Inequality of Opportunity in China s Labor Earnings: The Gender Dimension

28 China & World Economy / 28 50, Vol. 27, No. 1, 2019 Inequality of Opportunity in China s Labor Earnings: The Gender Dimension Jane Golley, Yixiao Zhou, Meiyan Wang* Abstract This paper investigates the inequality of opportunity in China s labor earnings, defined as the component of inequality determined by personal circumstances that lie beyond the control of an individual, of which gender is one, as opposed to the component determined by personal efforts. Using the Survey of Women s Social Status in China (2010), we measure the share of inequality of opportunity in the total inequality of individual labor earnings for people aged 26 55 years, and separately for six birth cohorts and for female and male subsamples. Gender is revealed as the single most important circumstance determining nationwide individual labor earnings, with one s region of residence, father s occupation, father s education, birth cohort and holding rural or urban hukou also playing significant roles. A further investigation into the roles of circumstances and personal efforts (including education level, occupation, Communist Party membership, migration and marital status) confirms that circumstances play an alarmingly high role in shaping labor earnings distribution in China, and reveals notable gender differences that cannot be attributed to personal effort alone. These results provide the basis for recommending ways to improve gender equality of opportunity in the future. Key words: China, gender, inequality of opportunity JEL codes: D31, D63, J16 I. Introduction Compared to the millennia-long history and culture that traditionally favored men, the establishment of the People s Republic of China has led to marked improvement in the socioeconomic status of Chinese women. During the Maoist era (1949 1976), agricultural collectivization and the Great Leap Forward brought millions of Chinese *Jane Golley, Acting Director, Australian Centre on China in the World, Australian National University, Australia. Email: jane.golley@anu.edu.au; Yixiao Zhou, Lecturer, School of Economics, Finance and Property, Curtin University, Australia. Email: yixiao.zhou@curtin.edu.au; Meiyan Wang, Professor, Institute of Population and Labor Economics, Chinese Academy of Social Sciences, China. Email: wangmy@cass. org.cn. This study was supported by the National Natural Science Foundation of China (Nos. 71473267 and 71642003).

Inequality of Opportunity in China s Labor Earnings 29 women into the productive sphere, where they were considered essential for socialist construction and required to hold up half the sky. The principle of gender equality was written into China s Constitution in 1954, entitling women to equal pay for equal work, with significant advances in marriage laws, paid maternity leave and other protective policies favoring women. In the post-mao era, successive leaders have affirmed China s commitment to the basic national policy of equality between men and women. This was evident in President Xi Jinping s address to the Global Leaders Meeting on Gender Equality and Women s Empowerment in September 2015, where he, like Mao, stressed women s important role in holding up half the sky, and called for global efforts to ensure that women would share equally in the achievements of development (MFA, 2015). However, as in all countries across the globe, achieving gender equality in China has been easier said than done. In urban China, despite a narrowing of the gender gap in educational attainments in recent years to the point where young urban Chinese women now out-educate their male contemporaries (Li, 2010; Zhang and Chen, 2014; Golley and Kong, 2018) there has not been a concomitant narrowing of the gap in individual earnings. Instead, a number of studies have confirmed an increase since the mid 1990s, much of which is attributed to gender discrimination rather than observable factors, such as gender differences in human capital or occupational choices (e.g. Wang and Cai, 2008; Zhang J. S. et al., 2008; Li et al., 2011). In rural China, gaps in both education and earnings are even more substantial (e.g. Zhang et al., 2007; Hannum et al., 2009; Zeng et al., 2012). Of course, gender is not the sole dimension of the income inequalities that have characterized China s rapid growth and development since the late 1970s welldocumented inequalities along regional, urban rural and socioeconomic divides have also been substantial. Recent overviews on the causes of income inequality in China include Gustafsson et al. (2008), Li et al. (2013), Knight (2014) and Zhou and Song (2016). Researchers in this area have also probed various causes in more detail. Kanbur and Zhang (2005) and Golley (2007) focused on regional inequalities; Sicular et al. (2007) examined rural urban inequalities; Piketty et al. (2017) looked at wealth inequality; and Golley and Kong (2013) probed inter-generational inequalities in education. But what if gender was the most significant of all of these? For women to have an equal share in the achievements of development, fundamental changes would need to occur. This paper sets out to explore the factors that have contributed to inequality of opportunity in China s individual labor earnings, with a particular focus on gender. The economic literature on inequality of opportunity begins with the premise that

30 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 the observed inequality in any particular economic outcome, such as earnings, can be attributed to two components. The first component derives from the different circumstances in which individuals find themselves, and over which they have no control for example, their gender, place of birth or the socioeconomic status of their parents. The second derives from the different levels of effort that individuals may exert to influence a given outcome for example, how hard they study, the occupation they choose, or whether they choose to marry. This distinction is crucial because inequality stemming from personal effort can, at least to some extent, be justified as being fair and reasonable, while inequality stemming from circumstances for example, being born male or female cannot. Various empirical methods have been used to estimate inequality of opportunity for a wide range of outcomes, including household per capita income and consumption expenditure (Marrero and Rodríguez, 2012); individual annual, monthly and long-run income (e.g. Bourguignon et al. 2007; Checchi et al., 2010; Björkland et al., 2012); educational attainments (Golley and Kong, 2018); and health outcomes (Jusot et al., 2013). Studies focused on income have revealed significant cross-country variation in the share of inequality of opportunity in total income inequality, ranging from close to 1/3 in Brazil and Guatemala (Ferreira and Guignoux, 2011) and 1/4 for India (Singh, 2012), to below 5 percent in Denmark, Finland, Germany, the Netherlands, Norway and Slovakia (Marrero and Rodríguez, 2012). 1 A number of different (and different numbers of) circumstances have been identified in these studies, including father s (or parents ) occupation and education; geographical location (region of birth for single-country analyses, country of birth for crosscountry analyses); and race, ethnicity or caste (in the case of India). For individuallevel analyses, male-only samples tend to be used, with a few exceptions. For example, Ferreira and Gignoux s (2011) analysis of individual labor earnings in five Latin American countries revealed a contribution of gender to total inequality ranging from just 0.2 percent in Colombia to 5.8 percent in Guatemala, with other family background circumstances being more important, particularly parental education and father s occupation. De Barros et al. (2009) found a similar share for gender in the inequality 1 See Brunori et al. (2013) and Ferreira and Peragine (2015) for more comprehensive surveys of the methods used and results found. In the one published study (to our knowledge) that applies these ideas to Chinese income, Zhang and Eriksson (2010) found an exceptionally high share of inequality of opportunity in individual income inequality, increasing steadily from 46 percent in 1989 to 63 percent in 2006. However, much of this is attributed to the inclusion of parental income in their set of circumstances, along with highly disaggregated measures of both parents education and occupation status which means that these results are not comparable with others in the literature, nor with ours.

