The Growth of Wage Inequality in Urban China, 1988 to 1999



Rising Returns to Skill, Labor Market Transition,

and the Growth of Wage Inequality in China*

Albert Park, University of Michigan

Xiaoqing Song, Beijing University

Junsen Zhang, Chinese University of Hong Kong

Yaohui Zhao, Beijing University

January 2008

Abstract

Using annual urban household survey data from 6 provinces in different regions of China, we analyze the rapid increase in inequality of China’s urban wages from 1988 to 2001. We describe within-group and between-group inequality trends and decompose inequality change into changes in the distribution of worker attributes, both observed and unobserved, and changes in the returns to those attributes. We find that growing regional wage disparities, rising returns to unobserved skills, and rising returns to education explain most of the inequality increase. Rising skill premiums are due primarily to increasing relative demand for skilled workers within industries rather than to demand shifts across industries. The timing of inequality change has been erratic, suggesting a key role played by reform policies.

*The authors acknowledge financial support from the Chinese University of Hong Kong (Direct Grant), the Research Grant Council of Hong Kong (N_CUHK417/01), National Natural Science Foundation of China, and a visiting research fellowship awarded to Yaohui Zhao by the World Bank. We thank participants of the 5th NBER-CCER conference on the Economy of China (July 2002) for valuable comments.

1. Introduction

Since economic reforms began in 1978, China has experienced one of the fastest increases in income inequality ever recorded.[1] Other transition economies also saw rapid inequality growth, but none have reached China’s level of inequality.[2] China’s largely successful rapid economic and social transformation makes China’s increasing inequality of particular interest, especially given the tension between widening disparities and the government’s continued espousal of a socialist ideology. However, in contrast to the U.S. where growing wage inequality has been studied in great detail, there has been surprisingly little analysis of rising inequality in China using detailed micro-level data over time.[3] In light of the fact that China accounts for roughly 20 percent of the world’s population, understanding the nature and causes of China’s recent growth in inequality is critical for understanding changes in world inequality. It also provides an important window into the inequality dynamics of transition and development.

Inequality trends in China reflect fundamental changes in the way labor is allocated and rewarded in China’s transitional economy. China has moved from a socialist planned economy with fixed wage scales and virtually no labor mobility to a market-based system featuring a dynamic non-state sector and an increasingly open labor market. We show in this paper that one key aspect of this transition has been a rapid increase in the returns to skill. This is to be expected if socialist planners compressed wage scales while market reforms have increasingly rewarded labor based on productivity. Despite these changes, important institutional barriers to labor mobility remain in China. Thus, it may be inappropriate to think of observed wages as being purely market-determined. China’s residential permit system continues to inhibit inter-region\al migration. Many state sector workers have been reluctant to switch jobs because the provision of non-wage benefits such as housing still tie them to their employers. Such forms of labor market segmentation have direct consequences for both inequality and the returns to skill.

In this paper, we analyze the changes in wage inequality that have occurred in China’s urban areas from 1988 to 2001 using annual household survey data from six provinces collected by China’s National Statistics Bureau. We focus on wages rather than earnings in order to better evaluate the performance of the formal labor market. Despite some limitations which we discuss below, the large-sample repeated cross-sectional data make it possible to go considerably beyond existing studies in analyzing the sources and timing of wage inequality changes in post-reform China.

The overall change in the distribution of wage income is illustrated in Figure 1, which plots kernel density estimates for the years 1988, 1992, 1994, 1997, and 1999. It is clear from this figure that mean incomes have increased steadily, while the distribution around the means has widened considerably. The most noticeable change in both the mean and variance of the distribution occurred between 1992 and 1994. Table 1 reports summary wage inequality measures by year for our dataset (described in greater detail below). By any measure, wage inequality increased substantially from 1988 to 1999. The gini coefficient increased from 0.244 to 0.359, the coefficient of variation from 0.486 to 0.752, the standard deviation of log wages from 0.456 to 0.658, and the Theil entropy index from 0.102 to 0.221.

Our goal in this paper is to describe and evaluate the sources of growing wage inequality in urban China. In particular, we focus on the importance of rising returns to skill and China’s transitional labor market institutions. We examine the returns to skill by looking at the returns to education and experience and the variation of wage regression residuals. The two institutional factors that we examine are ownership categories (state versus non-state) and inter-regional labor market segmentation. In a final section, we provide initial evidence on the relative importance of different supply and demand factors in explaining rising returns to education—a key dimension of skill differences.

[statement of main findings]

In section 2, we first introduce the labor market reforms and broader economic changes likely to affect wage inequality. After describing the data in section 3, in section 4 we first document the overall trends in the income distribution from 1988 to 1999. We then examine trends in within-group and between-group inequality, defining groups by education, experience, region, ownership, and sector. In section 5, we examine the importance to inequality of trends in observable characteristics of the labor force (quantity changes), changes in the returns to observable characteristics such as education and experience (price changes), and changes in the distribution of unobservable characteristics (residual changes). This leads us to a formal decomposition analysis of the relative importance of these different factors. In section 6, we conduct additional tests to examine the effect of supply and demand factors on rising returns to education. Section 7 concludes.

2. Economic reform and the labor market

In transition economies, the reform of labor market institutions has a profound effect on wage and inequality outcomes. Market-based reforms of the labor market are often implemented in combination with other major reforms that promote open product market competition, greater managerial autonomy, and restructuring of publicly owned enterprises. China introduced markets for outputs and inputs to most of the industrial sector in the mid-1980s, but nearly all urban jobs continued to be allocated by government labor bureaus until at least the early 1990s and in many areas until the late 1990s. Decentralization reforms in the mid-1980s gave greater decision-making autonomy and pay incentives to enterprise managers and allowed state firms to hire workers on a contract basis rather than provide permanent employment and provide a higher share of compensation in the form of bonuses (Groves et al., 1995). However, even by the early 1990s, state-owned firms were frequently prohibited from firing contract workers, and employment decisions within state-owned firms showed little responsiveness to changing market conditions (Benjamin, Brandt, and Yuen, 2001).

Substantial liberalization of the economy occurred during the rapid growth episode that followed Deng’s southern trip in 1992. During this period, many workers left state employment to jump into the ocean of the free market (xiahai) and many state-owned units expanded into a range of commercial activities. Finally, beginning in 1997, the government moved forward with aggressive restructuring and privatization of state-owned enterprises, leading to substantial layoffs, retirements, and exits from the labor force (Giles, Park, and Cai, 2003). From 1997 to 2001, over 45 million laborers left state sector employment.

Overall, these reforms have made China’s system of labor allocation increasingly market-oriented, which helps explain rising returns to skill and growing inequality. However, the reforms have been implemented in a gradual, halting fashion. And, as noted earlier there remain substantial barriers to labor mobility which may affect inequality outcomes. First, urban residents of one city cannot easily obtain permanent resident status in other cities, which is necessary for gaining access to public services such as education. In this case, it is the lack of market reform (to promote inter-regional labor mobility) that increases inequality by segmenting labor markets. Secondly, mobility between state and non-state sectors may be limited by hidden subsidies provided to workers by state enterprises and the slow progress of reforms intended to shift nonwage benefit provision (housing, pensions, health care) from employers to local governments or the market. If the non-state sector is more market-oriented, than such barriers to mobility could slow the overall development of the labor market. On the other hand, aggressive reforms within the state sector, especially in the late 1990s, could lesson the importance of state versus non-state employment.

