias. Our decomposition of the wage gap finds that the hukou designation explains about 17% of the wage differentials between urban and migrant workers in 2007.
2.
Data and descriptive analysis The data for this study comes from 2007 wave of the Chinese Household Income Project (CHIP). The CHIP-2007 is a nationwide survey that covers three types of households: urban local households (hereafter termed urban), rural households, and rural-urban migrant households, which are located in an urban area, but hold an agriculture hukou (hereafter termed migrant). The inclusion of migrant households in the CHIP-2007 allows for the examination of differences in socio-economic outcomes between urban local residents and rural-urban migrants. The CHIP-2007 data contains 5000 urban local households 2
ACCEPTED MANUSCRIPT
AC CE P
TE
D
MA
NU
SC
RI
PT
(20,632 urban residents) and 5000 migrant households (14,683 migrants). Geographically, the survey covers a total of 15 cities from these provinces: Shanghai, Jiangsu, Zhejiang and Hebei are in the eastern region; Anhui, Henan and Hubei are in the central region; and Chongqing and Sichuan are in the western region. Despite the richness of the data, there is no public census data about the provincial distribution of the migrant workers; therefore, no proper weights could be used to weight the sample. The data provides detailed socioeconomic information such as monthly earnings, working hours, work conditions, access to social security, education, work experience, occupation, industry, and enterprise ownership. Additionally, the data include demographic information about the respondents such as age, sex, marital status, and ethnicity. The Hebei province lacked rural-urban migrants’ observations; therefore, the Hebei province is not included in the sample. Our sample is limited to workers 16 to 64 years of age who earned a positive wage and had responses for all key variables in Table 1 and Table 2. The sample used for this study contains 5,506 urban hukou observations and 6,141 observations for rural migrants. We recode the data as follows. The nine provinces were recoded accord to regional location: eastern region, central region, and western region. We recode the twenty types of industries into three categories: monopoly industry, competitive industry, and other industries. We follow Yue et al. (2011) to identify the monopoly grouping. In addition, we recode the seven types of occupations into three categories: whitecollar, blue-collar, and service. Finally, we recode the 16 types of ownership into five categories: stateowned enterprise, collective-owned enterprise, private-owned enterprise, foreign-owned enterprise, and individual-owned enterprise. Table A.2. (Appendix A) displays our classification of these variables. Descriptive statistics for individual characteristics are provided in Table 1 and the employment distribution for the urban and migrant samples are provided in Table 2. The mean hourly wage of urban workers is twice that of migrant workers. Compared to urban workers, migrant workers are more likely to be younger and to be single. The mean age of migrant workers is 31.7, about eight years younger than urban workers. Sixty-four percent of migrant workers are married, which is 20 percentage points less than the marriage rate for urban workers. As for productivity-linked characteristics, migrant workers average about 3.4 fewer years of schooling than urban workers, and about eight fewer years in their current occupation than urban workers. Since education and work experience play a significant role for earnings, less education and shorter job tenure create a disadvantageous position in the labor market for migrant workers. In terms of industry distribution, 68 percent of migrants work in competitive industries compared to 27 percent of urban residents. Moreover, enterprises ownership distribution shows that state-owned enterprises are more likely to employ urban local labor rather than migrants. As a result, more than 80 percent of migrant laborers are employed in private-owned enterprises or were self-employed. In addition, more than 80 percent of migrant workers are employed in the eastern and central regions, which are more industrial and developed. With respect to occupational distribution, 62 percent of migrant workers hold service jobs while almost the same proportion of urban workers hold white-collar occupations.
3
ACCEPTED MANUSCRIPT Table 1. Individual characteristics of urban workers and rural migrants in 2007
9.70
3.05 10.39 12.83
31.68 58.09% 1.78% 64.30% 8.87 4.71 6.92
5506
2.76 4.99 5.64
Diff. Mean or % 8.4 -1% -1% 20% 3.4 8.3 7.3
6141
NU
Source: CHIP data, authors’ calculation.
10.06
PT
40.06 57.55% 0.98% 84.37% 12.29 12.98 14.24
SC
Age Male Ethnic minority Married Years of schooling Tenure in Current Occupation Hourly Wage N
Rural migrants Mean or % Std. dev
RI
Urban workers Mean or % Std. dev.
Table 2. Employment distribution and hourly wage of urban workers and rural migrants in 2007
D TE
AC CE P
Industry Monopoly Industries Competitive Industries Others Occupation White-Collar Blue-Collar Service Ownership State-owned enterprises Collective-owned enterprises Private-owned enterprises Foreign-owned enterprises Individual-owned enterprises Region Eastern Central Western
MA
Distribution (%) Urban Rural workers migrant 21% 27% 51%
5% 68% 28%
15.9 11.2 15.2
8.6 6.7 7.2
60% 17% 23%
10% 28% 62%
16.6 11.3 10.3
7.6 7.5 6.5
60% 6% 18% 5% 11%
9% 4% 40% 5% 41%
15.1 12.4 12.7 20.6 10.3
7.5 8.8 6.5 8.8 6.7
17.3 11.3 10.8
7.9 5.8 5.8
51% 52% 29% 32% 20% 16% Source: CHIP data, authors’ calculation. For variable definitions, please see Table A-2.
3.
Hourly Wage (mean) Urban Rural workers migrants
Method—extended Oaxaca-Blinder decomposition
3.1.