Inequality of Opportunity in China s Labor Earnings 31 of individual labor earnings in Mexico, at 3 4 percent. In their analysis of individual income inequality in Australia between 2001 and 2013, Martinez et al. (2017) similarly showed the relatively small contribution of gender to total income inequality of less than 6 percent (compared to more than 50 percent for father s occupation, the most important circumstance). These small contributions contrast starkly with the dominant role we find played by gender in the empirical analysis of China that follows. In the one study that focused specifically on gender, Hederos et al. (2017) found that it explained 13 percent of total inequality, making it the most important circumstantial determinant of long-term income in Sweden a point that we confirm for the case of China as well. 2 In separate analyses for each gender, they found that circumstances (excluding gender) explain more of male than female income inequality, leading them to conclude that there is greater equality of opportunity among women than men. The opposite turns out to be true for China. In these cited works, the role of personal effort has largely been overlooked, with the most common estimation method subsuming this into the error term (as will be explained further). One notable exception is by Bourguignon et al. (2007), a much-cited paper that examines individual (male) real hourly earnings across seven birth cohorts in Brazil. Their analysis explicitly considers the relationship between circumstances (race, region of birth, parental schooling and father s occupation) and three effort 3 variables: the individual s own schooling attainment; a migration dummy; and a variable for labor status (indicating whether the worker is a formal employee or employer, an informal employee or self-employed). We use this method to examine the ways in which the complex interaction between circumstances and effort differs across genders in the Chinese context. We find this approach highly complementary to the more standard analyses of the gender wage gap in China, of which the citations above are only a few. Many of these use the Oaxaca Blinder decomposition, in which wage differences between men and women are apportioned into differences between male and female coefficients (the unexplained portion) broadly interpreted as gender discrimination and differences between male and female characteristics (the explained portion), such as age, education 2 Based on their results in Table 6, which exclude IQ and non-cognitive ability from the set of circumstances, making these results more comparable to ours. However, they are not directly comparable as their circumstances also include parental income, along with number of siblings. 3 Effort is placed in inverted commas here to stress the fact that there is ongoing debate over what constitutes effort, particularly given its complex relationship with circumstances. For reading ease, we don t use inverted commas throughout the paper, but wish to stress that the application of the term effort in this literature is not perfectly aligned with the definition of the word as it is commonly understood.

32 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 level, occupation, marital status, industry and province (Blinder, 1973; Oaxaca, 1973; Heshmati and Su, 2017; Song et al., 2017). In contrast, the inequality of opportunity approach highlights the fact that gender is one of many circumstances that determine total income inequality both directly and indirectly via personal effort. The fact that we find gender to be the single most important contributor to earnings inequality in this context thus complements the need for ongoing research into the underlying causes of gender earnings inequality at the micro level. The paper proceeds as follows. Section II explains the concept of inequality of opportunity and presents the method we use to measure it, alongside a complementary method that enables a deeper (albeit imperfect) consideration of the interrelated roles of circumstance and effort in determining individual labor earnings. Section III presents the data and baseline regression results, which are used to calculate the share of inequality of opportunity in total income inequality, and also the partial contributions made by each of the circumstance variables. Having identified gender as the dominant circumstance contributing to earnings inequality in China, Section IV looks further at the relationship between circumstances and effort in the nationwide sample and in the male and female subsamples. Section V concludes with thoughts about what kind of equal opportunity policies might in fact enable Chinese women to hold up half the sky. II. Circumstances, Effort and Inequality of Opportunity In his pioneering work, Roemer (1993, 1998) defines equal opportunity as a situation in which the distribution of a given outcome is independent of circumstances; that is, all individuals who exert the same level of effort will achieve the same outcome, regardless of their circumstances. Dividing the population into groups of people or types with identical circumstances and measuring the extent to which this condition is not satisfied provides one measure of inequality of opportunity. This involves suppressing withintype inequality, which can be explained by variation within each type, attributed to variation in effort and calculating the extent of between-type inequality based on the mean levels for each type as an absolute measure of inequality of opportunity (IOA). For comparability across different datasets and countries, it is more common to focus on the share of between-type inequality in total inequality, or relative inequality of opportunity (IOR). To formalize these ideas, Roemer (1998) defined a finite population of individuals,, each of whom has achieved an economic outcome (in our case, annual labor earnings), y i, with distribution, {y i }. Earnings are assumed to be determined by a vector of circumstances, C i, and a vector of efforts, E i, which, as noted above, will be at least