In addition to labor market reforms, skill premiums in the labor market also depend on underlying supply and demand factors. In the U.S., explanations for rising wage inequality have focused on three factors: skill-biased technical change, international trade, and labor market institutions, especially unions (Katz and Autor, 1999). Evidence that rising inequality is associated with growing returns to skill, both observed and unobserved, even within relatively narrowly defined sub-sectors of the economy have led some to conclude that skill-biased technical change is the main culprit (Juhn, Murphy, and Pierce, 1993; Bound and Johnson, 1992).

China, like the U.S. in recent years, has seen rising skill premiums despite the fact that the workforce has become more educated over time. Although labor market reforms have been of first order importance, China has also witnessed other dramatic, transformative changes that have influenced labor demand. New technologies have been introduced rapidly, with foreign direct investment in China accounting for a substantial share of fixed investment capital [add numbers]. During the 1990s, China saw a dramatic increase in trade’s share of GDP, which would be expected to reduce skill premiums if China’s exports are intensive in low-skilled labor [add numbers]. In the final section of the paper, we provide the first analysis of which we are aware of the importance of different demand and supply factors in explaining growing skill premiums.

3. Data

Our data come from annual urban household surveys (UHS) conducted by the National Bureau of Statistics (NBS) from 1988 to 1999. The NBS urban sample frame includes urban permanent resident households in all urban areas, including cities of all sizes, and is designed to be representative at the provincial and national levels. The data set includes all NSB survey households from six provinces: Beijing, Guangdong, Liaoning, Shaanxi, Sichuan, and Zhejiang.[4] These six provinces are roughly representative of China’s different regions. Beijing is in North-Central China, Guangdong and Zhejiang are coastal provinces, Liaoning is in the Northeast, Shaanxi is in the Northwest, and Sichuan is in the Southwest.

Wages are defined to include base wages, bonuses, and subsidies, and to exclude capital and transfer income. The income and expenditure data are based on self-recorded diaries reported monthly, and so are likely to be more accurate than recall surveys. However, the NBS only includes annual wages in the data files and working hours are not reported. This precludes the possibility of constructing hourly wage rates. To reduce bias from variation in labor hours worked, we exclude individuals who are likely to be part-time workers, including students, the disabled, re-employed retired workers, workers younger than 16 and older than 60,[5] and self-employed workers. We further exclude all workers earning less than half of the minimum wage.[6] For prime-age adults in urban China, full-time work is the dominant form of employment and self-employment is relatively uncommon. Applying these criteria yields a sample of 80,312 workers over the 12 years. Table 2 describes the sample distribution and Table 3 presents descriptive statistics. We deflate incomes using provincial urban CPIs, with all incomes reported in 1988 yuan. Where appropriate, we weight the sample based on the sampling rate for each province, i.e., sample size divided by urban labor force, and by the number of working-age adults in the household, to correct for bias from household rather than individual sampling.

The data have several limitations which should be kept in mind in interpreting the results. First, the NBS urban survey is restricted to households that have urban residence permits (hukou), and so does not include migrants working in cities.[7] The survey also excludes workers residing in rural areas who are engaged in wage employment. This sampling approach results from China’s unique administrative separation of urban and rural residents. NBS uses separate sampling frames and questionnaires for their urban and rural surveys. Nonetheless, since most wage employment is in urban areas and most wage workers are urban residents, the data should accurately capture main trends in wage inequality. Strictly speaking, however, the results apply only to China’s registered urban residents.

Another limitation is that we do not have data before 1988, even though China’s economic reforms began in 1978. NSB did not rejuvenate its national survey apparatus until the mid-1980s and 1988 is the first year for which the sample and questionnaire data were comparable to later years. We note that nearly all of the major changes to China’s labor allocation system occurred after 1988, as did many other changes likely to influence retursn to skill (e.g., trade and FDI growth). Thus, the data should capture the main period of wage inequality increase.

4. Trends in wage inequality

4.1. Overall trends

In this section, we first provide an overall description of wage inequality. We then divide the sample into groups and examine changes in within-group and between-group wage inequality. Figure 2 describes the wages from 1988 to 1999 for different parts of the skill distribution in 1988. Real annual wages at the 10th percentile, the median, and the 90th percentile all increased during the past 12 years but the rates of increase were different. The 90th percentile more than tripled, increasing from 2,864 yuan in 1988 to 8,697 yuan in 1999 (in 1988 yuan), while earnings at the median more than doubled, increasing from 1,704 yuan in 1988 to 3,431 yuan in 1999. Earnings at the 10th percentile increased only by a half, from 910 yuan in 1988 to 1,391 yuan in 1999.

Figure 3 plots the annual percent growth in real wages from 1988 to 1999 for each percentile of the wage distribution. All percentiles experienced wage gains during the period, but higher percentiles experienced larger wage gains. Thus, inequality has not been a story of the rich getting richer and the poor becoming poorer, but rather the rich getting richer faster than the poor. To put the magnitude of these differences in perspective, consider that Juhn et al. (1993) found that the difference in wage growth at the top and bottom of the U.S. wage distribution to be about 45 percent from 1964 to 1988, or less than two percent per year. Figure 3 shows that in China this annualized difference was about eight percent per year, or four times greater than in the U.S.

Figure 4 depicts the trend of wage inequality by reporting log wage differentials between the 90th and 10th percentiles, the 90th percentile and the median, the 75th percentile and the 25th percentile, and the median and the 10th percentile. Using the 90th and 10th percentiles as an example, wage differentials are calculated as [pic]. By all of these measures, the rise in wage inequality was substantial. The 90th-10th percentile log wage differential increased from 1.15 in 1988 to 1.83 in 1999. Comparing the 90-50 and 50-10 differentials, we find that at the beginning of the period, the lower half of the income distribution had more dispersion than the top half, but this reversed after 1991. The 50/10 differential rose briefly in 1994 and again after 1997, approaching the level of dispersion in the top half of the distribution. In China, 1994 was a year of relative slowdown and monetary tightening, and 1997 saw the beginning of major state-owned enterprise restructuring.

From Figure 4, we see that inequality rose most rapidly from 1992 to 1994 but that the rising trend continued until the end of the period for which we have data. In Figure 5, we break down the percentile wage growth reported in Figure 3 into sub-periods, and see again that by far the most rapid increase in inequality occurred during the 1992 to 1994 period. The first period, 1988 to 1992 also saw relatively greater gains at the top end of the wage distribution. The most recent period, 1997 to 1999, is characterized by noticeably smaller gains for the bottom part of the distribution, and a general trend of the richer getting richer, except at the very top end of the distribution.

Table 4 presents additional measures of wage inequality for 5 different years: 1988, 1992, 1994, 1997, and 1999. Measured by the standard deviation of log annual wages, we find continuous increases in inequality, especially from 1992 to 1994 (from 0.50 to 0.63), when nearly all wage percentile differentials widen sharply. For example, the log wage differences between the 90th and 10th percentiles jump from 1.33 to 1.73 in these two years, accounting for 58 percent of the total increase from 1.15 in 1988 to 1.83 in 1999. There is little change in many of the percentile differentials from 1994 to 1997, but a resumption of divergence across the board from 1997 to 1999.

