Assimilation or Disassimilation? The Labour Market Performance of Rural Migrants in Chinese Cities Dandan Zhang Xin Meng August 31, 2007 Abstract Although significant earnings differentials between urban residents and rural migrants in Chinese cities have been revealed in many studies with single cross-sectional data, it is still not clear how the wage gap between the two groups of workers evolves over time. To fill the research gap, this paper uses a dynamic wage decomposition method and an economic assimilation model with repeated cross-sectional data from China Income Distribution Survey 1999 and 2002 to investigate changes in wage gap between urban residents and rural migrants from a dynamic perspective, and to explore the wage assimilation for new rural migrants towards their urban counterparts and its determinants in Chinese cities. Two technique problems associated with an economic assimilation model are solved: the lack of common support and identification problems. Two interesting results are found. First, there is a growing wage gap between rural migrants and urban residents across the two surveyed years, mainly caused by the decline in the rate of return to education for rural migrants. Second, rural migrants wages may generally assimilate to those of their urban counterparts, but well-educated rural migrants and those residing in capital cities (i.e., large cities) seem to have no significant advantage in the assimilation process. Both findings imply the current institutional arrangements in the China s urban labour market tend to favour less-educated rural migrants, which in turn may restrict new rural migrants from investing more in their human capital. Keywords: Assimilation, Rural-Urban Migration, Wage Differential, China JEL Classification: J30; J45; J61 Corresponding author: Dandan Zhang; Email: dandan.zhang@anu.edu.au. Economics Division, Research School of Pacific and Asian Studies, The Australian National University. Research Institute of Population and Labour Economics, Chinese Academy of Social Sciences. Economics Division, Research School of Pacific and Asian Studies, The Australian National University.
I Introduction The emergence of large scale spontaneous rural to urban migration is currently one of the significant phenomena in China s labour market. Motivated by the income disparity between the rural and urban sector, tens of millions of redundant rural workers that were gradually released from the lowproductivity agricultural sector have moved to the urban areas since the late 1980s (Meng, 2000; Zhao, 2000). Due to the estimate of the National Bureau of Statistics of China, there have been 132 million rural-urban migrants working in cities in 2006, accounting for around one third of the Chinese urban labour force. Over the next two decades, an additional up to 150 million rural workers are expected to move into the cities. Although rural-urban migrants have made make a significant contribution to Chinese economic growth during the past three decades, they suffer from discrimination in their host cities due to the current labour market institutional arrangements. These institutional barriers deteriorate the living standards of rural migrants in their host cities, making Chinese rural-to-urban migration a unique world phenomenon. One example of this is the guest worker system put in place by Chinese government for rural citizens, under which rural migrants usually hold a temporary work visa, restricts rural-to-urban migrants from attaining urban residency status (Du et al., 2006). The poor situation of rural-urban migrants in urban labour market, has been attracting the concern of both policy-makers and academics in China. Most relevant studies on rural-to-urban migration in China focus on investigating the relative labour market performance of rural migrants comparing with their urban counterparts. Many found that significant differences in wages and occupational attainments between rural migrants and urban residents, implying that rural-urban migrants occupied in disadvantaged positions in Chinese cities (see Meng, 2001; Meng and Zhang, 2001; Zhao, 2003; Du, et al., 2006; Park, et al., 2006). Although these previous studies furthered understanding of the rural-to-urban migration issue from a static perspective, none attempted to address it from a dynamic perspective due to lack of methodology and suitable data. This raises questions, such as how the wage gap between rural migrants and their urban counterparts may change over time, and whether rural migrants can recover from the initial wage disadvantage depending on the length of time they spend in the host city. Answers to these questions are important since they can help in understanding whether and how rural-urban migrants can settle down in cities as more and more rural-urban migrants come to the cities. According to the economic assimilation theory of migration, immigrants usually earn less than Page 1 of 53
the native-born upon arrival, but this earnings-gap can erode with time spent in the host region. As immigrants first arrive, they usually suffer from lack of local experience, such as years of schooling and work experience and so are disadvantaged in the local labour market. However, when they stay for a longer period, accumulation of country-specific human capital, such as job-specific training, language, knowledge of culture and customs can help them to overcome the market discrimination, making them comparable with their local counterparts (Chiswick, 1978). The rate at which the gap narrows is usually interpreted as a measure of economic assimilation (Meng and Gregory, 2006). Consistent with the theory, a large number of empirical studies on international migration in western countries have found supportive evidences: immigrants (including both skilled and unskilled immigrant groups) could assimilate with and finally catch up to their native counterparts with length of time residing in the host region (see Chiswisk, 1978; Borjas, 1985 and 1995; Friedberg, 1992; LaLonde and Topel, 1992; Baker and Benjamin, 1997). However, is the economic assimilation theory applicable for explaining the dynamic change in the wage of internal migrants in a developing country, like China? When the initial discrimination between rural-urban migrants and urban residents comes from government policies rather than market selection, can these rural-urban migrants