General decomposition steps The descriptive statistics do not provide a clear picture about which factors influence the wage gap. To answer this question, we decompose the hourly wage differential using the Oaxaca-Blinder method (Blinder 1973; Oaxaca, 1973). Controlling for demographic characteristics such as age, sex, ethnicity, marital status, region; productivity-linked variables such as education (years of schooling) and work
4
ACCEPTED MANUSCRIPT experience (tenure in current occupation); and employment distribution across industry, occupation, and enterprise ownership, we estimate wage equations separately for each group using the Mincer (1974) wage equation:
RI
PT
where g indicates the group (urban workers or migrant workers here), Y is a vector of the log hourly wage, and X is a matrix of individual characteristics. After running OLS, we get each group’s estimated income equation: (3)
NU
SC
where hats denote estimated constants and coefficients from separate equations. The wage estimates are used to decompose the differences in mean log hourly wage between the two groups. The differential equation is: which can be rewritten as ：
D
MA
(5) where bars indicate the mean values of wage or endowments. Economic interpretations will emerge from the decomposition equation (5) where we assume that is the non-discriminatory return vector for individual endowments. To explicitly express the economic meaning of Oaxaca-Blinder decomposition, we rewrite equation (5) again as:
AC CE P
TE
Equation (6) shows that the wage gap between the urban group and the migrant group can be decomposed into two parts. The first term represents the composition effect, while the second represents the wage structure effect. The first term shows how much of the wage gap comes from endowments differentials, namely, how much of the wage gap would disappear if the migrant group has the same characteristics as the urban group. The term shows how much the urban group is favored by the wage structure, while displays how much the migrant group loses by the wage structure. Finally, the formula shows the effect of omitted variables on the wage gap. The first part in equation (6) is called the explained part, the portion of the wage gap that can be explained by differences in endowments between the two groups. This portion of the differentials is not viewed as discriminatory. In contrast, the latter three terms constitute the unexplained part, meaning the portion of the wage gap that arises from the differential treatment of the same endowments that is not explained by standard productivity theory. This part is often interpreted as discrimination against the lower income group. By applying the Jann (2008) pooled method, we reduce the effect of “index number problem”. To address potential endogeneity problems, we control for the distribution of occupation, industry and enterprise ownership. Finally, to address potential selection bias (Lee, 2012; Messinis, 2013), we utilize the Heckman model to correct for selection bias of family assistants.
5
ACCEPTED MANUSCRIPT 3.2.
Index number problem-choosing
SC
RI
PT
3.2.1. Index number problem The decomposition results are dependent on the choice of β*, the non-discriminatory wage structure. Theoretically, β* should represent the wage structure in the absence of labor market discrimination. However, this kind of wage structure cannot be captured in the real world. The choice of β* is important for reducing bias when using the wage decomposition method to identify statistical discrimination (Oaxaca, 1973; Fortin et al., 2011). For example, assume the urban group represents the non-discriminatory wage structure, namely , the wage differentials equation can be written as：
D
MA
NU
(7) In equation (7), the first term measures the explained portion of the wage gap. The remaining two terms are the unexplained portion, the discrimination against migrant workers. Since may overvalue the wage structure in the absence of discrimination, the term may overvalue the explained portion; hence, discrimination would be undervalued. In contrast, assume the migrant group represents the non-discriminatory wage structure, the discrimination against migrant workers will be overestimated. The choice of the non-discriminatory wage structure may produce significant differences in decomposition results, which may explain why the Deng (2007) estimate of discrimination is 47 percent higher than the Guo and Zhang (2011) estimate of discrimination, even though both studies use the same CHIP-2002 sample.
AC CE P
TE
3.2.2. Taking weighted average of the estimated coefficients of subgroups as β* One solution of index number problem is to posit a new wage structure using a weighted average expression. Reimers (1983) thought neither group’s observed wage-offer function would be likely to exist in a non-discriminatory world. So, she gave each of the group’s wage structure a 50 percent weight to produce a detailed decomposition. , (8) Cotton (1988) argued that the non-discriminatory wage structure would be closer to the group that had the larger population, so he weighted the wage structures proportional to representation in the labor market. , Neumark (1988) argued that the non-discriminatory wage structure should be derived from a theoretical model of discriminatory behavior, namely, employer discrimination theory (Becker, 1957; Arrow, 1973). Therefore, he defined β* as: where X is the matrix of individual characteristics describing each type1 of worker, Ω2 is a diagonal matrix showing the population distribution of different types of workers, and Λ3 is a vector presenting the non-discriminatory wage for each type of worker. 1
He categorized the workers into J types according to labor skill. Ω=diag(U1+M1, ……，UJ+MJ), Ui,Mi present the workers number of the ith type, taking urban and migrant workers as different groups. 2
3
Λj=
6
ACCEPTED MANUSCRIPT Oaxaca and Ransom (1994) generalized equation (10) as: (11) arguing that the matrix could be used as the weight:
NU
SC
RI
PT
where and are the covariate matrices for urban workers and migrant workers respectively. Neumark (1988) demonstrated that the non-discriminatory wage structure computed using the Ω weight was identical to , the OLS estimate obtained from the pooled sample including both groups. Therefore, equation (2) and equation (3) can be written as: (13) where X is the vector of individual characteristics of the pooled sample. We should caution readers that here X does not include a group membership indicator. Hereafter is referred to as omega coefficients:
D
MA
The omega weighted approach is preferred to the weighting method only using numerical proportion of each group since the former captures differences of individual characteristics between the two groups. However, Fortin (2008) and Jann (2008) pointed out that the omega approach may transfer some of the unexplained component of the differentials into the explained component, resulting in the underestimation of discrimination. Therefore, they suggested using regression estimates from a pooled model with a dummy variable for group membership: (15) as the
TE
where the hukou dummy functions as the group membership indicator. Hereafter, we label pooled coefficients.