4.2. Within and between group trends

Next, we move beyond overall wage inequality to examine how inequality has changed within groups and between groups of the population. There are, of course, many ways to divide the working population into groups. In this section, we define groups based on the variables that reflect skill and labor market institutions: experience, education, region, ownership, and sector. For each criteria, we look at within-group inequality by presenting plots of wage growth by percentile within each subgroup, and examine between-group inequality by plotting mean wages of different subgroups over time.

Experience. Because our data do not report actual work experience, we define experience as potential experience, calculated as age minus years of schooling minus 6. We group workers into four experience groups: 1-10 years, 11-20 years, 21-30 years, and 31-40 years. Figure 6A presents annual average growth rates of real wages from 1988 to 1999 for each wage percentile of each experience cohort. Each line is upward sloping, signifying that inequality rose within each experience cohort. Also, lines for younger cohorts lie above those for older cohorts, revealing that younger cohorts experienced larger wage increases than older cohorts at every percentile. The slope of youngest cohort line is less steep than the others, suggesting that income inequality grew less among younger workers. Wages grew rapidly even at the lowest percentiles of the youngest workers (about 5 percent per year). This is quite different from the experience of younger workers in the U.S. (Juhn et al., 1994).

Figure 6B plots the mean wages of different experience groups over time. While wages of all groups rise substantially over time, as a percentage of initial wage means the increase is much greater for the youngest cohort, a pattern which should narrow wage inequality if their remain positive returns to experience. In the latter 1990s, experience group differences for all groups but the youngest workers appear to disappear altogether, suggesting that returns to experience fell over time, a trend which should be equalizing.

Education. The trends in wage differences within and between educational groups are both increasing. The upward slopes of the lines in Figure 7A illustrates that within each educational group, inequality has increased tremendously. The steeper slopes of the lines for college graduates and special or technical school graduates indicate that inequality increased more among those with greater educational attainment. In particular, there is remarkably high wage growth at the top end of the wage distribution of highly educated workers (surpassing 14 percent per year) that does not appear for other groups.

Wage differences across education groups also have widened (Figure 7B), especially between college graduates and other groups. One anomaly from Figure 7B is that primary school graduates have higher wages than middle school graduates in many years, likely due to the strong negative correlation between education and experience. Rising educational premiums could explain the increase in within-experience group wage differences. It is unlikely that rising experience premiums can explain increases in within-education group differences since experience premiums appear to have fallen over time. Figures 8A and 8B examine wage changes by percentile within education-experience cohorts. For simplicity, we restrict attention to the interaction of two cohorts—the youngest experience cohort (1-10 years) and an older experience cohort (21-30 years), and two education groups--college and above, and high school. We find that inequality is increasing even within groups of individuals with similar education and experience (Figures 8A and 8B). Wages are growing particularly fast for those at the positive tail of young college graduates.

Region. Figure 10A plots the annual real wage growth rate for each province for different percentiles of the wage distribution. Again, all of the lines are upward sloping, indicating that inequality has increased within each province. However, the lines are not nearly as steeply-sloped as that for the whole sample (Figure 3), suggesting that regional differences in wage growth may play a prominent role in overall inequality growth. What is particularly striking in both Figures 10A and 10B is the strong evidence of large mean differences in growth rates across provinces, with Guangdong’s wage growth far greater than the other provinces at every percentile (Figure 10A), and Guangdong’s mean wages rising to more than twice that of the closest province, Zhejiang. Guangdong’s median real wage growth rate has been a remarkable 9.6% over the 11-year period. Zhejiang is the second fastest growth province, followed by Beijing, Sichuan, Liangning and Shaanxi (only 3.7 percent per year). This ranking of provinces by growth rate is nearly identical to ranking by initial level of mean wages, implying substantial wage divergence spatially (Figure 10B).

Ownership. Many people point to dismantling of the state sector as a key to freeing up the labor market, so it may be informative to look carefully at wage patterns by ownership type. Figure 11A shows, surprisingly, that wage dispersion within the state-owned sector increased as much, if not more, than inequality within other ownership types, contradicting accounts of the state sector as rigid and prone to wage compression. In the state sector, wages of the 10th percentile grew by 2.1%, while wages of the 90th percentile grew by 9.9%. However, the pattern for other (non-government) enterprises is very different. The poorest 25 percent of workers had higher wage growth than the rest of the distribution, other than the top 5 percent. This may in part reflect changes in who works in the non-government sector. The percentage of workers employed in the non-public sector increased from only 3.1% in 1992 to 11.4% in 1999 (Table 3). Figure 11A also shows that at every percentile of their respective wage distributions, the state-owned units gained over their collective counterparts. Figure 11B confirms the growing gap between wages of state and collective workers, and also shows that wages in non-government-owned firms started lower than state and collective firms but surged past them in the early 1990s before converging somewhat with SOE wages by 1999.

Greater wage dispersion within the state sector could reflect labor market development, with wage differences better reflecting worker productivity differences. Evidence of rising returns to education within the state sector lends support to this explanation (Zhang and Zhao, 2002). However, it could also reflect growing differences in the economic performance of state-owned enterprises over time in a more competitive market environment, which could increase wage dispersion of a random nature if workers are not mobile.

Sector. The relative importance of within-sector and between-sector changes in inequality are frequently examined to distinguish among different causal explanations for inequality increase. Figures 13A and 13B present wage data for the five largest employment sectors, out of 13 sectors coded in the data consistently from 1988 to 1999. Figure 13A shows that as for other groupings, inequality rises substantially within each sector. Although there are not extremely large differences in average growth rates of wages across industries, both Figures 13A and 13B reveal that wages in high-skill sectors (government, education, and transportation/communication) rise faster than wages in manufacturing and retail trade.

Unobserved skills. By definition, unobservable skills cannot be measured directly. Here we follow the methodology of Juhn et al. (1993) who define unobserved skills to be the residuals from a regression of the logarithm of wages on a flexible specification of education and experience terms.[8] To account for substantial regional labor market segmentation in China, we also include regional dummy variables and allow them to interact flexibly with education and experience. The goal is to extract all of the “observable” information from the wage data, so that the residual can more justifiably reflect unobserved skills.

Table 5 presents the standard deviations of the residuals as well as various measures of residual inequality over time. The standard deviation increased substantially, from 0.33 in 1988 to 0.47 in 1999. As for overall inequality, the largest growth in residual inquality also occurred between 1992 and 1994. This suggests that unobservable factors are increasingly important in determining wages. Given the rapid changes in the labor market, this is not surprising since market mechanisms will increasingly reward workers by their productivity rather than just their observable credentials. This should not only be reflected in wage setting within firms, but also in employment choices, since more able, entrepreneurial workers are more likely to find employers that reward such characteristics.

Within-cohort inequality. One interpretation of changes occurring over time measured using repeated cross-sectional data is that the changes reflect differences in unobserved cohort characteristics as the sample increasingly includes more individuals born in recent years and fewer individuals born in earlier years. If this were the case, cohort differences could obscure changes in underlying factors generating inequality. One way to check how important such cohort effects may be is to follow birth cohorts over time to see how inequality within such cohorts changes over time and to compare these trends to overall changes in inequality. In Table 7, we report changes in wage inequality within synthetic cohorts where inequality is represented by log wage differences between workers at 90th and 10th percentiles. We present inequality by birth cohorts at an interval of six years. One can follow a birth cohort by moving horizontally from left to right and follow an experience group by moving upward along a diagonal. Within a birth cohort, the increase in inequality over time is due to an age/experience effect and a time effect. Within an experience group, the change in inequality is due to a cohort effect and a time effect. Thus, by comparing changes along the horizontal line and the diagonal, we can eliminate the time effect and directly compare the cohort and age effects.

