Inequality of opportunity in China: Evidence from pseudo panel data
Introduction
Popularized by Roemer, among other economists, the notion of equality of opportunity–also known as luck egalitarianism–provides the norm of a morally fair distribution of resources with which an individual should be endowed (Roemer, 1998).1 Underlying this notion is the idea that the part of inequality in outcomes that is attributed to circumstances is unjustifiable (Arneson Richard, 1989, Arneson Richard, 1990; Cohen, 1989; Dworkin, 1981a, Dworkin, 1981b; Rawls, 1971). Theoretical studies have shown that an unequal initial endowment can result in inefficient resource allocation and a waste of human capital (Banerjee & Newman, 1993; Galor & Zeira, 1993; Heathcote, Storesletten, & Violante, 2005). Empirical findings, though not conclusive, suggest that inequality of opportunity may hinder economic growth in the long run (Ferreira, Lakner, Lugo, & Özler, 2017; Marrero & Rodríguez, 2013).2 Faced with the rising income inequality worldwide, both economists and policy makers have realized the importance of leveling the play field for national and global development (Bourguignon, Ferreira, & Walton, 2007; Ferreira & Walton, 2005; Roemer, 1998).3
Although China ranks among the economies that have achieved the fastest growth rate in recent decades, it still faces the challenges of inequality of opportunity. For example, the rigid son preference leads families to allocate resources toward boys (Bowles, Gintis, & Groves, 2009); and the hukou system deprives rural migrant workers of the same public welfare as their urban counterparts (Afridi, Li, & Ren, 2015).4 Moreover, the initial gap can be further enlarged through intergenerational transfer (Li, 2007; Luo et al., 2019; Walder & Hu, 2009; Wang, Luo, Zhang, & Rozelle, 2017).
Using a nationally representative household survey – the China General Social Survey (CGSS) – this paper studies the degree to which the observed income inequality in China can be attributed to unequal opportunity. Following the literature that draws on Roemer's (1998) conceptual framework to study inequality of opportunity (IOP), we categorize the factors that could affect individuals' advantage: those that are at least partly subject to individuals' volition and choice (effort variables); and those that are due to circumstances beyond their control (circumstance variables). Adopting the parametric approach proposed by Bourguignon, Ferreira, and Menéndez (2007), we estimate the Mincer equation, regressing individuals' advantages on both the effort and the circumstance variables. We then use the point estimates to conduct a counterfactual analysis on the changes in income inequality when all the individuals in the sample share the same circumstances. The difference between the actual and the counterfactual income inequality provides a measure of IOP of China.
Among the studies that use the parametric method to measure IOP, the majority estimate the Mincer equation by OLS due to the lack of longitudinal household surveys, especially in developing countries (Bourguignon, Ferreira, & Menéndez, 2007; Choudhary, Muthukkumaran, & Singh, 2019; Ferreira, Gignoux, & Aran, 2011; Hassine, 2011; Hufe, Peichl, Roemer, & Ungerer, 2017; Marrero & Rodríguez, 2012; Singh, 2012; Suárez & Menéndez, 2018).5 Ferreira and Gignoux (2011) prove that in the presence of the issue of omitted circumstance variables, the IOP index calculated from OLS estimates of the Mincer equation is a lower-bound estimate of the true measure. A small number of recent studies draw on longitudinal household surveys to measure IOP and argue that the fixed-effects estimates produce an upper-bound estimate of the true IOP, assuming that all the circumstances are time-invariant. The upper-bound interpretation stems from the fact that all the individual fixed-effects are equalized in the counterfactual calculation, but they could include factors that cannot be considered circumstances (Carranza, 2020a, Carranza, 2020b; Hufe, Peichl, & Weishaar, 2019; Niehues & Peichl, 2014).
The pseudo panel, constructed from repeated cross-sectional surveys, provides an upward correction of the current lower bound and reduces the risk of overfitting to some extent.6 Deaton (1985) suggests defining cohorts with fixed membership and aggregate individual observations at the cohort level, so as to track the cohort fixed effects when the genuine longitudinal household survey is not available. The pseudo panel consists of cohort means. Deaton shows that the within estimators of the pseudo panel are consistent, once corrected for the measurement error. Realizing that typical efforts to address the downward bias by essentially increasing the number of types examined could result in upward distortion due to the overfitting issue, Brunori, Peragine, and Serlenga (2019) argue that the best model specification should balance the two sources of bias. By introducing the cohort fixed effects, the pseudo panel model can explain more heterogeneity in the outcome variable, while being more parsimonious than the model in which the full set of interactions of variables defining the cohort explicitly enters. Brunori et al. (2019) use the cross-validation method to find the best model specification, which may lie somewhere between the most parsimonious model, in which all the circumstance variables enter linearly, and the fully interacted model. The pseudo panel, by its construction, stands in the middle ground, thus providing a model that could be closer to the best model specification, in the sense of Brunori et al. (2019).