Endogeneity and Heckman correction In wage regression, selection bias is likely to arise when the assumption of random participation is violated. In our case, a subgroup of migrant workers called family assistants work for their family’s business, but may not report any income. If the reporting of income is random, we do not have selection bias. However, if there are characteristics that predict the likelihood that a family assistant will not report income, the random assumption is violated and we have selection bias. Since running a small family business is popular among migrants, the proportion of family assistants is higher for this group than for urban workers. Therefore, we combine the Heckman two-step model4with the wage differential decomposition method to correct this selection bias. To get the unbiased estimators, we estimate the probit regression:
AC CE P
3.3.
(16) where refers that the i observation if a wage is observed; while refers to a family assistant without pay, and is a vector of exogenous variables believed to impact the decision of migrant workers to be family assistants. Using estimator from equation (16), we generate the inverse of the Mills’ ratio: th
Using the inverse of the Mill’s ration from equation (17), we estimate: 4
The detailed explanation of Heckit model refers to Adrian Colin Cameron and P. K. Trivedi (2010).
7
ACCEPTED MANUSCRIPT (18) to generate semiparametric estimates of
, where
is the estimate correcting for selection bias.
4.
to decompose the wage
PT
Since family assistants without pay are rare in the urban group, we only use gap instead of , where m refers to the migrant group. Results
OLS regression Estimators We generate four sets of estimates of β,– regressing the following Mincerian hourly earning functions separately:
SC
RI
4.1.
NU
；
, by (19) (20) (21)
AC CE P
TE
D
MA
； (22) where X is a vector consisting of the explanatory variables age, education, work experience, work experience squared, hukou5, ethnicity, marital status, region, industry, occupation and ownership. Independent of the earnings function estimated, the results in Table 3 show a similar pattern for age, ethnicity, marital status and region. Younger and married workers earn more than older and single workers. Workers in the east region, which is more industrialized – with developed manufacturing, service, and financial sectors, earn more than workers in other regions. As for the productivity-linked variables, educational attainment and job tenure have a positive effects on wages for both groups. But returns to these characteristics for each group are not the same. One more year of schooling increases the hourly wage for urban workers by 6.4 percent, but only increases the hourly wage of migrant workers by 4.8 percent. An additional year of work experience increases hourly wages by 4.3 percent for migrant workers and 3.2 percent for urban workers. The estimates for employment distribution are complicated. Urban workers receive lower returns to blue-collar occupations than to white-collar occupations. The opposite is true for migrant workers – bluecollar occupations have a higher return than white-collar occupations. The lower return to white-collar occupations is explained by the occupation distribution of migrant workers. Specifically, nearly 80 percent of migrant white-collar workers are clerks. Of the three subcategories of white-collar occupations – clerks, managers, and professionals, clerks earned the lowest wages. In contrast, only 40 percent of urban white-collar workers are clerks. On average, urban white-collar workers earn 26 percent more than urban blue-collar workers and migrant white-collar workers earn 5.7 percent less than migrant blue-collar workers. Additionally, migrant white-collar workers are more likely to be employed in competitive industries than in state-owned or foreign-owned enterprises. Of the 671 migrant white-collar workers, only 63 workers (9.4%) work in monopoly state-owned enterprises (SOEs). In contrast, 22 percent of urban white-collar workers are employed monopoly SOEs. The distribution of industry and ownership have a similar effect on wage earnings for both groups. Generally, workers employed in the monopoly industry earn more than workers employed in the competitive industry. Urban and migrants workers employed in monopoly SOEs earn 21 percent and 7.7 percent more, respectively, than their counterparts employed in the competitive industry. Urban workers 5
hukou is only included in the pooled regression model 4) as indicator variable for group membership.
8
ACCEPTED MANUSCRIPT
AC CE P
TE
D
MA
NU
SC
RI
PT
employed in foreign-owned enterprises (FOEs) earn 25 percent more than urban workers employed in individual-owned enterprises (IOEs). Migrants workers employed in SOEs earn 8.7 percent % more than migrant workers employed in IOEs. Similarly, migrants workers employed in FOEs earn 16.7 percent more than migrant workers employed in IOEs. The coefficients from the estimates of equations (19) and (20) are so different for urban and migrant workers that choosing either as the non-discriminatory wage structure would create the index problem.
