Explaining the state-owned enterprise wage premium in China: is it unobserved heterogeneity?

Abstract

State-owned enterprises (SOEs) in China pay higher wages than private firms. Is it because SOEs pay their workers wage premium or because they employ high-quality workforce? Using the latest methods and data, this paper accounts for the unobserved heterogeneity and estimates the SOE wage premium for the period 1995–2013. It is found that the wage premium has diminished since 1995 and has become insignificant since 2000. The significant wage gap between SOEs and non-SOEs can be explained by the fact that SOEs recruit high-quality workforce in correlation with SOEs’ industrial composition. This paper also evaluates the instruments used in the previous studies and rejects them through validity tests. The evidence suggests that the labour market in China is not segmented by ownership.

Keywords:

• SOE wage premium
• unobserved heterogeneity
• instrument validity
• implied ratio

Notes

Acknowledgements

This paper uses several data sources, including CFPS (China Family Panel Studies), CHIP (China Household Income Project), and China Labour Statistics Yearbook. The CHIP data are provided by China Institute for Income Distribution at Beijing Normal University. The author thanks these data centres for allowing him to use the data. He also thanks two anonymous referees for helpful comments. The author takes responsibility for the errors if any.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 See Knight and Song (2003, 2008), Xue (2008), Cai, Park, and Zhao, (2008), Chi, Li, and Yu (2011), Démurger, Li, and Yang (2012), Ge and Yang (2014), Sheng and Zhao (2013), and Xia et al. (2014).

2 Descriptions of the CHIP surveys and key findings in these data can be found in the references cited in the CHIP Dataset website http://ciid.bnu.edu.cn/chip/index.asp?lang=EN (comment: the URL was recently changed). For CHIP samples, weights can be constructed following the instructions in Song, Sicular, and Yue (2013). The results change little when weights are used, reflecting the fact that each sample is representative of its population. Thus, to save space, the weighted results are not reported.

3 For more details of the dataset, including the sampling procedure, the data cleaning process, the definition of variables and the construction of cross-section and panel weights, see the CFPS User’s Manual (2010) at http://www.isss.edu.cn/cfps/d/file/p/8ad5b0d73614815cb99800d42b6b4bf7.pdf

4 Many previous studies classify urban collective firms as SOEs. However, recent studies argue that they are similar to private firms in terms of the economic environment and constraints and thus should be considered as non-SOEs (Meng 2000; Ge and Yang 2014).

5 is equivalent to condition 1 of Altonji, Elder, and Taber (2005a). Altonji, Elder, and Taber (2002) give the precise conditions and formal model leading to condition 1 of Altonji, Elder, and Taber (2005a).

6 This technique has been applied in many subsequent studies for different countries; e.g. Cavalcanti, Guimaraes, and Sampaio (2010) for Brazil, Goyal (2009) and Kingdo and Teal (2010) for India, and Altonji, Elder, and Taber (2005a) and Eren and Ozbeklik (2015) for the USA. It has not been applied to China in the similar context.

7 The test is robust to misspecification of the number of unobserved factors, provided that the specified number is less than the true number, i.e. there is at least one unobserved factor in , and that the covariance structure of the observable and unobservable characteristics is identical in all sectors (Heckman and Scheinkman 1987).

8 Because is log hourly wage, but the CHIP sample does not have panel data on working hours, to construct log hourly wage panel, it is assumed that the unobserved hours worked in the previous years are the same as the observed hours worked in the current year. For the CFPS panel, the observables vector X also includes lags of X if they do not cause the collinearity problem.

9 The instruments pass the weak instrument tests and over-identification tests (see Table A1).

10 The CFPS 2014 panel weight is used, ‘rswt_natpn1014’. The unweighted results are very similar to the weighted results.

11 According to my conversation with a human resources manager in a state-owned steel firm, they pay workers the standard market wage rate for the same industry and qualification.

12 In 1995, the effect of adding the proxies to the regression on the estimated SOE wage premium is smaller than in subsequent years because before the 1997 reform the contract type of workers and the profitability of firms in SOEs did not yet link to market conditions as much as in later periods (Meng and Kidd 1997).

13 Dong and Bowles (2002) use ‘marriage’. Zhao (2002) uses ‘number of children’. Zhang (2004) uses ‘job assigned by the government’. Xing (2005) uses ‘number of members in a household’. Zhang and Xue (2008) use ‘father working in an SOE’, ‘mother working in an SOE’, and ‘living in urban area vs. suburban area’.

14 The controls in the first stage are the covariates vector in Section two.

15 Failure to reject in an over-identification test does not imply that all the instruments are valid. This is because in this test the crucial exogeneity assumption of the just-identifying instruments is not testable. Recently, Huber and Mellace (2014, 2015) propose tests of instrument validity in just identified models using moment restrictions implied by valid instruments. Their test does not reject the instruments either. However, their test can have low power when the instrument is slightly invalid.

16 These results are the same as those of Zhang (2004). Zhang’s conclusions are doubted as the negative selection on unobservable worker characteristics seems to be at odds with positive selection on observable worker characteristics. Zhang (2004) couldn’t explain this.

Additional information

Funding

The CFPS is funded by 985 Program of Peking University and carried out by the Institute of Social Science Survey of Peking University.