Inequality of Opportunity in China s Labor Earnings 33 partially determined by circumstances, implying that y i = f(c i, E i (C i, v i ), u i ). The sample population can then be divided into types, which by definition contain individuals with identical circumstances. Formally, this requires dividing the population into K types, given П {T 1,, T K } with distributions {y k i}, where each individual in type k has identical circumstances:. The number of types is therefore П. A direct method for calculating inequality of opportunity suppresses within-type inequality by assigning the mean of their type to every individual and measuring the inequality in the distribution of those means. This yields an absolute scalar measure of inequality of opportunity, IOA = I({μ k i }), where {μ k i } is the distribution obtained from replacing each individual outcome, y k i, with its type-specific mean, μ k, and I( ) is an appropriate index of inequality. The associated relative measure is given by IOR = I({μ k i})/i({y i }). We choose mean log deviation, or generalized entropy (GE(0)), as our index of inequality. 4 These calculations would be straightforward using non-parametric methods, but this requires sample sizes far beyond what is available to us. Instead, we opt for what has become the standard parametric method, explained at length in Bourguignon et al. (2007) and Ferreira and Gignoux (2011), and summarized here. We begin by approximating the relationship between earnings, circumstances and effort with the following structural form:. (1). (2) Substituting Equation (2) into Equation (1) yields the reduced-form regression:, (3) where, y is labor earnings and C is a vector of discrete circumstance variables. Using the estimated coefficients β^ and the actual values of circumstances, we construct a distribution, {y^ }, where y^i = exp(β^ C i ). By replacing each y i with its prediction, given the vector of circumstances (which is identical for all individuals of the same type), all within-group inequality is eliminated, giving direct estimates of inequality of opportunity: IOA = I({y^ }) and IOR = I({y^})/I({y k i}). The vector of observed circumstances will only be a subset of all relevant circumstances that affect individual outcomes, constrained primarily by data availability. As long as some unobserved circumstances are correlated with the observed circumstances (the most obvious contender being IQ, which is almost certainly 4 For details on why GE(0) is the best measure, see Ferreira and Gignoux (2011).

34 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 correlated with one s father s education and one s own earnings), the estimated β^ coefficients will be biased, and cannot be interpreted as causal links between a given circumstance and the outcome. However, for the overall measure of inequality of opportunity, this is not important: adding more circumstance variables to the observed set would necessarily increase the estimates of IOA and IOR, implying that these are lower-bound estimates of the true inequality of opportunity that would prevail if all circumstance variables could be observed (Ferreira and Gignoux, 2011). Importantly, this lower-bound result holds regardless of the (potentially complex) relationship between circumstances and effort. To give an example that is relevant here, if rural Chinese men are more likely than rural Chinese women to work outside of agriculture and earn higher incomes as a consequence, this may partially be a result of the extra effort rural parents put into their sons careers, in the form of financial support and other kinds of encouragement (which we cannot measure). This would place an upward bias on the β^ coefficient for gender (with the excluded dummy being female), via this indirect impact of effort (unobserved in the error term). While this leads us to stress again that the estimates cannot therefore be treated as causal, we maintain that a coefficient biased in this way would be entirely appropriate for a genuine understanding of equality of opportunity because choices made by one s parents are beyond one s own control. 5 We are also interested in the partial contributions of each of the circumstance variables (e.g. gender, father s education, father s occupation, hukou and region). We follow Björkland et al. (2012) and Shorrocks (2012), using a Shapley-value decomposition to measure the contribution (in terms of correlation, not causation) of each circumstance to the observed inequality of opportunity. This method accounts for the well-known problem that inequality decompositions may depend on the order in which inequality from a particular circumstance is measured, and so uses an average across all possible orderings of all circumstances. To be more specific, the change in inequality that arises when a new circumstance, such as gender, is added to a set of circumstances depends on the sequence of inclusion of different circumstance variables. Therefore, the contribution of each circumstance is measured by the average change in inequality over all possible inclusion sequences. One advantage of the Shapley decomposition is that the sum of the Shapley value of each circumstance is equal to the total contribution of these circumstances to inequality of opportunity. 6 That said, we 5 See the debate between Roemer (1998) and Brian Barry, laid out in Roemer (1998, pp. 23 25) for a lengthier exposition of this point. 6 This is not the case for the method commonly used in the literature, as outlined in Bourguignon et al. (2007).

Inequality of Opportunity in China s Labor Earnings 35 still concede that these decompositions need to be treated with caution, given the biases acknowledged here. Formally, the change in our absolute measure of inequality of opportunity, IOA, when circumstance C is added to a subset M of circumstances is given by:, (4) where C T denotes the entire set of l circumstances and M is a subset of C T that includes all m circumstance variables, with the exception of C. IOA M is the absolute measure of income inequality for the subset of circumstances M and is the measure obtained after adding circumstance C to subset M. Therefore, the contribution of circumstance C to IOA is defined by:, (5) where. In Section IV we will return to the structural form presented in Equations (1) and (2), in which it is clear that an individual s earnings depend on his or her circumstances through two distinct channels: a direct channel, reflected in the coefficients on each circumstance in Equation (1); and an indirect channel, via the effect of his/her circumstances on his/ her effort in Equation (2). In their application to male earnings in Brazil, Bourguignon et al. (2007) attempted to separately estimate these direct and indirect effects; however, in a corrigendum they conceded that this is not possible (Bourguignon et al. 2011). While this means that Equations (1) and (2) cannot be used in the way originally intended, they remain useful for stressing the key point in what follows: that the relationship between circumstances and personal effort differs in significant ways across genders. III. Measuring Inequality of Opportunity 1. Data and Baseline Regressions We use the third wave of the Survey of Women s Social Status in China (2010), which was conducted by the Women s Studies Institute of China with joint sponsorship from the All-China Women s Federation and the National Bureau of Statistics. The complete database includes 29,694 observations from all of China s 31 provinces, each of which contains information on a randomly selected adult and child within each household. We choose annual labor earnings as our outcome of interest, including in the sample all individuals aged between 26 and 55 years of age with non-zero earnings. This yields a