9
ACCEPTED MANUSCRIPT Table 3. OLS regression coefficients of different samples using CHIP-2007 Urban
Rural
Omega
Pooled 0.1771***
Urban
AC CE P
TE
D
MA
NU
SC
RI
PT
-0.0027* -0.0046*** -0.0020** Age (-0.0012) (-0.0009) (-0.0007) 0.1134*** 0.1033*** 0.1166*** Married (-0.0259) (-0.0189) (-0.0155) -0.0006 -0.0117 0.0203 Han (-0.0804) (-0.0529) (-0.0457) 0.0637*** 0.0482*** 0.0661*** Education (-0.0031) (-0.0027) (-0.0019) 0.0324*** 0.0425*** 0.0376*** Experience (-0.0028) (-0.0033) (-0.002) -0.0006*** -0.0012*** -0.0007*** Experience2 (-0.0001) (-0.0002) (-0.0001) Region (Reference Category: Central) 0.4528*** 0.3195*** 0.3752*** Eastern (-0.0188) (-0.0159) (-0.0124) 0.021 0.0077 0.016 Western (-0.0232) (-0.0209) (-0.0158) Occupation (Reference Category: Blue-Collar) 0.2644*** -0.0574* 0.1839*** White-collar (-0.0231) (-0.0244) (-0.0156) 0.0577* -0.1455*** -0.0486** Service (-0.0288) (-0.0183) (-0.0158) Industry (Reference Category: Competitive) 0.2103*** 0.0766* 0.1500*** Monopoly (-0.0252) (-0.0328) (-0.0187) 0.1307*** -0.0537** 0.0274 Other (-0.0218) (-0.0188) (-0.0142) Ownership (Reference Category: IOEs) 0.0577 0.0874*** 0.1226*** SOEs (-0.0298) (-0.0259) (-0.0178) -0.0637 0.1804*** 0.0678** COEs (-0.0412) (-0.0349) (-0.0262) -0.006 0.0099 0.0006 POEs (-0.0315) (-0.0164) (-0.015) 0.2501*** 0.1674*** 0.2181*** FOEs (-0.0454) (-0.0349) (-0.0277) 0.8071*** 1.1699*** 0.7838*** Cons (-0.1057) (-0.0686) (-0.0561) N 5506 6141 11647 Note: Standard errors are given in parentheses. For variable definitions, please see Table A-2. ***, **, *denote statistical significance at the 1, 5, 10 percent level.
10
(-0.0157) -0.0041*** (-0.0007) 0.1175*** (-0.0155) 0.0131 (-0.0455) 0.0575*** (-0.0021) 0.0350*** (-0.002) -0.0007*** (-0.0001) 0.3795*** (-0.0123) 0.0136 (-0.0157) 0.1483*** (-0.0158) -0.0627*** (-0.0158) 0.1151*** (-0.0188) 0.0163 (-0.0142) 0.0799*** (-0.0181) 0.0389 (-0.0262) -0.0045 (-0.015) 0.2004*** (-0.0276) 0.9306*** (-0.0573) 11647
ACCEPTED MANUSCRIPT 4.2.
OLS regression with Heckman correction for migrants Since the coefficient of λ(·) is statistically different from zero, we use the Heckman model to correct for selection bias of migrant family assistants. The estimators after Heckman correction
PT
are presented in Table 4. Table 4. Heckman two-step regression coefficients for migrant workers using CHIP-2007
D
AC CE P
[95% Conf.
Interval]
0.0000 0.0000 0.9560 0.0000 0.0000 0.5640 0.0000 0.0000 0.1170 0.0000 0.1560 0.1910 0.0150 0.0000 0.5680 0.0010 0.0000
-0.0088 0.1074 -0.1640 0.0207 0.2830 -0.0436 0.0301 -0.0016 -0.1333 -0.1673 -0.0275 -0.0937 0.0190 0.0817 -0.0344 0.0766 1.1534
-0.0028 0.2371 0.1549 0.0417 0.3779 0.0799 0.0500 -0.0007 0.0149 -0.0557 0.1717 0.0187 0.1735 0.2895 0.0627 0.2852 1.5989
SC
-3.7600 5.2100 -0.0600 5.8400 13.6500 0.5800 7.8700 -4.7300 -1.5700 -3.9200 1.4200 -1.3100 2.4400 3.5000 0.5700 3.4000 12.1100
MA
0.0015 0.0331 0.0813 0.0053 0.0242 0.0315 0.0051 0.0002 0.0378 0.0285 0.0508 0.0287 0.0394 0.0530 0.0248 0.0532 0.1136
P>z
RI
z
NU
Std. Err.
TE
Coef. Log (hourly wage) Age -0.0058 Married 0.1723 Han -0.0045 Education 0.0312 Eastern 0.3304 Western 0.0182 Experience 0.0400 Experience2 -0.0011 White-Collar -0.0592 Service -0.1115 Monopoly 0.0721 Other -0.0375 SOEs 0.0962 COEs 0.1856 POEs 0.0141 FOEs 0.1809 Cons 1.3762 Selection to be workers with pay Age 0.0040 Male 0.7423 Married -0.3977 Education 0.0844 Cons 0.7295 Mills -0.8743 λ rho -1.0000 sigma 0.8743
0.0036 0.0588 0.0819 0.0100 0.1528
1.0900 12.6300 -4.8500 8.4600 4.7700
0.2740 0.0000 0.0000 0.0000 0.0000
-0.0031 0.6272 -0.5583 0.0649 0.4300
0.0111 0.8575 -0.2371 0.1040 1.0290
0.1501
-5.8300
0.0000
-1.1684
-0.5802
For variable definitions, please see Table A-2. 4.3.