In Table 7, the changes along the diagonal and the horizontal lines are quite similar. As an example, for the 1968 to 1972 market entry cohort, the 90th-10th percentile log wage differential increased from 1.26 in 1989 to 1.93 in 1994 and 1.86 in 1999, and the differential for new market entrants increased from 1.26 in 1989 to 2.07 in 1994 and 1.76 in 1999. Patterns for other birth cohorts are similar. The bottom two rows of the table present the average change across all birth cohorts and experience groups. On average, the changes in inequality within birth cohorts between 1989 and 1994 was 0.65, and within all experience groups in the same period was 0.71. Between 1994 and 1999, within birth cohort change was 0.06 and within experience group change was 0.09. [discuss]

5. Accounting for rising wage inequality

A main message of Section 4 is that wage inequality has increased significantly overall and within nearly every sub-group of the population. Differences between groups defined by education and region have also grown substantially. In this section we introduce a multivariate framework to more clearly identify the contributions to growing inequality due to different observable and unobservable worker attributes. The effect of any given factor can be decomposed into two parts: quantity effects and price effects. The former arises because of changes in the distribution of attributes in the population of workers (e.g., greater inequality in educational attainment), the latter because of changes in the value attached to different characteristics, e.g., higher returns to education over time. We examine these two effects in turn and then decompose the overall changes in inequality into changes in observable and unobservable quantities and prices.

5.1 Quantity effects

How have characteristics of the labor force changed over time, and can these changes help explain growing inequality? Table 3 provides summary information on key sample characteristics. The average age (and potential experience) of the population has increased over time, from 36.8 to 39.4, which could be due changing demographics associated with China’s strict one-child family planning policy. Aging of the sample could also be a factor, although this is unlikely given NSB’s periodic rotation of sampled households. With fewer young workers who make relatively less than older workers and so are at the tail of the wage distribution, this change is likely to reduce inequality.

There is a very dramatic increase in the educational attainment of the workforce. The share of workers with college education or above increased from 12 percent in 1988 to 26.8 percent in 1999 while those with junior secondary and primary education fell from 41 and 11.8 percent to 26.2 and 3.7 percent over the same period. These changes shifted much more of the distribution to the higher end compared to previous periods when it was centered strongly around junior and senior high school education. This could widen inequality, especially if college premiums are high.

The gender and regional distribution changes relatively little over time and so are not expected to contribute substantially to inequality. The percentage of men increases very slightly from 51.1 to 52.7 percent over the 11 years. There is slightly faster growth in the labor force size of richer provinces than poorer provinces, which would increase inequality. Finally, in terms of job types, in our sample we see only a small reduction in the size of the state sector, a decline in the collective sector, and an increase in the non-public sector. Since the non-public sector, which includes joint ventures, generally has higher wages, this change should be in inequality-increasing. We defer discussion of occupation and sector changes to later.

5.2 Price effects

In addition to changes in quantities, changes in prices of skills can also influence inequality trends. For example, the returns to education in urban China have increased dramatically from 1988 to 1999 (Table 6).[9] Due to the positive correlation between wages and education, this is likely to raise inequality. In this section, we compute the prices of education and other attributes, defining the prices as coefficients from wage regressions estimated for each year. We estimate two specifications for log wages. The first includes educational dummy variables, potential experience, experience squared, a gender dummy, and regional dummy variables as regressors. The second adds job-related dummy variables for ownership, occupation, and sector. To control for selection bias, we estimate Heckman selection correction models, where labor force participation is identified by marital status and number of children.

Results from both specifications show clearly that the returns to education have risen rapidly, with the largest increases for college education and above. With primary school education and below as the reference group, the coefficients for junior high school declined over time, those for senior high school increased slightly, those for technical school increased by 20 percentage points over the 11 years, and those for college and above increased by 31 percentage points.[10] The coefficients, or prices, in each year are plotted in Figure 14A.

In contrast to education, returns to potential experience declined over time (also see Figure 14B). The marginal return to a year of experience (evaluated at sample means) declined from 2.0% in 1988 to 1.3% in 1999. Age-earnings profiles (not shown) reveal that for both men and women, 1999 age-earning profiles were much more concave than those in 1988, with the largest wage increases occurring in the early period of work life, and wages flattening out as early as age 30.[11]

Growing regional wage differences can be seen in Figure 14C, which plots coefficients of provincial dummy variables with Sichuan being the reference province. We can see that Guangdong is the richest province and leads other provinces in income growth. Guangdong is followed by Zhejiang and Beijing, which also gained relative to Sichuan in the 1990s. The northeastern province Liaoning stagnated relative to Sichuan, while the northwestern inland province Shaanxi fell behind.

Next we examine wage differences due to job attribute variables included in the second specification. As expected, the coefficients of gender, education and experience all fall in magnitude in comparison to the specification excluding job attribute variables. Coefficients for the three sets of dummy variables are plotted in Figures 14D, 14E, and 14F. Compared to state-owned units, workers in urban collectives suffered wage losses over the period (Figure 14D). The sample sizes for non-public enterprises were too small before 1992 to have much meaning, but after 1992, the non-public sector consistently paid higher wages than state-sector workers.[12] Figure 14F shows that there were significant changes in relative wages across sectors over the period. The most notable gaining sectors were education, arts and media, and government (including semi-government organizations), which are sectors that have a high concentration of college graduates. Transportation, and post and telecommunications, two sectors that traditionally enjoy monopoly power and higher salaries, also gained in the 1990s, which may partly reflect improved technology in those sectors. The only sectors that did not show any wage gain over manufacturing was retail, food, and logistics.

5.3 Decomposition of the changes in wage inequality

Next, we decompose the changes in wage inequality into quantity and price effects, following the methodology in Wing (1997) and Juhn et al (1993). The quantity effects are changes in the distribution of worker attributes over time, and the price effects are changes in the returns to those attributes. We first estimate selection-corrected cross-sectional regressions of the log real annual wages on regional dummies, a gender dummy, education dummies, and experience and experience squared. The estimated equation can be written as follows:

[pic] (1)

Here, [pic]is the log of real annual wages for individual [pic] in year t.[pic] is the vector of worker characteristics described above (or quantities), βt are the regression coefficients (or prices), and [pic] are unobservable returns to skill that are uncorrelated with the Xit. As in Juhn et al (1991), we decompose the error term into two parts: the standard deviation of the residuals (σt) and the normalized residuals with mean zero and variance equal to one (θit). The former can be considered a price, the latter a quantity.[13] Then equation (1) can be re-written as follows:

[pic] (2)

In each year, we decompose log wage differentials at the 90th-10th, 90th-50th, and 50th-10th percentiles of the wage distribution. Each wage differential can be expressed as follows:

[pic] (3)

where [pic] is the log wage differential in year t, [pic]is the vector of the average differences in observable characteristics between workers at the two percentiles, [pic] is the inner product of the vector of differences in the observable characteristics and the vector of their corresponding returns, [pic] is the average difference in percentile rankings of the residuals, and [pic] is the product of the standard deviation and the average difference in percentile rankings.