36 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 nationwide sample of 15,974 individuals. Individuals are divided into six age cohorts (from 26 30 to 51 55 years old) to enable a comparison across those age groups (albeit not across time), which is the best available option for a single-country cross-sectional dataset and has been used in a number of prominent papers, including Bourguignon et al. (2007) and Singh (2012). 7 Laws enacted in 1978 set the mandatory retirement age for Chinese men at 60 years, and for women at 55 years for public servants and 50 years for all others which remained in place in 2012 (although these have since been raised). The inclusion of this oldest cohort illustrates the impact of one particular circumstance beyond women s control their retirement age. Drawing on the inequality of opportunity and China-specific inequality literature cited above, we propose the following set of circumstance variables, which will be applied to each of the six age cohorts: 1 Gender: male or female (taking female as reference); 2 Father s education: illiterate, primary school, junior high school and above (taking illiterate as reference); 3 Father s occupation: agriculture (including forestry and fisheries), low-skilled non-agriculture and high-skilled non-agriculture (taking agriculture as reference); 4 Hukou status at birth: rural or urban (taking rural as reference); 8 5 Region: east, central and west (taking west as reference). 9 In combination, this implies that for each age cohort we are dealing with K = 108 types, in the comfortable range of 72 108 types according to Ferreira and Peragine (2015). For example, one type (a relatively unlucky one, as it turns out) comprises rural women born in Western China with illiterate fathers who work in agriculture. In the nationwide regressions for the entire sample, age cohorts are included as the sixth circumstance as one clearly cannot choose in what year they are born. Table 1 presents the preliminary statistics for these variables. As expected, labor income varies substantially across people with different circumstances in ways that are 7 The first and second waves of the Survey on Women s Social Status in China were completed in 1990 and 2000, respectively. If we are able to obtain these at a later stage, we will complement this work with analysis across two decades of reform. 8 The survey includes village, town, town-city and city rural is equated with village while the others are all classified as non-rural. 9 We use the standard regional classifications for our research. Ideally, we would use region of birth as this variable, but this is not available in the survey, so we use region identified at the time of the survey. While this is problematic in the sense that some people will have migrated since birth (a matter of choice, not circumstance), this only accounts for 9 percent of the surveyed individuals. We considered this a reasonable sacrifice given the significant regional variations in levels of development across China, and the standard practice of including region as a circumstance in the comparable literature.

Inequality of Opportunity in China s Labor Earnings 37 largely consistent with expectations: people born in urban areas in Eastern China with more educated fathers in non-agricultural occupations out-earn those in the relevant categories below them, and by substantial margins. Average earnings are highest for the youngest cohort, falling across the age range, consistent with higher average educational attainments among younger generations. One outlier in terms of expectations is that average earnings in the western region are higher than in the central underpinned by higher female (but not male) earnings in the west. Table 1. Preliminary Statistics for Earnings and Circumstances Nationwide Female Male Female/male income (ratio) Number of individuals 15,974 9797 8914 Mean labor income (yuan) 19,696 15,241 23,730 0.64 Income Share Income Share Income Share Father s education Illiterate 12,877 27 8658 26 16,374 27 0.53 Primary school 18,785 39 12,922 39 23,975 40 0.54 Junior high and above 26,090 34 22,435 36 29,732 33 0.75 Father s occupation Agriculture, forestry and fishery 15,532 65 11,481 66 19,223 65 0.60 Low-skill non-agriculture 26,195 24 21,647 23 30,113 24 0.72 High-skill non-agriculture 30,760 11 23,864 11 37,407 10 0.64 Born in Rural 16,843 71 12,220 71 21,070 71 0.58 Urban 26,512 29 22,583 29 29,992 29 0.75 Birth region East 24,089 47 18,083 45 29,164 48 0.62 Centre 15,149 25 11,056 26 18,994 27 0.58 West 16,432 28 14,550 29 18,277 25 0.80 Age cohort 26 30 27,477 12 23,660 12 31,162 11 0.76 31 35 22,128 14 16,557 14 27,984 13 0.59 36 40 18,949 21 15,456 21 22,358 21 0.69 41 45 18,589 20 14,301 20 22,777 20 0.63 46 50 17,703 18 12,547 18 22,108 19 0.57 51 55 15,714 16 9477 15 19,572 17 0.48 Sources: Survey of Women s Social Status in China (2010) and authors calculation. Notes: Sample includes all surveyed individuals with non-zero labor income. Non-agricultural low-skill occupations include craft and related trade workers, service and sales workers and clerical staff; nonagricultural high-skill occupations include professionals and managers; rural is equated with village, urban with all other classifications town, town-city and city. The most notable findings are the striking gender differences in the raw data, summarized by the female/male earnings ratios in the final column of Table 1. These reveal that gender inequality exists for every circumstance, and at both ends of the socioeconomic spectrum although the ratios are noticeably lower at the lower end. Some of these disparities may