still catch up with their urban counterparts economically? The answer relies on some empirical analysis using China s rural-to-urban migration data. Based on previous studies, this paper aims to explore the wage gap between rural migrants and urban residents from a dynamic perspective by using both the trans-temporal decomposition method (Smith and Welch, 1989) and the economic assimilation model (Chiswick, 1978; Borjas, 1985 and 1995) with the repeated cross-section data from the China Income Distribution Surveys (CIDS) of 1999 and 2002. There is a methodological contribution to the literature from three aspects. First, the economic assimilation model is used to address the dynamic wage change of rural-urban migrants in China, which extends application of the model to cover the internal migration issue. Second, the propensity score matching method (Rosenbaum and Rubin, 1983) is used to solve the lack of common support problem, which base the comparison between rural migrants and urban workers on a sample with identical characteristics. Third, the year effect is decomposed with other independent variables, which helps to relax the constant year effect assumption based on a traditional economic assimilation model. Three questions will be posed in this study: how does the wage gap between urban residents and rural migrants evolve over time? Do rural migrants assimilate economically towards their urban counterparts? What are the factors affecting any migrant assimilation process? Page 2 of 53
As for the wage dynamic of rural migrants relative to their urban counterparts in urban China, the first finding is that there was a growing wage gap between rural migrants and urban residents during 1999 and 2002. The decomposition analysis reveals the widening wage gap was mainly caused by the decline in rate of return to education for rural migrants, whereas gaining experience (both total potential experience and experience in the host city) helps narrow the wage gap. This suggests local experience plays an important role in affecting the labour market performance of rural migrants in urban China. Furthermore, the economic assimilation analysis for the sample with common support shows that, as happens with international migrants in western countries, rural migrants wages may generally assimilate to those of their urban counterparts, although for well-educated rural migrants or those residing in capital cities there appears to be no significant advantage in the assimilation process. Both findings imply rural migrants in urban China may overcome the disadvantage caused by both the market selection and unfavourable institutional settings and assimilate to their urban counterparts economically as a result of the length of time spend in the host city. However, the current institutional arrangements in the China s urban labour market tend to favour less-educated rural migrants, which in turn may restrict new rural migrants from investing more in their human capital. The rest of this study is organized as follows. The next section presents a review of previous literature, particularly related to economic assimilation models and their empirical application to international migration issues, followed by discussion of rural-to-urban migration, its background and some relevant existing empirical results in China. Section 3 describes the data and Section 4 shows some summary statistics. A migrants/urban workers wage comparison will be conducted in Section 5, and the decomposition method is adopted to reveal the main factors accounting for the dynamic change in wage gap. Section 6 discusses the common support problem when comparing two groups of workers, models specifications of the economic assimilation model, as well as their corresponding regression results. The conclusions are given in section 7. Page 3 of 53
II Literature Review Economic Assimilation Immigration literature provides a large number of empirical models to analyse the differences in labour market outcomes of immigrants and natives, its causes and changes over time as immigrants adapt to the host country s labour market. As one of the most powerful techniques, the economic assimilation model has been widely used to address how the difference between immigrants and natives has changed over time for both international and internal migration during the past three decades. This section presents a review of this model and its empirical application to international migration data and discusses the applicability of the model for analysing the assimilation of rural-urban migrants in China. The purpose of this section is to set up the methodology for analysing rural-urban migration assimilation in urban China. The following is a review of several important papers in the development of economic assimilation models, including Chiswick (1978), Borjas (1985), Friedberg (1992), and LaLonde and Topel (1992). It is widely observed that, although initially lower than those of the native-born population in many western countries, as time spent in the host country increases immigrants earnings tend to catch up. This phenomenon of a falling gap between immigrants and native-born over time is usually defined as economic assimilation. The initial wage gap is thought to be explained by a lack of skill transferability across countries and the economic assimilation is generally attributed to the accumulation of specific experience by immigrants in the host country, such as job-specific training, language, culture and customs (Chiswick, 1978). Since the extent of economic assimilation can be helpful in understanding the changing trends of the economic status of immigrants in the host country, a lot of research, especially in the U.S., Canada and Australia, has measured the economic assimilation of immigrants. The economic assimilation model was originally developed by Chiswick (1978). His model, based on the use of cross-section data, aimed to analyse the relative economic performance of immigrants in the host country s labour market. The core empirical function of the model in Chiswick (1978) can be written as logw age i = X i β 0 + β 1 I i + β 2 Y SM i + ε i (1) Page 4 of 53