The change in wage inequality between two years t and [pic] can be expressed as follows:

[pic] (4)

The first term on the right hand side captures the effect on wage differentials of changing distributions in observable characteristics given fixed prices. The second term is the effect of changing prices of observable skills holding constant the distribution of observable skills. The third term measures the effects of changes in the average percentile rankings of wage residuals given a fixed price of unobserved skills, and the final term represents the effects of changes in the price of unobservable skills holding the distribution of those skills fixed.

Decomposition results are presented in Table 8. We present results for the 90th-10th, 90th-50th, and 50th-10th wage percentile differentials for the period 1988 to 1999, as well as the results for the 90th-10th wage percentile differential for four sub-periods. Each percentile group is defined to include all individuals within 10 percentile rankings. For example, the 90th percentile includes individuals with percentile rankings ranging from 80 to 100. In panel A, we present results for the specification described above, and in panel B, we add additional variables for ownership, occupation, and industry to see if changes in the distribution of job types and the returns to different jobs explains a substantial share of inequality growth. We recognize that the coefficients on these variables must be interpreted with caution since they are endogenous to job choices which may reflect individual unobservable characteristics.

We focus first on the results in panel A of Table 8. First, examining the changes in the log wage differentials (first row), we find that most of the total increase in wage inequality as measured by the 90th-10th percentile wage differential (0.712) occurred in the top half of the distribution (0.452 change in the 90th-50th wage percentile differential compared to 0.259 for the 50th-10th wage percentile differential), and much of the increase from 1988 to 1999 occurred during the 1992-1994 period (0.378), followed by the 1988-1992 period (0.213).

Second, all of the increase in inequality is attributable to changes in prices. Changes in quantities actually reduced the 90th-10th percentile wage differential over this period (-0.099 for observables and -0.027 for unobservables), but were overwhelmed by changes in prices (0.554 for observables and 0.284 for unobservables). The most important price changes contributing to wage inequality growth were the increase in regional wage differences, which alone accounted for a majority of the 90th-10th percentile log wage differential growth from 1988 to 1999 (0.472). Regional differences grew especially fast in the top half of the wage distribution, and especially in the earliest periods (0.271 from 1988 to 1992 and 0.152 from 1992 to 1994). The second most important observable price factor was education, which accounted for 0.097, or 13.6 percent, of the total 90th-10th percentile wage inequality growth. Unlike the regional wage differences, the effect of changes in the returns to education was most pronounced in the most recent period, 1997 to 1999, which is consistent with large recent increases in the returns to education (Zhao et al., 2002). Among the quantities, changes in experience had a substantial negative effect on inequality growth (-0.248), especially during the first two periods (1988 to 1994).

We also note the substantial contribution of the returns to unobserved skills to rising wage inequality. Increased variance of the residuals increased the 90th-10th log wage differential by 0.284, or 40 percent of the total increase. This increase was particularly pronounced in the second period, 1992 to 1994, and dissipated to less than one percent in the last period, 1997 to 1999.

In panel B of Table 7, we do the same decomposition, but add three job-related variables to the specification: ownership, occupation, and industry. Although self-selection into jobs complicates interpretation, these variables help indicate whether changes in the distribution of job types in the economy or in the relative returns to different jobs can help explain the inequality increase. For the most part, the results are similar to those for the simpler specification. Quantity changes in job characteristics negatively affect inequality, while changes in the relative returns to different job characteristics contribute positively to inequality. Of the three job-related variables, the price effects are strongest for industry (0.043), followed by occupation (0.023) and ownership (0.016). The combined job price effects are 0.082, or 11.5 percent of the increase in the 90th-10th percentile log wage differential. The gains are concentrated in the 1992 to 1994 period. The lack of an ownership effect suggests that labor market changes largely encompassed the state sector over this period, so that flows of workers out of the state sector and changes in the state-nonstate wage differentials were not key determinants of overall wage inequality growth.

6. Explaining rising returns to skill

The decomposition results validate our expectation that changes in labor market institutions are increasing the returns to education and skills, and that these changes are playing a prominent role in growing urban wage inequality. This is especially true for the recent period 1997 to 1999, when changes in the returns to education explain more than half of the total inequality increase (Table 8A). From 1988 to 1999, the increase in the college to high school premium as measured by wage regression coefficients are particularly large, growing from 13.8 percent to 38.9 percent (Table 6). However, the change in this premium is somewhat erratic over time. There are huge jumps in 1994, 1998, and 1999, and declines in 1991, 1995, and 1997, all years of relative slow economic growth.

Our goal in this section is to investigate the likely importance of different supply and demand factors in explaining the rising returns to education. We first explore whether changes in relative supply can explain the trend. Here the answer is clearly no since in China, like in the U.S., India, and Latin America, recent years have witnessed rising relative wages of college versus high school graduates even as the relative supply of college versus high school graduates has grown substantially.[14] Figure 15 plots the relative wage and relative supply of college versus high-school educated workers over time. Both are upward sloping, especially the relative supply line. To show that the lack of correlation between relative supply and relative wage generalizes beyond these two education groups, in Table 10, we report the trends for all education groups and find that the same positive correlation between relative supply and relative wage holds, especially for the lowest education group. In Figure 16, we plot relative wage and relative supply changes for 32 groups broken down by gender, education, and experience. Here, too, if anything there is a positive rather than negative relationship. Thus, the rise in the college-high school wage premium must be due to rising relative demand for college-educated workers.

In considering the importance of relative demand changes, it is useful to write down the following expression for the log of relative wages when one assumes constant elasticity of substitution between workers of different skill levels:

[pic] (5)

Here, wi(t) and xi(t) are mean wages and supply of workers of type i at time t, σ is the elasticity of substitution, and D(t) is relative demand. There are two approaches to estimating relative demand changes based on equation (5) given available data on wi(t) and xi(t). First, one can make an assumption about the parameter value σ and recover the change in demand by treating equation (5) as an identity. Alternatively, one can estimate the equation directly, proxying D(t) with a time trend and/or variables thought to affect relative demand. A byproduct of this approach is that one can recover an estimate of the elasticity of substitution from the coefficient of the log of relative supply.

Figure 17 plots the changes in relative demand based on estimates of the elasticity of substitution that fall in a conventional range (1 to 4). As expected, all of the lines are sharply upward sloping, reflecting a very substantial increase in the demand for college educated workers relative to high school-educated workers. However, as noted before with respect to educational premiums, the changes in relative demand over time are somewhat erratic, falling in 1990, 1991, 1993, and 1997.

Next, we estimate equation (5) directly. We regress relative wages on relative supply and other variables expected to affect relative demand, using both national and provincial data (for 6 provinces). We first proxy demand changes with a time trend. Then, we add the following three variables: FDI as a share of fixed investment, trade as a share of GDP, and the SOE share of industrial employment. Using the provincial data, we also estimate specifications which control for provincial fixed effects and year fixed effects.