38 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 reflect the fact that women work fewer hours which, given different retirement ages, is certainly the case in the oldest cohort with the lowest female/male earnings ratio of just 0.48. Another explanation is the different occupational structure for men and women for example, 40 percent of women work in agriculture, compared to 28 percent of men (calculated based on the sample used in our study). But this is not the whole story. Table 2 presents the baseline results for the nationwide sample, and for the female and male subsamples, based on Equation (3). As seen in column (1) in Table 2, being male is associated with labor earnings that are 0.54 log points higher than for females. 10 All coefficients take on their expected signs and relative magnitudes: for example, they increase with father s education and occupation, with urban hukou status and from west to east. The substantial gender differences seen in Table 1 are confirmed in the gender-specific regressions in columns (2) and (3) in Table 2. The adjusted R-squared for the female subsample is considerably higher than for men providing the first indication that women s earnings are more affected (i.e. constrained) by their circumstances than men. This is also suggested by the higher magnitudes of most coefficients in the female regressions. Table 2. Circumstantial Determinants of Labor Earnings Independent variable Nationwide Female Male Circumstance variable (1) (2) (3) Male 0.54*** Father primary 0.22*** 0.24*** 0.21*** Father junior high and above 0.34*** 0.37*** 0.30*** Father low-skill non-agriculture 0.30*** 0.36*** 0.24*** Father high-skill non-agriculture 0.41*** 0.43*** 0.39*** Urban 0.35*** 0.47*** 0.23*** Central region 0.09*** 0.03*** 0.19*** Eastern region 0.42*** 0.31*** 0.52*** Age 31 35 0.065* 0.039 0.079* Age 36 40 0.016 0.040 0.0015 Age 41 45 0.064* 0.089* 0.051 Age 46 50 0.12*** 0.19*** 0.069 Age 51 55 0.34*** 0.44*** 0.25*** Constant 8.9*** 8.7*** 9.4*** Observations 15,974 7592 8382 Adjusted R 2 0.248 0.234 0.174 Sources: Survey of Women s Social Status in China (2010) and authors calculation. Note: * p < 0.05, ** p < 0.01, ***p < 0.005. 10 For the log-linear form, the difference between the excluded dummy, x (e.g. male) and the included one, y, (e.g. female) is given by: lnx lny = ln(x/y). If x differs from y by a factor of 1+e, then ln(x/y) = ln(1+e), which is approximately e, the percentage difference between x and y, but only when e is small. The exact percentage change is given by exp(e 1), which for the example given here amounts to 75 percent.

Inequality of Opportunity in China s Labor Earnings 39 Table 3 presents the baseline results for each age cohort for the nationwide sample and the female and male subsamples. Panel A confirms the significance of being male for all cohorts, with higher coefficients for older cohorts. Panels B and C confirm substantial gender differences for all age cohorts, with the vast majority of coefficients being higher for women than men. The youngest cohort is particularly striking, with coefficients of 0.34 and 0.63 for women and 0.13 and 0.29 for men whose fathers have primary school education (compared to being illiterate) and work in low-skill non-agriculture (compared to working in agriculture), respectively. The adjusted R-squared values are consistently higher for the female cohorts than for their male counterparts, confirming that this select set of circumstances explains more of female earnings than it does of male earnings, across all age groups. Table 3. Circumstantial Determinants of Labor Earnings by Age Cohort Cohort 26 30 31 35 36 40 41 45 46 50 51 55 Independent variable Panel A: Nationwide Male 0.42*** 0.49*** 0.51*** 0.51*** 0.61*** 0.69*** Father primary 0.21** 0.28*** 0.24*** 0.24*** 0.18*** 0.22*** Father junior high and above 0.33*** 0.45*** 0.40*** 0.40*** 0.29*** 0.21** Father low-skill non-agriculture 0.46*** 0.35*** 0.28*** 0.28*** 0.16** 0.27*** Father high-skill non-agriculture 0.61*** 0.44*** 0.46*** 0.46*** 0.34*** 0.34*** Urban 0.25*** 0.30*** 0.28*** 0.28*** 0.36*** 0.58*** Central region 0.04*** 0.11*** 0.10*** 0.10*** 0.10*** 0.09*** Eastern region 0.45*** 0.48*** 0.40*** 0.40*** 0.45*** 0.37*** Constant 8.97*** 8.96*** 8.89*** 8.89*** 8.83*** 8.43*** Observations 1889 2336 3460 3460 2839 2125 Adjusted R 2 0.249 0.261 0.231 0.231 0.232 0.279 Panel B: Female Father primary 0.34** 0.27*** 0.33*** 0.20*** 0.18** 0.26*** Father junior high and above 0.49*** 0.52*** 0.47*** 0.26*** 0.30*** 0.23 Father low-skill non-agriculture 0.63*** 0.38*** 0.33*** 0.34*** 0.19* 0.31** Father high-skill non-agriculture 0.65*** 0.46*** 0.50*** 0.27** 0.45*** 0.29 Urban 0.39*** 0.36*** 0.40*** 0.50*** 0.54*** 0.91*** Central region 0.15*** 0.04*** 0.08*** 0.14*** 0.03*** 0.01* Eastern region 0.27*** 0.45*** 0.36*** 0.21*** 0.33*** 0.18* Constant 8.68*** 8.89*** 8.76*** 8.80*** 8.73*** 8.29*** Observations 928 1192 1709 1643 1308 812 Adjusted R 2 0.261 0.244 0.217 0.166 0.18 0.206 Panel C: Male Father primary 0.13 0.29*** 0.16** 0.27*** 0.18** 0.21*** Father junior high and above 0.21* 0.38*** 0.32*** 0.31*** 0.28*** 0.22* Father low-skill non-agriculture 0.29*** 0.31*** 0.21*** 0.20** 0.15* 0.27** Father high-skill non-agriculture 0.54*** 0.42*** 0.42*** 0.38*** 0.26** 0.35** Urban 0.14 0.23*** 0.16** 0.27*** 0.21*** 0.43*** Central region 0.26*** 0.20*** 0.12*** 0.30*** 0.22*** 0.17*** Eastern region 0.63*** 0.53*** 0.43*** 0.53*** 0.56*** 0.49*** Constant 9.63*** 9.51*** 9.52*** 9.41*** 9.52*** 9.20*** Observations 961 1144 1751 1682 1531 1313 Adjusted R 2 0.188 0.191 0.144 0.139 0.135 0.183 Sources: Survey of Women s Social Status in China (2010) and authors calculation. Note: *p < 0.05, **p < 0.01, ***p < 0.005.