where W age i gives the wage rate of person i in the host country; X i is a vector of socioeconomic characteristics (often including age and education); I i is a dummy variable set to unity of person i who is foreign-born; and Y SM i gives the number of years the immigrant has resided in the host country (the United States in this case) and is set to zero if i is a native. Equation (1) shows how the wage difference between immigrants and natives is determined using cross-section data. Since the age effect has been controlled and normally assumed to be identical between natives and immigrants, the coefficient β 2 measures the differential value associated with the time spent in the host country s labour market versus the time spent in the immigrants source country. With Equation (1), Chiswick (1978) used 1970 U.S. Census data to examine how quickly immigrants adapted to the U.S. labour market. Two fundamental results were revealed and include the earnings of immigrants are significantly lower than the earnings of immigrants who have been in this country for longer periods (but tend to catch up over time), and the relatively rapid growth of immigrant earnings over time leads to the existence of an overtaking age, at which point the earnings profiles of the native and the foreign-born cross (Borjas, 1985, p.464). This provided empirical evidence for the first time to explain the higher increasing rate of immigrant age-earning profiles in the U.S. which Chiswick (1978) attributed to labour market selection for able immigrants and immigrants having a higher motivation than native-born to work longer and harder. Although Chiswick (1978) contributes to setting up the baseline model for economic assimilation study and to properly explaining the Americanisation of immigrants, his early model has an important shortcoming. Borjas (1985) first questioned the empirical validity of Chiswick (1978) and suggested Chiswick s use of cross-section data could have over-estimated the true assimilation impact as it does not take appropriate account of quality differences across immigrant cohorts. Using the 1970 and 1980 Public Use Samples from the U.S. census, Borjas (1985) argued that the phenomenon of rapid wage assimilation between immigrants and native-born observed by Chiswick (1978) was more likely to be due to the cohort effect rather than the accumulation of immigrants specific experiences in the host country. Specifically, since immigration policy, the size and national origin of the immigrant flow had changed substantially in the U.S. during the period 1970-1980, the structure and quality of immigrants was different over time. If these changes generated a lower quality immigrant flow, the cross-section correlation indicating more recent immigrants earn less would say little about Page 5 of 53
the process of wage assimilation, but would instead reflect innate differences in ability or skills across cohorts. This suggests the use of only one cross-section data may create biased results if there exist big differentials across immigrant cohorts, and this estimation bias may be upward or downward depending on the changes in immigrants quality. As suggested by Borjas (1985), the solution for the above problem is to introduce the cohort variable into Equation (1). Doing so raises the identification problem since there will be perfect multi-collinearity amongst year, cohort and year since migration. Thus, longitudinal data should be used, where a particular worker is tracked over time, or, equivalently, a number of randomly drawn cross-sectional data (at least two waves) 1 should be used with some specific restrictions imposed on the relationship between age, cohort and year since migration, so that specific cohorts can be tracked across the survey years. By pooling the 1970 and 1980 Public Use Samples from the U.S. Census, Borjas (1985) developed a synthetic panel data economic assimilation model with both an immigration equation and a native equation. Immigrant equation: L lnw age lt = X lt φ it + δ i Age lt + αy SM lt + βcohort lt + γ it Y ear lk + ε lt (2) t=t Native equation: L lnw age lt = X lt φ nt + δ n Age lt + γ nt Y ear lk + ε lt, (3) t=t where W age lt is the wage of person l in cross-section t; X lt gives a vector of socioeconomic characteristics; Age lt is the worker s age at the time of survey; Cohort lt (i.e. year-of-arrival) gives the calendar year in which immigrants arrived in the host country; Y SM lt (i.e. year-since-migration) is the number of years the immigrant has resided in the host country; and Y ear lt gives the dummy variable indicating whether person l was drawn from cross-section t. Note that a more general model would allow for non-linearities in the age, year-since-migration, and year-of-arrival variables, as well as variation in the coefficient vector (φ, δ) over time, and differences in the coefficient α across immigrant categories, by introducing some interaction terms. Equations (2) and (3) show how the wage assimilation between immigrants and native-born is determined using the synthetic panel data. Although the interpretation for the coefficients was 1 The pooled cross-sectional data are also called the synthetic panel data, coined by Clark and Lindley(2006). Page 6 of 53
similar as in the cross-section data model, the measurement of these coefficients was significantly different since the cohort of immigrants was controlled. For example, parameter α measures the differential value of a year spent in the host country versus a year spent in the source country and the economic assimilation is defined independently as α = (δ i + α) δ n. Moreover, since Cohort lt + Y SM lt = Y ear lk, some restrictions should be imposed to separate the assimilation, cohort and period effects and carry out regressions with Equations (2) and (3). Borjas (1985) proposed restriction of the period effects being identical for immigrants and native: γ it = γ nt, or trends in aggregate economic conditions change immigrant and native wages by the same percentage. His reason was that the period effects for immigrants are usually exogenous to the immigrant wage determination system (Borjas, 1985). With this restriction, the two stage OLS method can be used - the native function can be first estimated to obtain γ nt and then the same period effect can be imposed on the immigrant equation - to estimate the assimilation effect α. With the economic assimilation model using the synthetic panel data, Borjas (1985) extended the early empirical study on the economic assimilation between immigrants and native-born by avoiding the complexity caused by the cohort effect due to the use of