Table xx presents the results. Because of the small number of observations (12 years for the national data and 72 province-year observations for the provincial panel) and our inability to control effectively for endogeneity, the results should be interpreted with caution. The statistical relationships thus reflect correlations in the data rather than causal relationships. Relative supply is significantly positively related to relative wages in all specifications, contrary to theory. This suggests that there are important omitted variables correlated with both variables (e.g, the timing of reforms), or that supply is responding to relative wages. When we add the additional regressors, nothing is significant using national data. Using provincial data, the coefficient for the FDI variable switches signs and is statistically insignificant across specifications. The coefficients on trade/GDP and SOE employment share are consistently negative, and in most cases are statistically significant. They are most negative and most significant in the preferred specification which includes a time trend and provincial fixed effects. Both results accord well with our expectations. Trade is expected to reduce skill premiums, as is a strong state sector in industry. Overall, the results suggest that globalization (FDI and trade) is not strongly correlated with rising skill premiums, and that reform of labor market institutions is important.

Next, following much of the literature, we attempt to quantify the relative importance of between-sector changes and within-sector changes on rising relative demand for college-educated workers. Some have argued that changes in demand associated with changing product preferences due to rising incomes, or changing production structure reflecting greater trade and specialization, should be manifest more in between-sector changes than in within-sector changes.[15] Alternatively, within-sector changes are likely to be driven more by skill-biased technological change. In the Chinese case, within-sector changes might also reflect changes in labor market institutions that affect all industries. Here, we are limited somewhat by the data, which only categorizes workers into 13 broad employment sectors, and does not, for example, distinguish among different sub-sectors within manufacturing. The sectors, ranked by number employed, are as follows: industry, commerce/trade, government, education and arts, transportation, housing, social services, construction, science, finance and insurance, mining, agriculture, and others.

We can break down between-sector and within-sector changes in several ways. First, we can decompose changes in the variance of wages into within-sector wage variation and between-sector wage differences, based on the following formula:

[pic] (6)

[pic]

The four terms in (6) measure the changes in wage inequality associated with changes in within-sector wage variation, changes in employment shares across sectors with different wage variances (composition effect), changes in wage variation across sectors, and changes in employment shares across sectors with different mean wages (composition effect). Table 11 reports the results for the whole period 1988 to 1999 as well as sub-periods. Nearly all of the inequality change is due to increased wage variance within industries.

In Table 12, we examine industrial employment shares for all workers and for college-educated workers, and show how these have changed from 1988 to 1999. The sectors that grew by more than 20 percent over this period, in order of the magnitude of increase, are finance and insurance, construction, agriculture, government, and transportation. Of these, only finance/insurance and government have a high concentration of college-educated workers. There also does not appear to be a strong correlation between sectors in which the share of college-educated workers increased and the initial skill-intensity of the sector. However, the lack of clear evidence of employment shifts toward skill-intensive sectors could be due to the level of sector aggregation.

We can decompose changes in relative demand into between-sector and within-sector components using two alternative methods. The first follows Katz and Murphy (1992), who define between-sector changes in demand for a group of workers as follows:

[pic] (7)

Here, [pic]is the increase in demand for workers of type k, Ek is the average share of workers that are type k across all years, Ejk is the number of workers of type k in sector j, Ej is the total number of workers in sector j, and [pic]is the average share of workers in sector j of type k. In order to calculate within-sector changes in demand, we assume that within-sector shifts are captured by movements across occupations within industries. Then we redefine subscript j in equation (7) to represent an occupation-sector category. The data codes 8 occupations and 13 sectors, or 104 occupation-sector categories. Then the change in demand using this group definition is defined as the total change in demand; the within-sector demand change is calculated as the difference between the total demand change and the between-sector demand change. The results of this exercise are presented in Table 13. The main finding is that the within-sector changes are much more important than the between-sector changes in explaining rising relative demand for more educated workers.

If we assume constant returns to scale, the change in relative demand for college versus high school graduates can be defined as the change in the share of the total wage bill paid to college graduates, where the total wage bill equals the sum of the college and high school wage bills. This change in the wage bill division can be decomposed into between-sector and between-sector components, based on the following simple formula:

[pic] (8)

Here, Sc is the share of the total wage bill paid to college-educated workers, [pic]is the share of industry j wages paid to college-educated workers, [pic] is the average share of total wages paid to sector j across all years, and [pic]is the average share of the industry j’s wages paid to college-educated workers. The first of the two terms on the right-hand side of equation (8) is a measure of within-sector demand changes, and the second captures between-sector demand changes. The decomposition results are presented in Table 14, which shows that nearly all of the change in relative demand for college-educated workers is attributable to within-sector changes.

7. Conclusions

In this paper we have documented an enormous increase in wage inequality among urban workers in China during the period 1988 to 1999. We highlight some of the main findings and then offer some final thoughts on remaining questions and directions for future research. Key findings are the following:

• Most of the inequality increase occurred in the top half of the wage distribution, with wage growth of the rich and highly educated growing particularly rapidly.

• The period of most rapid inequality increase was 1992 to 1994 when China substantially liberalized economic activity and experienced high growth. The most recent period, 1997 to 1999, has also seen substantial increases in inequality.

• There was a substantial increase in within-group inequality which is robust to numerous group definitions. The returns to unobservable skills as measured by regression residuals explains much of the increase in overall inequality.

• The wage inequality increase is due entirely to changes in the returns to worker attributes over time. Quantity effects are negative, if anything, dominated by changes in the distribution of experience. The main price effects are growing regional differences and higher returns to education.

• Many of the changes in inequality have occurred within the state sector as well as the non-state sector. Inequality has not been driven by a collapse of the state sector and large labor flows to the non-state sector.

• Rising skill premiums have been driven by increased relative demand for highly educated workers within industries rather than from changing demand and production among industries.

There remain important unanswered questions about the recent growth in urban wage inequality in China. First, it is difficult to distinguish between the importance of skill-biased technological change (as in the U.S.) versus labor market reforms as sources of rising skill premiums within industries. It will be of great interest to pursue research using more appropriate data to more directly link wage premiums to sources of technology, such as FDI. Second, why did inequality and skill premia increase so erratically over time? What specific policy and environmental factors explain the rapid inequality increase episode of 1992 to 1994? Third, what specific barriers or factors can explain such large and growing regional wage differential in China? In this paper, we have documented a number of provocative stylized facts which we hope will help define the agenda for future research.

References

Benjamin, Brandt and Yuen (2001). Employment Dynamics in State Owned Enterprises During Economic Transition, mimeo, University of Toronto.

Bound, John, and George Johnson (1992). “Changes in the Structure of Wages in the 1980s: An Evaluation of Alternative Explanations,” American Economic Review 82: 371-392.

Groves, Theodore, Hong Yongmiao, John McMillan, and Barry Naughton (1994). “Autonomy and Incentives in Chinese State Enterprises,” Quarterly Journal of Economics 109(1) : 185-209.

Juhn, Chinhui, Kevin M. Murphy, and Brooks Pierce (1993). “Wage Inequality and the Rise in Returns to Skill,” Journal of Political Economy101(3): 410-442.

Giles, John, Albert Park and Fang Cai (2003). How Has Enterprise Restructuring Affected China’s Urban Workers?, mimeo, Michigan State University.

Katz, Lawrence, and David Autor (1999). “Changes in the Wage Structure and Earnings Inequality,” in O. Ashenfelter and D. Card, eds., Handbook of Labor Economics, Volume 3, Elsevier Science B.V., pp. 1463-1555.