Our list of effort variables consists of education, work experience in the nonagricultural and formal sectors, and Chinese Communist Party (CCP) membership. The circumstance variables include gender, hukou at birth, and a set of paternal characteristics that existed when the respondent was age 14. As some of the effort variables and all the circumstance variables are time-invariant, the cohort within estimator does not identify the coefficients of these variables. Therefore, we further apply the Hausman-Taylor estimator to the pseudo panel sample to estimate the coefficients of observed time-invariant covariates in the Mincer equation.
We find, according to the estimates of the Mincer equation, that the circumstance variables have a greater impact on individuals' income advantage than effort variables have. Counterfactual analysis shows that equalizing only observed individual circumstance variables brings about a very limited reduction in income inequality; however, there is a significant drop in income inequality when we equalize both observed circumstances and estimated cohort fixed effects, indicating the important contribution from cohort-level circumstances to the observed income inequality. In our discussion in Section 4, we further decompose the full sample by region and by generation to study the spatial and generational variation of IOP in China. The results show that the western provinces have higher IOP than the eastern and central provinces and that the IOP of younger cohorts is lower than that of older cohorts. We also find heterogeneities in the contributions of individual circumstances to income inequality across subsamples.
Our analysis based on the pseudo panel estimates – which allows us to distinguish between individual and cohort circumstances – enriches the literature that studies the inequality of opportunity in China. The previous research focuses primarily on the contribution of individual circumstances. Zhang and Eriksson (2010) were the first to measure IOP in China between 1989 and 2006, using the OLS estimates of the Mincer equation. Their results show the significant effect of parental characteristics on income inequality. Golley and Kong (2018) find that the hukou system is the most important circumstance determining the inequality of opportunity in education in China. Similar to Golley, Zhou, and Wang (2019), we find that gender and paternal attributes contribute more than hukou of birth to the observed income inequality. Moreover, our results suggest that circumstances at the regional and temporal levels have a stronger influence than individual circumstances have on China's current income inequality.
The rest of the paper is organized as follows. In Section 2, we first introduce the conceptual framework; we then discuss the identification strategy and describe the data. In Section 3, we present the pseudo panel estimates and the results of the counterfactual analysis for the full sample. In Section 4, we discuss the variations in IOP across regions and across age cohorts. We conclude and discuss potential extensions for future research in Section 5.
Section snippets
Identification strategy
This section starts with the conceptual definition of equality of opportunity.7 For individual i in a population of size N (i ∈ {1, …, N}), we can distinguish between all the variables that could affect his or her advantage yi. That is, we can identify circumstances that are
Determinants of income advantage
Table 2 shows the pseudo panel estimates of the structural form of the Mincer equation (Eq. 5) for the full sample. Among effort variables, education and CCP membership are both positively associated with individual income. The impact of years of schooling on individual income is statistically highly significant and has a fairly large economic magnitude: one additional year of education increases log income by 0.101 standard deviation. Years of experience in the nonagricultural sector, however,
Discussion
The analysis in the previous section focused on the determinants of income advantage and IOP for the full sample. In this section, we explore how IOP in China varies across regions and across generations. We find that the leading contributors to IOP are different across regions and that IOP exhibits a decreasing trend over generations.
Conclusion
We construct a pseudo panel sample from the China General Social Survey to calculate the inequality of opportunity in China, which is measured by the difference between actual income inequality and counterfactual income inequality when the heterogeneities in the circumstance variables are purged. The pseudo panel enables us to distinguish between individual and cohort circumstances, and we argue that equalization of cohort circumstances provides an upward correction of the current lower bound
Acknowledgements
Xinchen Dai would like to thank Jean-louis Arcand for his guidance during Ph.D studies. We would also like to thank Salvatore Di Falco, Jaya Krishnakumar, Xavier Ramos, Lore Vandewalle, and Martina Viarengo for their helpful comments. We are grateful to the editor Le Wang and the two anonymous referees whose constructive comments helped us to greatly improve the paper. This research was supported by the Postdoc Program of Zhejiang Gongshang University (No. 284223), the National Social Science
References (59)
- et al.
Social identity and inequality: The impact of China’s hukou system
Journal of Public Economics
(2015) - et al.
The long-term health and economic consequences of the 1959–1961 famine in China
Journal of Health Economics
(2007) Panel data from time series of cross-sections
Journal of Econometrics
(1985)- et al.
Inequality of opportunity in China’s educational outcomes
China Economic Review
(2018) - et al.
Equality of opportunity and luck: Definitions and testable conditions, with an application to income in France
Journal of Public Economics
(2009) Family background, financial constraints and higher education attendance in China
Economics of Education Review
(2007)- et al.
Inequality of opportunity and growth
Journal of Development Economics
(2013) Famine, fertility, and fortune in China
China Economic Review
(2011)- et al.
Inequality of opportunity and household education expenditures: Evidence from panel data in China
China Economic Review
(2019) - et al.
Minimum MSE estimation of a regression model with fixed effects from a series of cross-sections
Journal of Econometrics
(1993)