40 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 2. Inequality of Opportunity in China s Individual Labor Earnings Table 4 presents the measures of inequality of opportunity in China s labor earnings based on the method laid out in Section II, using Equation (3). The top two rows of each panel provide scalar measures of total inequality in labor earnings using GE(0) and the Gini coefficient, for the entire sample (All) and for each cohort. Gini coefficients are included to stress the point that inequality across this sample is undeniably high: above 0.50 in all but two of the age cohort subsamples. Panels B and C reveal that inequality is higher within the female subsample in all cohorts, except for the 31 35 year-old cohort, regardless of the inequality measure used. The IOR provides the best gauge in terms of international comparisons to assert the point that the share of inequality of opportunity in total inequality in China s individual labor earnings is also undeniably high: with an IOR value of 0.25 for the nationwide sample, and with all cohort-level IORs above 0.20, with just one exception (in Panel A). These nationwide IORs are substantially higher than the male-only IORs in Panel C, reflecting that gender itself is a significant determinant of inequality of opportunity. Furthermore, the female subsample IORs are substantially higher than male IORs in all age cohorts, peaking at 0.29 for 31 35 year-old women (Panel B in Table 4), contrasting most sharply with an IOR of just 0.14 for 46 50 year-old men (Panel C in Table 4). Table 4. Inequality of Opportunity in China s Individual Labor Earnings All By birth cohort 26 30 31 35 36 40 41 45 46 50 51 55 Panel A: Nationwide Total inequality Mean log deviation (GE(0)) 0.55 0.70 0.50 0.46 0.52 0.51 0.62 Gini 0.53 0.59 0.51 0.55 0.52 0.51 0.56 Inequality of opportunity Absolute (IOA) 0.14 0.14 0.14 0.12 0.07 0.13 0.13 Relative (IOR) 0.25 0.20 0.28 0.26 0.14 0.25 0.21 Panel B: Female Total inequality Mean log deviation (GE(0)) 0.57 0.78 0.45 0.51 0.50 0.51 0.60 Gini 0.54 0.62 0.47 0.52 0.51 0.52 0.56 Inequality of opportunity Absolute (IOA) 0.14 0.16 0.13 0.12 0.10 0.11 0.16 Relative (IOR) 0.25 0.21 0.29 0.24 0.19 0.22 0.27 Panel C: Male Total (outcome) inequality Mean log deviation (GE(1)) 0.48 0.60 0.49 0.38 0.48 0.43 0.54 Gini 0.50 0.56 0.51 0.44 0.50 0.48 0.53 Inequality of opportunity Absolute (IOA) 0.09 0.09 0.09 0.06 0.07 0.06 0.10 Relative (IOR) 0.19 0.15 0.18 0.16 0.15 0.14 0.19 Sources: Survey of Women s Social Status in China (2010) and authors calculation. Notes: GE, generalized entropy; IOA, absolute inequality of opportunity; IOR, relative inequality of opportunity.

Inequality of Opportunity in China s Labor Earnings 41 Table 5 presents the Shapley-value decompositions of IOA for the nationwide sample and for each birth cohort (noting that these decompositions are identical for IOR). This reveals that gender is the single largest circumstance contributing to inequality of opportunity in China s labor earnings, accounting for 28 percent of the nationwide IOA. This is followed by one s region, father s occupation, father s education, birth cohort, and being rural or urban, in that order. In four of the six birth cohorts, gender is the largest contributor to inequality of opportunity, peaking at 38 percent of the IOA for the 41 45 year-old cohort. It ranks second for the two remaining cohorts (age 26 30 and 36 40), with father s occupation being the largest contributor in both cases. Table 5. Shapley-value Decomposition of Individual Circumstances (Contribution to Inequality of Opportunity, %) All By birth cohort 26 30 31 35 36 40 41 45 46 50 51 55 Gender 27.6 30.0 34.2 22.6 38.0 37.9 32.4 Father s education 17.7 10.0 14.4 20.0 13.5 11.9 12.9 Father s occupation 18.0 31.3 22.7 22.8 17.7 13.1 16.0 Born rural or urban 8.7 1.9 7.3 13.0 18.3 7.2 17.5 Region 18.4 24.9 21.5 21.7 12.5 29.9 21.2 Age cohort 9.7 Gender share in total inequality, GE(0) 7.0 6.0 9.6 5.9 5.3 9.7 6.8 Sources: Survey of Women s Social Status in China (2010) and authors calculation. Note: GE, generalized entropy. For comparison with other countries, the bottom row of Table 5 presents the share of gender in total inequality (as opposed to its share in inequality of opportunity). At 7 percent for the nationwide sample, the contribution of gender to China s earnings inequality is the highest of any share found in the international literature to date. While these results are not directly comparable because of the different datasets and circumstance variables included in each paper (e.g. De Barros et al., 2009; Ferreira and Gignoux, 2011; Martinez et al., 2017), it still seems reasonable to claim that China has a serious inequality of opportunity problem, with gender playing a very significant role. IV. Circumstances and Efforts: A Closer Look We now return to the method laid out in Section II based on Equations (1) and (2), to explore the relationship between circumstance and personal effort in more detail.