Katz, Lawrence and Kevin Murphy (1992). “Changes in Relative Wages, 1963-1987: Supply and Demand Factors,” The Quarterly Journal of Economics, 107: 35-78

Khan, Azizur, and Carl Riskin (1998). “Income and Inequality in China: Composition, Distribution and Growth of Household Income, 1988 to 1995,” The China Quarterly 154: pp. 221-53.

Li, Zhao, and Zhang (1997).

Riskin, Carl, Renwei Zhao and Shi Li (2001). China’s Retreat from Equality – Income Distribution and Economic Transition. New York: M. E. Sharpe.

Rutkowski, Jan (2001). “Earnings Inequality in Transition Economies of Central Europe: Trends and Patterns During the 1990s,” mimeograph.

Sanchez-Paramo, Carolina, and Norbert Schady (2002). Off and Running? The Rising Demand for Skilled Workers in Latin America, mimeo, The World Bank.

Wing, Suen (1997). “Decomposing Wage Residuals---Unmeasured Skill or Statistical Artifact?” Journal of Labor Economics 15(3): 555-566.

World Bank (1997). Sharing Rising Incomes, Washington D.C.: The World Bank.

Zhang, Junsen and Yaohui Zhao, (2003) “Economic Returns to Education in Urban China, 1988-1999,” paper presented at ASSA meeting, Washington, January 3-5.

Zhao, Yaohui (2002). “Earnings Differentials between State and Non-State Enterprises in Urban China,” Pacific Economic Review, Vol. 7, No. 1, 181-197

Table1: Trends in Inequality, 1988 to 2001

|Year |1988 |1989 |1990 |1991 |1992 |

|Standard deviation of log wages |0.45 |0.50 |0.63 |0.65 |0.70 |

|Percentile differentials: | | | | | |

|90-10 |1.08 |1.31 |1.72 |1.71 |1.92 |

|90-75 |0.26 |0.41 |0.45 |0.47 |0.47 |

|90-50 |0.51 |0.72 |0.89 |0.93 |0.96 |

|75-50 |0.25 |0.31 |0.44 |0.46 |0.49 |

|75-25 |0.53 |0.61 |0.86 |0.86 |0.97 |

|50-10 |0.57 |0.59 |0.82 |0.79 |0.95 |

|50-25 |0.27 |0.30 |0.42 |0.41 |0.48 |

|25-10 |0.29 |0.30 |0.40 |0.38 |0.47 |

|Observations |6087 |7853 |6752 |6641 |5404 |

Source: NBS, Urban household survey, 1988-2001.

Table 5. Inequality of Residual Wages, 1988-2001.

| |

|A. Parsimonious specification: Without controlling for occupation, industry and ownership |

|1988 |

|year |

|Birth year |1989 |1994 |1999 |

|1978-82 | | |1.81 |

|1973-77 | |2.17 |1.87 |

|1968-72 |1.30 |1.87 |1.86 |

|1963-67 |1.13 |1.88 |1.96 |

|1958-62 |1.11 |1.63 |1.94 |

|1953-57 |1.01 |1.61 |1.80 |

|1948-52 |1.00 |1.61 |1.80 |

|1943-47 |1.00 |1.64 |1.70 |

|1938-42 |0.93 |1.48 |1.65 |

|1933-37 |0.84 |1.45 | |

|1928-32 |0.88 | | |

|Average Changes within Cohorts and Experience Levels |

|Average Change |1989-94 |1994-1999 | |

|Within Cohorts |0.61 |0.09 |　 |

|Within Experience Levels |0.68 |0.12 |　 |

Table 8_A. Decomposition of Changes in Wage Inequality, 1988 to 2001.

|Percentile differentials: |90-10 |90-50 |50-10 |90-10 |90-10 |90-10 |90-10 |

|Periods: | |1988-2001 |1988-92 |1992-94 |1994-97 |1997-2001 |

| |Log wages |1.008 |0.797 |0.211 |0.311 |0.444 |0.104 |0.149 |

|Quantity changes (observables): |-0.136 |-0.065 |-0.071 |0.008 |-0.037 |0.000 |-0.060 |

| |Region |0.111 |0.060 |0.052 |0.113 |0.010 |0.020 |-0.039 |

| |Gender |-0.015 |-0.023 |0.008 |-0.006 |-0.001 |-0.005 |-0.005 |

| |Education |0.047 |0.035 |0.012 |-0.004 |0.044 |-0.008 |0.022 |

| |Experience |-0.279 |-0.137 |-0.142 |-0.096 |-0.090 |-0.008 |-0.039 |

|Quantity changes (Residuals): |0.040 |0.042 |-0.002 |-0.030 |0.051 |0.001 |0.261 |

|Price changes (observables): |0.520 |0.422 |0.098 |0.283 |0.218 |-0.032 |0.005 |

| |Region |0.438 |0.359 |0.079 |0.312 |0.174 |-0.022 |-0.019 |

| |Gender |-0.001 |0.000 |-0.001 |0.002 |0.004 |0.003 |-0.009 |

| |Education |0.104 |0.079 |0.025 |0.016 |0.049 |-0.011 |0.044 |

| |Experience |-0.021 |-0.015 |-0.006 |-0.047 |-0.008 |-0.002 |-0.011 |

|Price changes (Residuals): |0.584 |0.397 |0.187 |0.057 |0.080 |0.039 |0.165 |

Table 8_B. Decomposition of Changes in Wage Inequality, 1988 to 2001.

|Percentile |90-10 |90-50 |50-10 |90-10 |90-10 |90-10 |

|differentials: | | | | | | |

| |Log wages |1.008 |0.797 |0.211 |

|1990 |4.9 |5.2 |4.6 |4.0 |

|1995 |7.5 |6.9 |6.5 |6.3 |

|2000 |10.6 |10.0 |9.3 |9.3 |

Note: The returns are derived by interacting the schooling variable with experience dummy variables.

Table 10. Change in Relative Wage and Relative Supply of Different Educational Groups, 1988 to 2001.

| |1988-92 |1992-94 |1994-97 |1997-2001 |1988-2001 |

|Changes in log relative wage: | | | | | |

|College and above |3.8 |8.6 |5.1 |10.0 |27.5 |

|Special or technical |6.1 |0.9 |0.4 |13.4 |20.8 |

|Senior high |0.3 |2.7 |3.6 |-1.1 |5.5 |

|Junior high |-6.8 |-6.6 |-7.3 |-12.4 |-33.0 |

|Changes in log relative supply: | | | | |

|College and above |36.0 |9.9 |2.7 |19.5 |68.1 |

|Special or technical |7.2 |0.9 |-8.6 |-6.8 |-7.3 |

|Senior high |5.5 |2.4 |1.8 |0.0 |9.8 |

|Junior high |-23.3 |-7.6 |-4.2 |-25.6 |-60.7 |

Table 11. Effects of Industrial Composition on Wage Inequality.

|　 |　 |Within-industry |Between-industry |

| |Total |Variance |Composition |Wage |Composition |

|Years |Changes |Changes |Effect |Changes |Effect |

|1988-01 |0.300 |0.261 |0.001 |0.039 |-0.001 |

|1988-92 |0.046 |0.041 |-0.003 |0.008 |-0.001 |

|1992-94 |0.157 |0.141 |0.002 |0.014 |0.000 |

|1994-97 |0.021 |0.033 |-0.001 |-0.011 |0.001 |

|1997-01 |0.076 |0.046 |0.002 |0.029 |-0.001 |

Table 12. Employment Shares of the Overall Sample and College Labor by Industry.