the cross-section data. The empirical results from the new model showed the earnings of male immigrants grew at a much slower rate than suggested by previous studies using cross-sectional data. This provided some empirical evidence for Borjas s (1985) hypothesis that there had been a secular decline in the quality of male immigrants admitted to the U.S. during the 1970s. This study not only challenged the results from Chiswick (1978) on the unconditional convergence of wages between immigrants but also created a new empirical framework for analysing the economic assimilation. Friedberg (1992) elaborated on the economic assimilation model by introducing the variable of immigrant s age at migration and another identification condition. This helps merge the two equations in Borjas s (1985) model into one through introducing the appropriate interaction terms between nativity, the survey year dummy and other explanatory variables, thereby simplifing the regression in the empirical study. Specifically, Friedberg (1992) argued that the model shown in Equations (2) and (3) ignored an important aspect of immigrant wage determination: the role of age at arrival in the host country. For example, the U.S. data suggested a strong negative correlation between age at arrival and entry earnings. Thus, the identification problem, in the Borjas (1985) model does not disappear, even after imposing his identification condition but becomes more severe, when the entry Page 7 of 53
wage of immigrants depends on age-at-migration. To deal with the worsened identification problem, Friedberg (1992) suggested putting the immigrant s age at migration into the regression and imposing another identification condition: the age coefficient is the same in both the immigrant and native samples. Consider an extended form of Equation (2): L lnw age lt = X lt φ it + δ i Age lt + αy SM lt + βcohort lt + θm lt + γ it Y ear lk + ε lt, (4) t=t where M lt gives the immigrant s age at migration. As before, the parameter vector (α, β, γ) in Equation (4) cannot be identified. In addition, the inclusion of the age-at-migration variable leads to another identity: M = Age Y SM. With these two identities, there will be perfect collinearity even after imposing restrictions of the same period effects for immigrants and natives. Thus, an additional restriction must be imposed on the data. That is the coefficient of the age variable is the same for immigrants and natives (Friedberg, 1992). The estimation of the system in Equations (4) (for immigrants) and (3) (for natives) then requires that δ i = δ n and γ it = γ nt, for all t. Although both Borjas (1985, 1995) and Friedberg (1992) suggested two separated regression models for immigrants and natives, the one equation regression (like Equation 1 incorporating the cohort effect) has dominated empirical studies for exploring the economic assimilation of immigrants. Economists usually employ one single equation to simplify the estimation with larger samples in some existing literature (see Baker and Benjamin, 1997; McDonald and Worswick, 1999; Blau et al., 2002; Antecol et al., 2004; Meng and Gregory, 2005; Clark and Lindley, 2006). To relax the strong restriction imposed on the single equation, some interaction terms between nativity, survey year and other characteristics (such as age and education level) may be added into the regression model, to allow those independent variables to change across nativity, and different survey years. Based on the contributions of Borjas (1985, 1995) and Friedberg (1992), there have been a large number of studies exploring the economic assimilation of immigrants in the U.S.. However, Borjas (1999, p.1721) states that even after the analysis has allowed for the possibility of cohort effects, there seems to be a great deal of confusion in the empirical literature about whether immigrants in the United States experience a substantial degree of economic assimilation. To explain this phenomenon, LaLonde and Topel (1992) raised the hypothesis that the confusion in empirical results Page 8 of 53
might come from a conceptual disagreement over the definition of assimilation. To support their claim, LaLonde and Topel (1992) proposed a very different definition of the process from the traditional definition for economic assimilation: assimilation occurs if, between two observationally equivalent (foreign-born) persons, the one with greater time in the U.S. typically earns more (LaLonde and Topel, 1992, p.75). As they argued, the new definition, reflected in the above economic assimilation model of Borjas (1985) (see, Equations 2 and 3), can be simply defined as parameter α, the coefficient of years-since-migration in the immigrant earnings function. Compared with the traditional definition of economic assimilation, say α, is defined by comparing the economic value (in terms of the host country s labour market) of a year spent in the host country relative to a year spent in the source country (Bojas, 1999, p.1721). That is, the base group in the new definition of economic assimilation is that of the immigrant rather than the native-born. Through using this new definition, LaLonde and Topel (1992) found significant economic assimilation between the old and new waves of immigrants. This helps provide a possible explanation as to why different empirical studies sometimes arrive at different conclusions about economic assimilation: the ambiguous choice of a base group in previous studies makes their empirical results incomparable. It is therefore important to choose a consistent definition for the index of economic assimilation when comparing various study results, though there is wide disagreement on which definition of economic assimilation should be used. LaLonde and Topel (1992) argue that the basic assumption for constant period effect may be violated in some cases when the wage distributions, and the human capital stocks or skill levels are significantly different for the two comparison groups. In this case, estimate results will be biased based on a traditional economic assimilation model. LaLonde and Topel (1992) recommended comparing immigrants to native-born who would have experienced similar wage changes in inequality and relative skill prices. However, they did not provide a method to find the optimal base group, or in other words to equalize the period effect. Beginning with LaLonde and Topel (1992), economic assimilation models evolved to be a mature methodology for analysing changes in the earnings gap between immigrants and natives. Three preliminary conditions of the economic assimilation model must be satisfied before it can be used empirically. One is that there is enough information to distinguish between immigrants and native-born both theoretically and empirically then the evolution of their wage can be tracked over time. Second Page 9 of 53