The Oaxaca-Blinder decomposition with Heckman correction The Oaxaca-Blinder method decomposes mean wage differences into two components: one shows the composition effect which could be attributed to the differences in mean endowments of different groups; other shows the discriminatory effect which could be attributed to the differential returns to endowments. Furthermore, the aggregate composition and discriminatory effects can be decomposed to detailed effect of each independent variable. Table 5 has four sets of aggregate and detailed decomposition results. The odd columns, (1), (3), (5), and (7), report decomposition results without correction for selection and the even columns, (2), (4), (6), and (8), report the Heckman adjusted coefficients. The total estimated log urban-migrant hourly wage 11
ACCEPTED MANUSCRIPT
AC CE P
TE
D
MA
NU
SC
RI
PT
differentials are in Row 2. Row 3 reports the aggregate decomposition results attributable to endowment differences and row 4 reports how much of the total differential is explained by the differences in endowment. Similarly, row 5 reports the aggregate decomposition results attributable to differences in “treatment” to endowments – discrimination and row 6 reports how much of the total differential is explained by differences in “treatment” to endowments. The effect of the index number problem can be seen by examining row 6 in columns (1) and (3). The unexplained contribution to the total wage differentials using the migrants’ coefficients is 3.2 times higher than when the urban coefficients are used, 55 percent compared to 17 percent. In comparison with the pooled method, the omega method is likely to transfer some of the unexplained part into the explained part. Comparing column (5) to column (7) in row 6, we find that only 12 percent of the total wage gap is attributed to discrimination when using the omega coefficients, which is 15 percentage points lower than estimates using the pooled coefficients. The decomposition results using the urban coefficients and the pooled coefficients are closest to each other. However, using the urban coefficients underestimates discrimination by 10 percentage point as compared to estimates using the pooled coefficients. Correcting for selection bias produces significantly different decomposition results. Correcting for selection bias narrows the raw log hourly wage gap from 0.6461 to 0.5525. Since we only correct the selection bias for migrant workers, the narrowing of the wage gap comes from the increase in wages for migrant workers after the correction. Comparing row 5 and row 6 for each pair of odd and even columns reveals how correcting for selection bias impacts the unexplained portion of the decomposition. For example, comparing column (8) to column (7) in row 6, the contribution of discrimination decreases from 27 percent to 17 percent after correcting for selection bias. The detailed decomposition shows that a large part of the explained wage gap can be attributed to differences in education and work experience. For instance, the mean difference in the years of schooling explains 32 percent of the total wage gap, and tenure in the current occupation accounts for another 25 percent of the total wage gap. Differences in education and work experience explain about 57 percent of the wage gap between urban and migrant workers. Differences in occupation distribution, industry distribution and ownership distribution, explain another 29 percent of wage gap between urban and migrant workers. Because the choice of the omitted group is arbitrary, the contributions of dummy variables are “uninterpretable.6 Therefore, the decomposition results of region, industry distribution, occupational distribution and ownership distribution in the unexplained part reported in Table 5 cannot be interpreted. In general, compared to urban workers, migrant workers are favored in term of work experience, and discriminated against in term of education. For example, the return to education for migrant workers is less than the return to education for urban workers, which accounts for 60 percent of the wage gap. In contrast, the return to work experience for migrant workers is greater than the return to work experience for urban workers, which narrows the wage gap by 2 percent. Older workers earn less than younger workers in both groups; however, the impact of age on wages is less for migrant workers than it is for urban workers, which narrows the wage gap by 20 percent.
6
Blinder (1973, footnote 13), Oaxaca (2007), Jones (1983) mentioned the arbitrary of the choice of omitted group. Fortin et al. (2011) summarized this identity problem.
12
ACCEPTED MANUSCRIPT Table 5. Oaxaca-Blinder Decomposition Results (CHIP-2007)7 (1)
2
E[ln(wu)]- E[ln(wm)]
2’
Adjusted
(2)
(3)
(4)
(5)
Urban Coef.
Adjusted Coef.
Migrant Coef.
Adjusted Coef.
0.6461***
0.6461***
0.6461***
0.6461***
0.5525***
0.6461***
0.5525***
Explained
Omega Coef.
PT
Reference Group:
RI
1
(6)
(7)
(8)
Adjusted Coef.
Pooled Coef.
Adjusted Coef.
0.6461***
0.6461***
0.6461***
0.5525***
0.5525***
-0.0234**
-0.0234**
-0.0396***
-0.0498***
-0.0171***
-0.0278***
-0.0351***
-0.0380***
Education
0.2107***
0.2107***
0.1592***
0.1031***
0.2185***
0.1871***
0.1901***
0.1770***
Experience
0.1370***
0.1370***
0.0847***
0.0762***
0.1510***
0.1440***
0.1396***
0.1377***
Ethnicity
0
0
-0.0001
0
0.0001
0.0001
0.0001
0.0001
Marital Status
0.0243***
0.0243***
0.0222***
0.0370***
0.0251***
0.0334***
0.0252***
0.0301***
Region
-0.0131***
-0.0131***
-0.0095***
-0.0095***
-0.0109***
-0.0110***
-0.0112***
-0.0111***
Industry
0.0638***
0.0638***
0.0009
0.0038
0.0310***
0.0301***
0.0228***
0.0238***
Occupation
0.1083***
0.1083***
0.0263**
0.0127
0.1086***
0.1032***
0.0964***
0.0955***
MA
NU
SC
Age
0.0282**
0.0282**
0.0440***
0.0475***
0.0615***
0.0579***
0.0410***
0.0428***
Total
0.5359***
0.5359***
0.2881***
0.2210***
0.5679***
0.5170***
0.4690***
0.4578***
4
Contribution
83%
97%
45%
40%
88%
94%
73%
83%
Age
0.0594
0.0970**
0.0756
0.1234**
0.0531
0.1014*
0.0711
0.1116**
Education
0.1401***
0.2925***
0.1917***
0.4001***
0.1323***
0.3162***
0.1607***
0.3263***
Experience
-0.0194
-0.0102
0.0329
0.0505*
-0.0334*
-0.0173
-0.022
-0.0109
Ethnicity
0.0109
0.0039
0.011
0.0039
0.0108
0.0037
0.0108
0.0038
Marital Status
0.0064
-0.0371*
0.0085
-0.0497*
0.0057
-0.0461*
0.0055
-0.0428*
Region
0.0738***
0.0663***
0.0702***
0.0626***
0.0716***
0.0641***
0.0718***
0.0642***
Industry
0.0604***
0.0558***
0.1232***
0.1158***
0.0931***
0.0895***
0.1014***
0.0958***
Occupation
0.1573***
0.1371***
0.2393***
0.2327***
0.1570***
0.1422***
0.1692***
0.1499***
Ownership
-0.0159
-0.0194
-0.0317
-0.0387
-0.0492*
-0.0492*
-0.0287
-0.034
Constant
-0.3628***
-0.5691***
-0.3628***
-0.5691***
-0.3628***
-0.5691***
-0.3628***
-0.5691***
5
Total
0.1102***
0.0167
0.3580***
0.3315***
0.0782***
0.0355***
0.1771***
0.0947***
6
Contribution
17%
3%
55%
60%
12%
6%
27%
17%
N
11647
11647
11647
11647
11647
11647
11647
11647
PT ED
Ownership 3
AC
CE
Unexplained
7
Note: Jann (2008) provides Stata code which is very easy to apply an extended Oaxaca-Blinder decomposition method with Heckman two-step correction. For variable definitions, please see Table A-2.