42 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 Drawing on Bourguignon et al. (2007), and our understanding of China s context, we propose the following five variables for personal effort: 1 Own education: a dummy variable for those who have attained junior high or below, or senior high school and above (taking junior high or below as reference); 2 Own occupation: a dummy variable representing those who work in the agricultural or non-agricultural sector (taking agricultural sector as reference); 3 Migration: a dummy variable for those who have ever worked or ran a business in a town or city different from where your hukou is for more than half a year? or who have never migrated (taking never migrated as reference); 4 Communist Party membership: a dummy variable for yes or no (taking not a party member as reference); 5 Marital status: married or not married, with the latter including people who have never married, are divorced or whose spouse has passed away (taking married as reference). Own education is included to reflect the fact that schooling above junior high has never been compulsory in China, and to some extent reflects an individual s choice, or effort (although, as it turns out, a substantial part of that choice is in fact explained by one s circumstances and most obviously, one s gender). Likewise, it is also a matter of choice to some extent at least to migrate and/or work in occupations outside of the agricultural sector. Communist Party membership is included to reflect the effort required to gain such membership, and because the earnings returns as a result of being a Party member have been shown to differ along gender lines (Shu and Bian, 2003; Appleton et al., 2009). 11 The inclusion of marital status is in recognition of the fact that couples have interdependent preferences that affect their household income decisions in ways that often imply different employment choices for married men and women, and therefore notably different earning outcomes (see Zhang Y. P. et al., 2008; Cook and Dong, 2011). 12 Table 6 presents the summary statistics for these five variables for the nationwide sample and the female and male subsamples, all of which are consistent with expectations: higher levels of education, non-agricultural occupations, Communist Party 11 Shu and Bian (2003) showed that Communist Party membership is associated with earnings that were 6 and 10 percent higher for men and women, respectively, in 1995, while Appleton et al. (2009) revealed that wage rewards for Party membership have in fact increased over time. 12 Cook and Dong (2011) attribute the decline in female labor force participation rates in China to the intensified pressure on women arising from their dual responsibilities as (unpaid) family carers and income earners during the economic transition. Zhang Y. P. et al. (2008) concur on this point, demonstrating that the observed gender earnings gap is strongly related to family status, with married women and mothers facing the most significant disadvantages in the (urban) labor market.

Inequality of Opportunity in China s Labor Earnings 43 membership and non-migration are associated with higher earnings (because almost all migrants are from rural areas and have significantly lower earnings on average than their non-migrating urban counterparts). Notably, marriage is associated with higher average earnings for men, but lower earnings for women. Table 6. Preliminary Statistics for Personal Effort Variables Nationwide Income Share Female Income Share Male Female/male income Income Share (Ratio) Own education Junior high or below 13,531 60 9372 63 17,662 58 0.53 Senior high school and above 29,029 40 25,296 37 31,966 42 0.79 Own occupation Agricultural, forestry and fishery 8566 36 6327 40 11,055 33 0.57 Non-agricultural 26,011 64 21,210 60 29,879 67 0.71 Political party Not a Communist Party member 18,126 85 14,107 89 22,117 81 0.64 Communist Party member 28,577 15 24,506 11 30,722 19 0.80 Migration status Ever migrated 18,387 9 10,779 8 24,161 9 0.45 Never migrated 19,818 91 15,618 92 23,686 91 0.66 Marital status Married 19,543 91 14,867 92 23,828 91 0.62 Not married 21,235 9 19,294 8 22,800 9 0.85 Sources: Survey of Women s Social Status in China (2010) and authors calculation. Table 7 presents the regression results based on Equation (1), which add these effort variables to the reduced-form regressions presented in Table 2. This yields a considerable increase in the adjusted R-squared values, with the bulk of this increase coming from own occupation and own education, and only minor increases for Communist Party membership, migration and marriage. 13 Working in a non-agricultural occupation is associated with earnings that are 0.76 log points higher than working in agriculture, while having senior high school education or above is associated with a 13 In regressions adding these effort variables separately to the nationwide sample, the adjusted R-squared increases from 0.248 (column (1) in Table 2) to 0.367 for own occupation, 0.310 for own education, 0.272 for Communist Party membership, 0.250 for marriage and negligibly for migration. These results are not presented in separate tables due to space limits and are available from authors upon request.

44 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 0.38 log point income boost. The coefficients on Communist Party membership and migration are positive and significant, but small. Table 7. Determinants of Labor Earnings: Circumstances and Efforts Independent variable Nationwide Female Male Circumstance variable (1) (2) (3) Male 0.44*** Father primary 0.11*** 0.11*** 0.12*** Father junior high and above 0.14*** 0.15*** 0.13*** Father low-skill non-agriculture 0.035 0.057* 0.017 Father high-skill non-agriculture 0.13*** 0.13*** 0.14*** Urban 0.058** 0.10*** 0.016 Central region 0.02*** 0.07*** 0.12*** Eastern region 0.29*** 0.18*** 0.40*** Age 31 35 0.043 0.028 0.045 Age 36 40 0.013 0.020 0.021 Age 41 45 0.028 0.026 0.047 Age 46 50 0.081** 0.089* 0.085* Age 51 55 0.23*** 0.25*** 0.22*** Effort variable Non-agricultural occupation 0.76*** 0.79*** 0.72*** Senior high and above 0.38*** 0.44*** 0.31*** Communist Party member 0.20*** 0.22*** 0.19*** Migration 0.092*** 0.023 0.14*** Married 0.16*** 0.065 0.26*** Constant 8.34*** 8.35*** 8.81*** Observations 15,974 7592 8382 Adjusted R 2 0.400 0.392 0.337 Sources: Survey of Women s Social Status in China (2010) and authors calculation. Note: *p < 0.05, **p < 0.01, ***p < 0.005. As expected, coefficients on many of the circumstances fall (compared to those in Table 2), an indication of the correlations between these and the two key effort variables, own education and occupation (although none too high to suggest that multicollinearity is a serious problem). Being male continues to have the highest positive association with earnings for the nationwide sample. As seen in columns (2) and (3) in Table 7, working outside agriculture, attaining senior high school education and having Communist Party membership are associated with higher returns for women than men, while migration and marriage are insignificant for women but are associated with higher earnings for men. The key point is that these effort variables matter, and they matter in different ways for each of the genders. Finally, Table 8 presents the results of probit regressions for each of the binary effort variables conditioned on the set of circumstance variables, as in Equation (2). Nearly all coefficients take on their expected signs and relative magnitudes. Most importantly,