is that there is enough information on year since migration and years of arrival for immigrants, and third is to specify the counterfactual group. Imposing these three conditions, this analysis aims to apply the economic assimilation model to examine internal migration in China, something that has not yet been done in previous studies. Assimilation, Rural-Urban Migration in China, and Some Empirical Evidence Although previous economic assimilation studies help in advancing our knowledge on economic assimilation, most focus on the international migration issue in developed countries while ignoring the internal migration issue in developing countries, especially those characterized by a dual economy. The main reason is that wage assimilation for internal migration is not usually treated as an important issue for an integrated labour market. However, China differs from other developing countries in experiencing a long-term segregation between rural and urban labour markets. This provide an opportunity to fill the research gap and analyze the economic assimilation of domestic migration in a dual economy instead of international migration by taking into account the special case of rural-tourban migration in China. The survey below follows some related studies on rural-urban migration, income disparity between rural migrants and urban residents, and two-tier labour markets in urban China. China has adopted the policy of sacrificing agriculture and subsidizing industries as its economic development strategy. To fulfil this development strategy, population mobility has been strictly controlled since the beginning of the 1950s (Zhao, 2002). As a consequence, the rural and urban economies have been isolated for about 40 years. This segregation was mainly implemented through the Household Registration System (Hukou System), which artificially segments people into agricultural and non-agricultural Hukou. Despite their different explanations for the population mobility control policy, many previous studies on rural-urban migration argued that the strict enforcement of rural-to-urban migration controls has segregated the urban labour market from the rural one and generated an institutional bias towards urban residents from the early 1950 till the end of 1980s (see Lin, Cai and Li, 1996; Cai, 2001; Cai, Du and Wang, 2001; Zhao, 2002; and Cai and Wang, 2004). However, as economic reform was initiated in the late 1970s, millions of rural labourers were released from agricultural production. Since labour mobility was tightly controlled by the Hukou system until the middle 1980s, these redundant rural workers were originally absorbed by the rural Page 10 of 53
non-agricultural sector. However, during the early 1990s, the rapid growth development of the urban economic sector, especially the private and informal sectors, created demand for rural labour in urban areas. Thus, both supply and demand forces pushed the government to relax the restrictions on ruralurban migration (Meng and Zhang, 2001). As a result, the population mobility control lost its initial effectiveness and urban labour market reform and rural and urban economies began to be linked. Attracted by higher urban incomes, after 40 years of segregation tens of millions of rural workers flowed into the cities. Although there is no accurate estimate of the scale of migration population in China, it is usually believed that around 132 million migrant workers are currently residing in cities, accounting for one third of the Chinese urban labour force. Although the strict restrictions on rural-urban migration have gradually eased, rural migrants in urban China are still treated differently from their urban resident counterparts. Thus, migrants only have temporary permission to work in a host city with no access to the social benefits available to their urban counterparts. As a consequence, these migrants have to be involved in long-hours-low-payment jobs and so can not be substitute of urban resident workers in the urban labour market. There have been many studies investigating the income disparity between rural-urban migration and urban residents in China (see Meng, 2001; Meng and Zhang, 2001; Zhao, 2003; Du et al., 2006; Park, et al., 2006). Most found a significant difference in occupational attainment and wages between rural migrants and urban residents. For example, Meng (2000) and Meng and Zhang (2001) argue the wage differentials between rural migrants and urban residents is due mainly to occupational segregation. Meng (2000) suggests a two-tier labour market exists in China featured by typical urban residents working in higher-ranked jobs with higher wages, and more government subsidies, relative to rural migrants. Migrant workers as an excluded group are usually hired in low income and unwanted 3D jobs, i.e., dirty, dangerous, and disgraced tasks. There are also significant wage differentials between urban residents and rural migrants. Using survey data in Shanghai from 1995 and 1996, Meng and Zhang (2001) further explored occupational segregation between rural and urban workers. Their results show rural migrants are much less likely to obtain white-collar occupations than their urban counterparts. After controlling for individual characteristics such as education and experience, the majority of the wage gap between the two groups of workers is accounted for by occupational segregation rather than the difference of human capital. Page 11 of 53