13
ACCEPTED MANUSCRIPT
AC CE P
TE
D
MA
NU
SC
RI
PT
Conclusion The purpose of this study is to examine wage inequality between “agricultural hukou” (migrants) and “nonagricultural hukou” (urban residents) holders employed in urban labor market in China. We apply Jann (2008) pooled method to the estimation of log hourly wages controlling for distribution of occupations, ownerships and industries and correct for the selection bias of family assistants. Our findings support the conclusion that discrimination against migrant workers, existed in 2007. Based on the decomposition results using Jann (2008) pooled method, we find that discrimination against migrant workers explains 27 percent of the wage gap. After correcting for the selection bias of migrant family assistants, the wage gap narrows and discrimination against migrant workers decreases to 17%. Differences in employment distributions across industry, occupation, and the ownership of enterprises also play a significant roles in explaining the wage gap. About 30 percent of the wage gap can be contributed to these job distribution differences between urban and migrant workers. We find that differences in educational attainment – measured by years of schooling, and job tenure – measured by years in the current occupation, explain a large portion of the urban-migrant wage gap. These two productivity-linked characteristics account for more than half of the wage gap, 57%. Policies proposed to increase education attainment and the work experience of migrant workers might reduce the wage gap. However, the functionality of discrimination hypothesis (Darity, 2001) suggests that the degree of anti-migrant discrimination may intensify as migrants’ profile of productivity-linked characteristics becomes more similar to that of urban residents. Therefore, the effect of such polices may have a lower bound or even a U-shape effect with respect to the reduction of the wage gap. Additionally, since institutional discrimination crowds migrants into low-end jobs regardless of their educational attainment (Cai, 2007; Meng, 2012), migrant workers may choose to work rather than invest in human capital. Hence, the combined effects of pre-market discrimination, labor market segregation and in-market discrimination against migrant workers have contributed to the continued wage gap between urban and migrant workers. China, the most populous country, is expected to encounter labor shortages (Cai, 2010; Knight et al., 2011; Meng, 2012). Discrimination against migrant workers may aggravate rather than alleviate the labor shortage since it harms the efficiency of the labor market.
14
ACCEPTED MANUSCRIPT Appendix A Table A.1. Previous studies on the decomposition of wage gap between urban workers and rural migrants Results
Method
Wage variable
Meng and Zhang (2001) Liu (2005)
The data of Shanghai in 1995
More than 100 percent of the earnings differentials are attributable to discrimination More than 59 percent of the income differentials are attributable to discrimination.
Brown et al. (1980) decomposition method The Oaxaca and Ransom (1994) method
Hourly wage
Wang (2003)
The data of six cities in 2000
The Oaxaca method
Xie and Yao (2006)
The data of Zhejiang Province in the years 2003-2004 CHIP-2002
76 percent of the income differentials are attributable to discrimination 55.2 percent of the differential due to discrimination.
Discriminatory proportion of the wage gap was as high as 60 percent The population differential could interpret 138% of the wage gap which is 92 percentage point higher than hourly wage effect
Oaxaca-Blinder method Microsimulation method
Surprisingly, across all income quantiles the migrants’ group seems to be the favored one in labor market. Discrimination has declined to 54 percent from 2000 Only 10 percent of wage gap could be attributed to discrimination
The one percent sample of national data in 2005
The data of Zhejiang Province in 2007
CHIP-2002
Wang (2007)
The data of five cities in 2005 The data of five cities in 2005
Lee (2012).