|　 |1988 |2001 |Percent change, |1988 |2001 |Percent change, |

| | | |1988-2001 | | |1988-2001 |

|Sector |The overall sample |College and above |

|Agriculture |0.58 |1.21 |108.26 |0.98 |1.60 |64.03 |

|Industry |34.23 |34.11 |-0.35 |21.23 |22.38 |5.44 |

|Mining |9.22 |0.52 |-94.31 |4.95 |0.77 |-84.52 |

|Construction |2.91 |3.83 |31.54 |2.51 |3.79 |51.07 |

|Transportation |6.34 |7.59 |19.57 |2.61 |3.98 |52.33 |

|Commerce/trade |14.50 |14.66 |1.09 |5.14 |9.95 |93.59 |

|Public utilities |6.34 |8.95 |41.05 |1.44 |5.84 |305.61 |

|Social services |4.41 |4.72 |7.17 |7.89 |6.15 |-22.06 |

|Education, arts, and media |6.84 |6.94 |1.52 |18.91 |12.99 |-31.33 |

|Scientific research |3.51 |1.41 |-59.74 |12.41 |2.74 |-77.93 |

|Finance and insurance |1.39 |3.54 |154.20 |2.57 |7.02 |173.03 |

|Government |8.57 |10.95 |27.73 |18.81 |21.63 |14.98 |

|Other industry |1.15 |1.56 |36.25 |0.54 |1.15 |112.78 |

Note: Following table 15 of Katz & Autor.

Table 13. Between-Industry and Within-Industry Changes in Relative Demand for Workers, by Education Group, 1988 to 2001.

| |Total |Between-industry |Within-industry |

| |1988-01 |1988-92 |1992-94 |

|1988-01 |0.187 |0.174 |0.012 |

|1988-92 |0.087 |0.080 |0.007 |

|1992-94 |0.035 |0.026 |0.009 |

|1994-97 |-0.001 |0.008 |-0.009 |

|1997-01 |0.065 |0.055 |0.010 |

Note： Sánchez-Páramo and Schady，The World Bank，2002.

Table 15. Within-Industry and Between-Industry Changes in Relative Demand for College versus High School Graduates, 1988 to 2001 ([pic]=1) (newly added)

| Period |Total |Within industry |Between industry |

|1988-01 |1.42 |0.68 |0.74 |

|1988-92 |0.56 |0.30 |0.26 |

|1992-94 |0.25 |0.09 |0.16 |

|1994-97 |0.11 |0.06 |0.05 |

|1997-01 |0.43 |0.20 |0.24 |

Note: Katz and Autor, 1999.

Figure 1: Kernel density estimates of log real annual wages for selected years.

[pic]

Figure2: Real Annual Wages by Percentiles, 1988-2001.

[pic]

Figure 3: Log Real Wage Changes by Percentile, 1988-2001

[pic]

Figure 4: Log Wage Differentials by Percentiles, 1988-01.

[pic]

Figure 5: Annual Average Wage Growth by Percentiles.

[pic]

Figure 6A: Real Wage Growth by Wage Percentile by Experience Group, 1988 to 2001.

[pic]

Figure 6B: Real Wages by Experience Group, 1988-2001.

[pic]

Figure 7A: Real Wage Growth by Wage Percentile by Education Group, 1988-2001.

[pic]

Figure 7B: Real Wage by Education Group, 1988 to 2001.

[pic]

Figure 8A: Real Wage Growth by Wage Percentile by Education Group for Young Experience Group (1-10 Years), 1988 to 2001.

[pic]

Figure 8B: Real Wage Growth by Wage Percentile by Education Group for Older Experience Group (21-30 Years), 1988 to 2001.

[pic]

Figure 9A: Real Wage Growth by Wage Percentile by Gender, 1988 to 2001.

[pic]

Figure 9B: Real Wages by Gender, 1988 to 2001.

[pic]

Figure 10A: Real Wage Growth by Wage Percentile by Province, 1988 to 2001.

[pic]

Figure 10B: Real Wage by Province, 1988 to 2001.

[pic]

Figure 11A: Real Wage Growth by Wage Percentile by Ownership, 1992 to 2001.

[pic]

Figure 11B: Real Wages by Ownership, 1988 to 2001.

[pic]

Figure 12A: Real Wage Growth by Wage Percentile by Occupation, 1988 to 2001.

[pic]

Figure 12B: Real Wages by Occupation, 1988 to 2001.

[pic]

Figure 13A: Real Wage Growth by Wage Percentile by Sector, 1988 to 2001.

[pic]

Figure 13B: Real Wages by Sector, 1988 to 2001.

[pic]

Figure 14A: Returns to Education by Levels, 1988-2001.

[pic]

Figure 14B: Returns to Potential Experience, 1988-2001.

[pic]

Figure 14C: Coefficients of Provinces, Reference Sichuan, 1988-2001.

[pic]

Figure 14D: Coefficients of Ownership, Reference State-sector, 1988-2001.

[pic]

Figure 14E: Occupation Coefficients, Reference Manual Workers, 1988 to 2001.

[pic]

Figure 14F: Coefficients of Sectors, Reference Manufacture, 1988-2001.

[pic]

Figure 15: Relative Wage and Supply Differentials for College vs. High School, 1988 to 2001.

[pic]

Figure 16: Price and Quantity Changes for 32 Groups, 1988 to 2001.

[pic]

Figure 17: Alternative Implied Relative Demand: College vs. High School, 1988 to 2001.

[pic]

-----------------------

[1]The World Bank (1997) finds that China’s overall gini coefficient grew from 0.288 in 1981 to 0.388 in 1995, from 0.176 to 0.275 in urban areas and from 0.242 to 0.333 in rural areas. Khan and Riskin (1998) report that the overall gini coefficient grew from 0.382 in 1988 to 0.452 in 1995, from 0.233 to 0.332 in urban areas and from 0.338 to 0.416 in rural areas. Using National Statistical Bureau data covering 18 years from 1978 to 1995, Li, Zhao and Zhang (1997) found that the Gini coefficient increased from 0.16 to 0.28 in urban areas and from 0.21 to 0.34 in rural areas.

[2]For example, Rutkowski (2001) reports that the income ratio of the 90th to 10th percentiles grew from 3.1 to 6.3 in China from 1988 to 1999, from 3.4 to 4.2 in Hungary from 1990 to 1997, from 2.9 to 3.5 in Poland from 1991 to 1999, and from 2.4 to 5.0 in Romania from 1991 to 1999.

[3]An important exception is the set of studies from the China Income Project based on surveys in 1988 and 1995 (Riskin et al., 2000).

[4] All members of the households are included in the survey. Although Beijing is a city, it enjoys the same administrative status as a province.

[5] Age 60 is official retirement age for male managers. Female workers retire at 50 and male workers and female managers at 55.

[6] Juhn et al. (1993), Katz and Murphy (1992) and Katz and Autor (1999), among others, also apply this sample exclusion rule. Although China started implementing a minimum wage system since 1995, we have information on minimum wages only for the last two years, 1998 and 1999. Because China experienced rapid growth rate of real incomes in our data period, we discount the average of real minimum wage in 1998 and 1999 by the wage growth rate to derive the implied minimum wage in previous years.

[7] Because urban resident permits :; ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download