More recently, Du et al. (2006) state that rural migrants are only allowed to stay in cities temporarily rather than settle down permanently. This study describes rural-urban migrants usual objective is to earn more and save as much money as possible before returning home. Accordingly, rural migrants have a short-term focus. Using two different data sources in Chinese cities conducted in 2001 and 2002, Du et al. (2006) explored the impact of the guest worker system on the wellbeing (such as earnings and health) of migrant workers in urban China. They found more migrants live in poverty than urban residents and that migrants work 50 percent more hours per week than their urban counterparts. This may reduce migrant poverty in the short run but may also adversely affect their long term health condition. Park et al. (2006) also focuses on the influence of rural migrants on urban poverty and inequality in China. Contrary to the results of Du et al. (2006), using data from 12 cities in 2004 and 2005, they find relatively small differences in the poverty rates of rural migrants and urban residents. They explain this as owning to higher labour force participation rates and longer hours worked for migrant households. Their study shows the significant differences between migrants and urban residents are actually their non-income benefits and accessibility to social welfare systems. Another branch of studies on rural-urban migration focuses on migration duration and return migration. They generally find that after a few year stay in cities rural migrants seem to have a strong willingness to stay rather than return to the countryside. Even without the legal status, migrants prefer to stay as long as circumstances permit. Field research in Anhui and Sichuan provinces in 1998 (RCRE, 1999) suggests that, when forced to leave one city, rural migrants usually go to another and only return home as a stopover between these two migration trips. Using a household survey conducted in 1995, Hare (1999) finds very few permanent returnees in a county in Henan province. Zhao (2002) examines the causes of return migration using a rural household survey conducted in six Chinese provinces in 1999. She finds a limited scale of return migration. Most of these migrants had not returned home permanently upon leaving their jobs; and returning migrants tended to be older, married, better educated, and with a spouse who had not migrated. This phenomenon suggests the guest worker system has not prevented the transfer of labour into urban areas. Although previous studies have made a large contribution to the issue of rural-urban migration from a comparative static perspective, few have addressed the issue from a dynamic perspective, e.g., the economic assimilation in wages for rural migrants in the host city. This anaylsis aims to use the dynamic wage decomposition technique and economic assimilation model to examine whether migrants Page 12 of 53
can assimilate to their urban counterparts based on the length of time residing in host cities and its determinants. III Data and Summary Statistics The data used comes from the China Income Distribution Surveys (CIDS) of 1999 and 2002, conducted by the Institute of Economics at the Chinese Academy of Social Sciences. Each survey was made independently, and covers 13 cities (in 6 provinces) in 1999 and 28 cities (in 12 provinces) in 2002, respectively. Since the main interest here is to compare the relative wages change of rural migrants with those of their urban counterparts, only rural migrants and urban residents in the same cities should be tracked over time. Hence, 11 of those 13 and 28 cities which are covered by both surveys, have been chosen as the database. Those cities include 5 large cities (Beijing, Shenyang, Zhengzhou, Chengdu and Lanzhou) and 6 medium and small cities (Jinzhou, Xuzhou, Kaifeng, Pingdingshan, Nanchong and Pingliang), all distributed in different regions of China. Both CIDS in 1999 and 2002 cover a large sample of households for urban residents as well as rural migrants, which were randomly sampled from their corresponding population in each city. The questionnaires for urban residents and rural migrants are generally comparable, and the data collected from those surveys contain detailed employment and earning information from the sample. Meanwhile, some migration information such as year of arrival in the host city and time spent in the host city for rural migrants, which are crucial to the economic assimilation model, are also included. 2 Distributions of the sample for rural migrants and urban residents in CIDS across 11 cities are reported in Appendix Table 10 and 11, respectively. As shown in Appendix Tables 10 and 11, since the CIDS enlarged the number of surveyed cities from 13 (in 6 provinces) in 1999 to 28 (in 12 provinces), the sample size for both rural and urban workers in each city has changed greatly across the two surveyed years. In order to get rid of the impact of this change, the data are adjusted by using their host cities weights. Furthermore, the relevant sample is trimmed due to the research purpose of this study. More specifically, individuals are defined as those employed aged between 16 and 65. The sample for urban workers is restricted to those who hold non-agricultural Hukou and for rural migrants 2 As a repeated cross-sectional data set, CIDS allows for the control of both years since migration and cohort effects when there is a need to take into account both quality of the cohort and assimilation effects. Page 13 of 53
to those who arrived in the host city at over 16 years old. 3 Additionally, all observations with missing information (including missing values for earnings, hours worked and other important information) are deleted. Thus, the final dataset contains 4786 and 2131 observations for urban residents and 789 and 1232 observations for rural migrants in 1999 and 2002, respectively. Although CIDS in 1999 and 2002 are suitable for the study, there are still a few limitations which need to be pointed out. First, there might exist some biased sampling problem in CIDS. According to the CIDS sampling strategy, rural migrants are chosen through residential committees, which have information on migrants registered as temporary residents as well as those who are unregistered but living in the community (i.e., those operating small businesses). This neglects migrant workers living on construction sites and in factory dormitories, who normally do not register with any residential committees, and therefore generates some selection bias. The evidence for this concern is that the proposition of married migrant couples and owners of private firms or self-employed for rural migrants, calculated from the sample, seems over-presented compared with those calculated from the 2000 Census data. 4 Second, CIDS did not include the question as to how many hours urban workers worked in 1999, though such information appears in all other questionnaires. As this study focuses on rural and urban workers wage instead of income, the number of hours worked is essential for calculating workers hourly wage. To solve this problem, the number of hours urban workers worked annually in 1999 is estimated by a simple labour supply model with urban workers data in 2002. Since there is a strong assumption behind this estimation - the regression structure doesn t change for urban workers across years, the hourly wage of urban workers in 1999 might be over- or under- estimated. 