Xing (2008)
Yao et al. (2008)
Not mentioned
Occupation
Taking population proportion as the weight
Not mentioned
Hourly wage Annual earnings
None
Using rural migrants as reference group Reporting a simple average of eight possible results
Not mentioned
Unconditional quantiles method and FFL method
Monthly wage
Occupation, ownership and industry
Not mentioned
Brown method
Occupation
Not mentioned
Not mentioned
Oaxaca-Blinder method
Hourly wage Hourly wage
None
Discrimination only accounted for less than 10 percent of the wage differentials The migrant group is the favored group
Oaxaca-Blinder method
Hourly wage
Control for selectivity in working participation using Heckman methods Not mentioned
Oaxaca-Blinder method
Monthly wage
Using urban workers as reference group
Not mentioned
The percentage of discrimination is as high as 40 percent.
Brown method
Hourly wage
Occupation, ownership and industry Occupation, ownership and industry Occupation
Reporting a simple average of two possible results Using urban workers as reference group
Taking population proportion as the weight
Not mentioned
None
NU
SC
Total annual gross income Hourly wage
The Oaxaca-BlinderCotton method
Hourly wage
MA
Messinis (2013)
Reporting a simple average of four possible results Reporting four results with four different weights respectively Not mentioned
PT ED
CHIP-2002
Selectivity problem
CE
Démurger et al. (2009)
Index problem
AC
Deng (2007)
The data of Beijing in 1995
Employment variables controlled Occupation distribution
PT
Data
RI
Study
15
Occupation
Ownership of the enterprises
Not mentioned
Not mentioned
Applying Lee method to control for selectivity in sector and estimate selectivity-corrected earnings function, but finding no quantitative difference Control for selectivity in high education using matching or IV methods
ACCEPTED MANUSCRIPT
Table A.2. Classifications of region, industries, occupations and enterprise ownerships. Categories
Types Included
Central
Anhui, Henan, Hubei
Western
Chongqin, Sichuan
PT
Shanghai, Jiangsu, Zhejiang, Guangdong
Competitive
Construction Industry; Wholesale and Retailing; Accommodation and Catering; Residential Services and Other Services Agriculture, Forestry, Animal Husbandry and Fishery; Mining Industry; Manufacturing; Real Estate Industry; Leasing and Commercial Service Industry; Scientific Research, Technical Service and Geological Prospecting Industries; Water Resources, Environment and Public Accommodation Management Industry; Education; Sanitation, Social Security and Social Welfare; Culture, Sports and Entertainment; Government and Public Administration; International Organization
Occupatio n
White-Collar Blue-Collar Service
SC
Others
RI
Production and Supply Industry of Power, Fuel Gas and Water; Transport, Warehousing and Post Industries; Financial Industry; Information Transmission, Computer Services and Software Industry
NU
Industry
Monopoly
Managers; Professionals; Clerks Agricultural Manual Workers; Industrial Manual Workers
MA
Region
Eastern
Commercial and Service Industry Workers
Collective-owned enterprises
Solely Collective-Owned Enterprises; Collective Holding Enterprises; Collective Holding Joint Venture Solely Private-Owned Enterprises; Private Holding Enterprises; Private Holding Joint Venture
Private-owned enterprises
D
Solely State-Owned Enterprises; State Holding Enterprises; State Holding Joint Venture; Government Agencies and Party Agencies; State and Collective Institutions
TE
Ownership
State-owned enterprises
References
AC CE P
Foreign-owned enterprises Solely Foreign-Owned Enterprises; Foreign Holding Joint Venture Individual-owned enterprises Self-Employed Individuals Source: CHIP-2007 urban survey questionnaire.
Arrow, K. (1973). The Theory of Discrimination. In O. Ashenfelter and A. Rees(Eds.), Discrimination in Labor Markets. Princeton, NJ:Princeton University Press. Becker, G. S. (1957). The Economics of Discrimination, Chicago: University of Chicago Press. Blinder, A. S. (1973). Wage Discrimination: Reduced Form and Structural Estimates. The Journal of Human Resources, 8(4), 436-455. Cai, F. (2007). Rural Urban Income Gap and Critical Point of Institutional Change. Economic Change and Restructuring, 40(1), 189-206. ____. (2010). Demographic Transition, Demographic Dividend, and Lewis Turning Point in China. China Economic Journal, 3(2), 107-119. Cai, F.& Du, Y. (2011). Wage Increases, Wage Convergence, and the Lewis Turning Point in China. China Economic Review, 22(4), 601-610. Cameron, A. C.& Trivedi, P. K. (2010). Microeconometrics Using Stata, College Station, Tex: Stata Press. Cotton, J. (1988). On the Decomposition of Wage Differentials. The Review of Economics and Statistics, 70(2), 236243. Démurger, S., Gurgand, M., Li, S.& Yue, X. (2009). Migrants as Second-Class Workers in Urban China? A Decomposition Analysis. Journal of Comparative Economics, 37(4), 610-628. Darity, W. J. (2001). The Functionality of Market-Based Discrimination. International Journal of Social Economics, 28(10/11/12), 980-986.