Inequality of Opportunity in China s Labor Earnings 45 the pseudo R-squared values are highest for the own-education and own-occupation regressions, and both of these are higher for the female subsample than the male subsample underpinned by larger magnitudes on almost all coefficients in the female subsample regressions. This adds further weight to our finding that circumstances matter more for women: not only directly but also indirectly through their effect on two prominent and identifiable forms of effort (own education and own occupation). Circumstances Table 8. Probit Regressions of Efforts Determined by Circumstances Own education Own occupation Communist Party member Migration Marriage Panel A: Nationwide Male 0.20*** 0.28*** 0.37*** 0.14*** 0.063* Father primary 0.33*** 0.20*** 0.24*** 0.013 0.16*** Father junior high and above 0.75*** 0.35*** 0.48*** 0.11* 0.15*** Father low-skill non-agriculture 0.49*** 1.15*** 0.12*** 0.25*** 0.14*** Father high-skill non-agriculture 0.71*** 0.78*** 0.30*** 0.22*** 0.068 Urban 0.74*** 1.25*** 0.20*** 0.87*** 0.38*** Central region 0.11** 0.21*** 0.11 0.05*** 0.199 Eastern region 0.19*** 0.49*** 0.091** 0.16*** 0.13*** Observations 15,974 15,974 15,974 15,974 15,974 Pseudo R 2 0.214 0.280 0.059 0.097 0.062 Panel B: Female Father primary 0.32*** 0.28*** 0.27*** 0.071 0.082 Father junior high and above 0.77*** 0.44*** 0.52*** 0.064 0.026 Father low-skill non-agriculture 0.63*** 1.07*** 0.22*** 0.27*** 0.21*** Father high-skill non-agriculture 0.75*** 0.76*** 0.30*** 0.29** 0.099 Urban 0.87*** 1.30*** 0.34*** 0.81*** 0.48*** Central region 0.05** 0.15*** 0.027 0.13*** 0.16** Eastern region 0.16*** 0.53*** 0.082 0.16** 0.0046 Observations 7592 7592 7592 7592 7592 Pseudo R 2 0.269 0.303 0.072 0.099 0.071 Panel C: Male Father primary 0.35*** 0.13*** 0.24*** 0.030 0.21*** Father junior high and above 0.73*** 0.27*** 0.46*** 0.14* 0.25*** Father low-skill non-agriculture 0.38*** 1.25*** 0.052 0.23*** 0.087 Father high-skill non-agriculture 0.68*** 0.81*** 0.31*** 0.17 0.030 Urban 0.62*** 1.20*** 0.093* 0.91*** 0.33*** Central region 0.15 0.27*** 0.17 0.03** 0.227 Eastern region 0.21*** 0.46*** 0.099* 0.17*** 0.24*** Observations 8382 8382 8382 8382 8382 Pseudo R 2 0.172 0.256 0.037 0.097 0.075 Sources: Survey of Women s Social Status in China (2010) and authors calculation. Note: ***p < 0.01, **p < 0.05, *p < 0.1.

46 Jane Golley et al. / 28 50, Vol. 27, No. 1, 2019 V. Conclusions This paper set out to explore the various factors that explain total inequality of individual labor earnings in China, distinguishing between circumstances that are essentially determined at birth from the efforts or choices that people make over the course of their childhood and adult lives. Measurements of the contribution of circumstances using the predominant empirical approach revealed an alarmingly high relative share of inequality of opportunity in nationwide individual labor earnings of 25 percent: one of the highest shares found in the existing international literature to date. Even more alarming was the fact that gender topped the list of circumstances that shape economic advantage above father s occupation and education, rural or urban status and region of birth. Furthermore, we showed that inequality of opportunity is higher among Chinese women than among Chinese men, across all age cohorts. Simply put, circumstances matter more for women in determining their position along the labor earnings distribution, and not in positive ways, on average at least. Importantly, we have not attributed causality to any of these circumstances, and we concede that biases exist in possibly all of them, because of omitted variables the most obvious of which is IQ. However, in the absence of any convincing evidence that men and women have different IQs, or that IQ is transmitted differently from father to son than from father to daughter, we argue that the substantial gender differences observed in the analysis here confirm substantial differences in the opportunities available to Chinese men and women. This point was compounded by the final part of our analysis, which identified five key effort variables that were in turn determined by circumstances, with one s own education and occupational choices being the most prominent. The collective role of circumstances (other than gender) in shaping efforts (or choices) was more important in the female than male subsamples, further confirming significant biases in the earning opportunities for Chinese men and women. These results provide some basis for recommending ways in which gender equality of opportunity could be improved in China in the future. The fact that women s earnings and education, as well as occupational choices, are impacted more than men s by their father s education and occupational status suggests that it would be eminently reasonable for educational policies to specifically target girls from poor, rural families as a starting point for expanding their earning opportunities. Measures to assist young women exiting agricultural employment would also likely improve their earning potential in the future. Ongoing reforms to the hukou system to ensure that rural migrants are not discriminated against in urban areas in terms of their access to jobs and social welfare and the pay they receive would also go some way toward equalizing the opportunities they face compared