5 Third, whether the four-year-time-span covered by CIDS (i.e. from 1999 to 2002) is long enough for studying the dynamic wage gap between rural migrants and urban residents is another concern. Given that there is no better trans-temporal data available and that the dramatic increase in rural- 3 Note that 3 observations, which were recorded as having 15 years-of-schooling for migrants workers, are dropped due to the limited observations. 4 CIDS has a larger proportion of married (range from 87 to 93 percent) and self-employed rural migrants (accounting for 40 to 50 percent of the total sample) than common belief. According to the 2000 Census data, the proportion of married in the 11 cities as in CIDS is only 70 percent. Since the 2000 Census did not include questions as to whether a migrant is working as a wage or salary earner or as self-employed, the actual occupational distribution for migrant population is unknown in Chinese cities. Even though, the proportion of self-employed is too high according to common sense, the only explanation for this phenomenon is that married migrant couples are more likely to rent a room in the urban residents communities, which reflects the selection bias made in CIDS 5 For example, if the actual labour supply for urban residents increases (decreases) from 1999 to 2002, the hours worked in 1999 was overestimated (underestimated). As a result, the hourly wages for urban residents in 1999 are underestimated (overestimated). Page 14 of 53
urban migration took place around 2000s, the CIDS would be the only choice for carrying out such a study in current time. The descriptive statistics of urban residents and rural migrants by men and women in 1999 and 2002 are shown in Table 1. 6 Some interesting facts, such as characteristics of migrant workers, changes in initial wages of rural and urban workers, as well as workers mobility across occupations and industries can be summarised as below. First, rural migrants were much younger and less educated than their urban counterparts, and they came into cities only recently. As shown in Table 1, the migrants in the sample were on average around 7 years younger and 3 to 4 years less educated than their urban counterparts. On average, male migrants, had resided in the host cities for 5.8 years in 1999 and 7.8 years in 2002, respectively, whereas female migrants had one year less of duration in cities (4.6 years in 1999 and 6.1 years in 2002). Around 50 percent of male migrants and 66 percent of female migrants arrived after 1995. Second, rural migrants earned significantly less than their urban counterparts in terms of both annual and hourly earnings, and the gaps between the two groups widened across the years. In 1999, the annual earnings for urban male workers was 8152 yuan,on average, while the amount for male migrants is 6412 yuan, which accounts for 78.7 percent of that of urban workers. By 2002, the earning (in 1999 yuan) of urban workers increase by roughly 70 percentage to reach 13781 yuan, while the number for migrant men increases by 43 percentage point to 9163 yuan. Hence, the ratio of rural migrants over urban residents has dropped from 78.7 percent in 1999 to 66.5 percent in 2002. Although female migrants earnings were generally less than for migrant men, the gap for annual earnings between them and their urban counterparts did not drop as much as for migrant men. The ratio between women migrants and urban workers decreased by only 2.6 percentage points from 78.7 percent to 76.1 percent. When taking into account the changes in hours worked, the hourly wage gap is further enlarged (drop from 66.9 percent to 44.4 percent) since migrant workers work 1.5 times as much as urban workers per annum. Third, there is significant segregation between rural and urban workers in ownership, industry and occupation. For example, urban workers were mainly employed in the state-owned sector (65.4 percent in 2002), whereas most rural migrants worked in urban private and individual sector (72 percent in 2002). In terms of industry distribution, a round 35-37 percent of urban male workers were 6 Details concerning the definition for the key variables appear in the Data Appendix. Page 15 of 53
based in manufactoring industry, while 43-45 percent of rural migrants are concentrated in Wholesale, Retail trade and Restaurant, and another 19-23 concentrated in the Social Service sector. It turns out that rural migrants were more likely to work in Tertiary Industry than their urban counterparts. As for occupational distribution, most rural migrant men are either operating a small private enterprise, are self-employed (accounting for 41.8-51.7 percent) or employed as unskilled workers (accounting for 36.8-32.2 percent). On the other hand, urban workers were more likely to be high level officers, professionals or clerks. These findings may suggest rural migrants and urban workers are segregated into two labour markets, and that rural migrants can only access limited jobs in the host city. However, this significant segregation between rural and urban workers cannot be sustainable in the long term due to along with the market integration reform. It seems that more urban workers and rural migrants were sharing similar jobs across the two surveyed years. More specifically, from 1999 to 2002, one of the significant changes was that the frequency of urban workers in the state-owned sector fell from 89.5 to 65.4 percent, with large scale employment for urban male workers moving to the private sector, due to state-owned enterprises restructuring from the mid 1990s. In addition, more urban workers entered into Tertiary Industry and worked as unskilled workers. These facts reveal that unskilled urban workers, who had been gradually released from the state sector and exposed to labour market competitiveness