16
ACCEPTED MANUSCRIPT
AC CE P
TE
D
MA
NU
SC
RI
PT
Fortin, N., Lemieux, T.& Firpo, S. (2011). "Chapter 1 - Decomposition Methods in Economics," A. Orley and C. David, Handbook of Labor Economics, Elsevier, 1-102. Fortin, N. M. (2008). The Gender Wage Gap among Young Adults in the United States: The Importance of Money Versus People. Journal of Human Resources, 43(4), 884-918. Jann, B. (2008). The Blinder-Oaxaca Decomposition for Linear Regression Models. The Stata Journal, 8(4), 453479. Jones, F. L. (1983). On Decomposing the Wage Gap: A Critical Comment on Blinder's Method. The Journal of Human Resources, 18(1), 126-130. Knight, J., Deng, Q.& Li, S. (2011). The Puzzle of Migrant Labour Shortage and Rural Labour Surplus in China. China Economic Review, 22(4), 585-600. Lee, L. (2012). Decomposing Wage Differentials between Migrant Workers and Urban Workers in Urban China's Labor Markets. China Economic Review, 23(2), 461-470. Li, S. (2010). The Economic Situation of Rural Migrant Workers in China. China Perspectives, 2010(4), 4-15. Liu, Z. (2005). Institution and Inequality: The Hukou System in China. Journal of Comparative Economics, 33(1), 133-157. Meng, X. (2012). Labor Market Outcomes and Reforms in China. The Journal of Economic Perspectives, 26(4), 75. Meng, X.& Zhang, J. (2001). The Two-Tier Labor Market in Urban China. Journal of Comparative Economics, 29(3), 485-504. Messinis, G. (2013). Returns to Education and Urban-Migrant Wage Differentials in China: Iv Quantile Treatment Effects. China Economic Review, 26(0), 39-55. Mincer, J. (1974). Schooling, Experience, and Earnings, New York: National Bureau of Economic Research; distributed by Columbia University Press. National Bureau of Statistics of China. Statistical Communiqué of the People's Republic of China on the 2014 National Economic and Social Development, 2015-02-26， http://www.stats.gov.cn/english/PressRelease/201502/t20150228_687439.html, 2015-02-26. Neumark, D. (1988). Employers' Discriminatory Behavior and the Estimation of Wage Discrimination. The Journal of Human Resources, 23(3), 279-295. Oaxaca, R. (1973). Male-Female Wage Differentials in Urban Labor Markets. International Economic Review, 14(3), 693-709. Oaxaca, R. L. (2007). The Challenge of Measuring Labor Market Discrimination against Women. SWEDISH ECONOMIC POLICY REVIEW, 14 199-231 Oaxaca, R. L.& Ransom, M. R. (1994). On Discrimination and the Decomposition of Wage Differentials. Journal of Econometrics, 61(1), 5-21. Reimers, C. W. (1983). Labor Market Discrimination against Hispanic and Black Men. The Review of Economics and Statistics, 65(4), 570-579. Yue, X., Li, S.& Sicular, T. (2011). High Incomes in Monopoly Industries: A Discussion*. Social Sciences in China, (02). Deng, Q. (2007). Earnings Differential between Urban Residents and Rural Migrants: Evidence from OaxacaBlinder and Quantile Regression Decompositions. Chinese Journal of Population Science, 2, 8-16. (邓曲恒.
(2007).
城镇居民与流动人口的收入差异——基于oaxaca-Blinder和quantile方法的分解.
中国人口科学, (2), 8-16.) Guo, F, & Zhang, C. (2011). The Effect of Education Discrimination and Household Discrimination on the Wage Differential between Urban Workers and Migrant Workers. Issues in Agricultural Economy, 06, 35-42. (郭凤鸣, 张世伟. (2011). 教育和户籍歧视对城镇工和农民工工资差异的影响. 农业经济问题, 6,35-42.) Pan, J & Wei, H. (2013)The Citizenization of Rural Migrants. (Eds.). Annual Report on Urban Development of China,Vol.6. Beijing: Social Sciences Academic Press. (潘家华, 魏后凯. (2013). 农业转移人口的市民化, 中国城市发展报告. 北京: 社会科学文献出版社. ) Wang, Mei. (2003). Wage Differential during Transition: the Quantitative Analysis of Discrimination. Quantitative & Technica Economics, 5, 94-98.
17
ACCEPTED MANUSCRIPT (王美艳. (2003). 转轨时期的工资差异:歧视的计量分析. 数量经济技术经济研究, (5), 94-98.) Wang, Mei. (2007). Changes of Discrimination against Migrant Workers in China's Urban Labor Market. Chinese Labor Economics, 1, 109-119. (王美艳. (2007). 城市劳动力市场对外来劳动力歧视的变化. 中国劳动经济学, (1), 109-119.)
PT
Xie，S & Yao, X. (2006). The Quantitative Analysis of Wage Discrimination against Rural-Migrant Workers. Chinese Rural Economy, 4, 49-55.
RI
(谢嗣胜, 姚先国. (2006). 农民工工资歧视的计量分析. 中国农村经济, (4), 49-55.) Xing,C.(2008). Earning Differential between Urban Workers and Rural-Migrant Workers. Management World, 5, 55-64.
SC
(邢春冰. (2008). 农民工与城镇职工的收入差距. 管理世界, (5), 55-64.) Yao, X., Li, L., & Zhang, H..(2008). Wage Discrimination and Occupational Segregation against Rural LaborersEvidence of Zhejiang Province. An Academic Edition of ManaMaga, 3, 227-239.
AC CE P
TE
D
MA
NU
(姚先国, 李莉&张海峰. (2008). 农民工工资歧视与职业隔离——来自浙江省的证据. 管理学家(学术版), (3), 227-239.)
18
ACCEPTED MANUSCRIPT
AC CE P
TE
D
MA
NU
SC
RI
PT
Highlights Young married workers earn more than older workers Urban workers have a higher return for job tenure Migrant white-collar workers earn less than migrant blue-collar workers Migrant endowment weights underestimate the explained portion of the decomposition Differences in education (32%) and work experience (25%) are key factors in explaining the wage gap
19