and were likely to be competing with rural migrants. The dynamic statistics of labour market outcomes according to year of arrival (i.e. cohort) in host cities for men and women are presented in Table 2, which reveals a broad picture about assimilation. In this table, the change of average earnings, hours worked and years of schooling are compared from two aspects: across-cohort difference in each year or within cohort difference across years. The top panel shows the dynamic change for migrant men in 1999 and 2002, while the bottom panel indicates the same women. The table disaggregates migrants by four cohort subgroups. It shows that, both annual earnings and hourly wage appear to be higher for earlier than later cohorts in each year, which indicates the most assimilated migrants will be those who have been in the host city the longest. Also, when tracking over the same cohort across years (i.e. within-cohort growth), the annual earnings for each cohort has increased, except for the cohort that arrived before 1989 for men. However, in terms of hourly wage, however, the 1990-94 and before 1989 cohort for men and the earliest cohort for women experienced a decrease in hourly wage whill wages for the others cohorts increased across the years. Table 2 also presents the quality differences (i.e. years of schooling) across cohorts. It shows the education level tends to be a bit higher for the more recent migrant cohort for Page 16 of 53
both men and women; and the years of schooling seems to have a slight increase for migrant men who arrived before 1995, while for migrant women, the years of schooling overall drops for all cohorts. The previous discussion focuses mainly on the difference in the mean wage and its change over time. However, an analysis of the mean wage may not completely characterize the differences between urban residents and rural migrants in whole distributions. To understand the changes in the wage gap across years by quintiles, the real wage 7 distribution for rural migrants and urban workers is illustrated in Figure 1, where the density of wage distributions for rural migrants and urban workers in 1999 and 2002 by men and women are separately plotted. The figure shows that, both for men and women, wage distributions for rural migrants are displaced leftward with respect to those of urban residents, indicating non negligible wage differentials between the two groups. This implies migrants became further concentrated at the bottom tail of wage distribution for urban residents during the period. Figure 2 shows the changes in real hourly wage gap between urban and migrant workers from 1999 to 2002 separated by male and female, which can help in understanding changes in the wage gap across distributions. In the graph, each point along the lines indicates the wage increase of urban residents minus the wage growth for rural migrants at the same percentile point in their own wage distribution. It displays that, for both men and women, the wage gap widened at various points in the wage distribution over time. The wage gaps also increase consistently across the distributions of wages. Moreover, the widening in wage gap for men is more severe than for women through the entire wage distribution. Thus, the analysis for the wage distributions and wage gaps concludes the wage distribution has changed across years; more specifically, relative wage changes from 1999 to 2002 seem to favour more-skilled workers. As a result, the disadvantage situation for unskilled migrant workers may deteriorate relative to their urban counterparts. IV Dynamic Wage Decomposition So far, the descriptive statistical analysis shows significant earnings differentials between urban residents and rural migrants and a large widening of the raw wage gap between the two groups across years. For example, in 1999, migrant men earned 66.9 percent of urban male workers in terms of hourly earnings, whereas in 2002 they only earn 44.4 percent. It is not clear why the situation of rural migrants worsens or what factors contributed to the widening of the wage gap. Before embarking on 7 Log hourly wages have been converted to 1999 yuan using CPI. Page 17 of 53
addressing the wage assimilation issue in this section, these questions will be answered by adopting the decomposition approach of Smith and Welch (1989). Decomposition Method In the spirit of Oaxaca s (1973) one period decomposition, Smith and Welch (1989) developed an alternative approach to examine how the changes in means and returns to the explanatory variables combined to affect changes in the wage gap over time. This study adopts their decomposition method to analyze the attributes of some key variables to the change in the wage gap between migrants and urban workers from 1999 to 2002 in China s urban labour market. The change in the wage ratio of rural migrants relative to urban residents proposed by Smith and Welch (1989) can be expressed by the following equation: (lnw u 02 lnw r 02) (lnw u 99 lnw r 99) = ( β u 02X u 02 β r 02X r 02) ( β u 99X u 99 β r 99 Xr 99) = β u 99 [(Xu 02 X r 02) (X u 99 X r 99)] (i) (5) +( β u 02 β r 02)(X r 02 X r 99) +(X u 02 X r 02)( β u 02 β u 99) (ii) (iii) +X r 99[( β u 02 β r 02) ( β u 99 β r 99)], (iv) where, the superscripts u and r refer to urban residents and rural migrants, respectively; subscripts 02 mean year 2002 and 99 is year 1999; lnw indicates the mean log earnings; β presents the estimated coefficients from wage regression; and X = is a vector of means for personal endowments. This approach decomposes the raw change of the wage gap between urban residents and rural migrants into the following four components. The first term (i) in Equation (5) measures the predicted change in wage ratio due to change in mean characteristics, valued at 1999 urban residents parameter values. For example, if difference in years of schooling between urban workers and rural migrants diminished over time, the wage gap between them would narrow. The change due to variation in mean attributes is regarded as the main effect (Smith and Welch, 1989). The second term (ii) represents identity interaction (i.e. identity of urban residents or rural migrants), which measures the additional change in the wage gap predicted by the change in the attributes of rural migrants, fixing the difference of returns between urban residents and rural migrants in 2002. For example, if Page 18 of 53