Migration and Income Mobility of Rural Households in China a. Chong-En Bai, b1. Wenkai Sun, b2. Xianghong Wang a. School of Economics and Management, Tsinghua University b. School of Economics, Renmin University of China Abstract: This paper examines the dynamics of, and the impact of rural migration on income mobility in rural China. We use income transition matrix to measure income mobility and adopt some econometric models to estimate the effect of migration on income mobility. Some interesting findings are obtained. First, annual income inequality in rural China had been increasing before 2002 but then decreased after 2003, while income mobility was relatively stable from 1986 to 2006. Second, both the descriptive statistics and econometric analyses verified that migration played an important role in income mobility. Households with migrant workers have greater upward mobility. Third, households of interior provinces and poor households of coastal provinces have higher tendency to have migrant workers. Migrant workers account for a higher proportion of the labor force in the households of interior provinces. The direction of the migration is mostly toward regions with higher income levels. Our findings help to explain how migration affects income mobility and income inequality. Key Words: Income Mobility; Income Inequality; Rural Migration Chong-En Bai: Department of Economics, School of Economics and Management, Tsinghua University, Beijing, PRC, 100084. (tel: 86-10-62773183, email: baichn@sem.tsinghua.edu.cn); Wenkai Sun: School of Economics, Renmin University of China (email: sunwk@ruc.edu.cn); Xianghong Wang: School of Economics, Renmin University of China (email: shwang06g@gmail.com).
Migration and Income Mobility of Rural Households in China 1. Introduction Many studies have found that China s income inequality, both in rural and urban areas, have been widening during its rapid economic development since 1978 (Wang and Fan, 2005; Lipton and Zhang, 2005; e.g.). Our observation, however, indicates that the annual and permanent inequality in rural China have started to decrease since 2003. This paper tries to examine the causes of this changing trend through rural migration and income mobility in rural China. This is made possible by a survey conducted by the Ministry of Agriculture in rural China between 1986 and 2006. Income mobility has been increasingly recognized as an important indicator in the analysis of income inequality and development issues in recent years (Solon, 1992; Dercon, 2000; Khor and Pencavel, 2006). Milton Friedman first discussed it in 1962 (Friedman, 1962). It measures inter-temporal changes of households or individuals relative income positions in the economy. Greater income mobility indicates that the economy gives the low-income individuals more opportunities to move up on the income ladder and thus helps to reduce long-run permanent income inequality. Compared with the usual annual income inequality approach, income mobility helps to study the dynamics of inequality and the long-run income inequality issue. Studies on income mobility have generally examined intergenerational mobility (Chadwick and Solon, 2002; Björklund and Chadwick, 2003), determinants of income mobility (Woolard and Klasen, 2005), poverty traps in a country (Solon, 1992;
Dercon, 2000; Olga, 2000; etc.), gender differences (Gang et al, 2002), and cross-country comparisons ( Khor and Pencavel, 2006). In China, studies of income mobility have focused on intra-generational mobility. Yin et al (2006) investigated urban income mobility between 1991 and 2002, and found decreasing mobility over the years. Khor and Pencavel (2006) compared income mobility of urban individuals in China and the United States between 1991 and 1995. They found that China s mobility was greater and that poorer individuals in urban China had benefited more from the economic growth than their wealthier counterpart. Zhang, Mi, Huang (2007) found that, during 1987 to 2002, the probability for the poorest 25 % to move up to higher income status had increased but the upward mobility of those of the middle-income had gradually become stagnant. In the latest study, Ding and Wang (2008) used the data from the China Health and Nutrition Survey (CHNS) to measure absolute income mobility of household in China. They found that household income mobility in China remained at a high level from 1989 to 2000. Since urban income is generally higher than rural income, it is reasonable to expect migration from the rural areas to the urban areas to help increase upward income mobility of the poor and reduce income inequality. Based on China Statistical Yearbook of 2007, average income in the urban areas is three times higher than in the rural areas. There have been many studies on migrant workers and the role of migration in improving farmer s income in rural China. Cai et al (2001) and Cai (2007) found that the registration system reform had given the farmers much more employment opportunities in the urban areas. Giles (2006) and Sun et al (2007) found
that labor mobility had played an important role in improving the living standards of migrant workers. To our knowledge, however, no research has directly studied the impact of labor mobility on income mobility in China. This paper studies the income mobility of rural households in China between 1986 and 2006, with a focus on the impact of rural migration on income mobility. We use a dataset obtained from the Survey Department of the Research Center on the Rural Economy (RCRE), Ministry of Agriculture, covering 6 provinces from 1986 to 2002 and 23 provinces from 2003 to 2006. We first use several methods, including income transition matrix and correlation coefficients, to measure income mobility within certain periods. Then we use conditional transition matrix and Multinomial Logit model (MLM) to estimate the effects of migration on rural households income mobility. The impact of rural migration on income mobility has strong policy implications and has not been studied much in detail in China. The data set we use has a wide and long coverage, allowing us to examine both the dynamics of income mobility with the most updated data as well as to make regional comparisons on income mobility. Our study has obtained the following general findings: Since 1986, income mobility has stayed relatively high over the years. From 2003 to 2006, both permanent and annual income inequality have decreased among rural households. Migrant workers within the rural households are important in explaining this trend. The lower the households initial relative income ranking, the higher is the probability for the households to have migrant workers. Households with migrant workers have greater upward income
mobility. In addition, there are obvious differences between coastal and interior provinces. 1 The interior provinces have higher income growth rates and more households with migrant workers. This leads to decreasing income gap between interior and coastal provinces and boosts income mobility. The rest of this paper proceeds as follow: Section 2 introduces the data set and methods; Section 3 summarizes all descriptive statistics; Section 4 presents analyses of income mobility with the whole data set, measured in income transition matrix and statistical indices based on the transition matrix; Section 5 reports the econometric results for analyzing the impact of migrant worker and income mobility. Section 6 concludes the paper. 2. Data and Methods 2.1 Data source The panel data set used in this paper is from the survey conducted by the Research Center on the Rural Economy (RCRE), Ministry of Agriculture. This survey was started in 1986, and collected comprehensive household information (including household types, demographic information of household members, household production data, asset, and credit). For the period between 1986 and 2002, excluding 1992 and 1994 when the survey was not conducted, our data set contains six provinces including Liaoning, Shandong, Hubei, Guangdong, Yunnan, and Gansu. For the period between 2003 and 2006, our data set includes 23 1 The costal provinces include Beijing, Tianjin, Liaoning, Jiangsu, Zhejiang, Fijian, Guangdong, Shandong, Shanghai and Hainan.
provinces and more variables. The dynamics of income mobility is best studied with data set that has continuous annual observations. Because of the missing two years and the survey changes since 2003, we divide the analysis into three continuous time frames: 1986-1991, 1995-2002 and 2003-2006. The three periods contain respectively 5416, 4098, and 12889 households. If we only include households that stayed in all these three periods, we are left with 2095 households. When we analyze the impact of migration, we focus on the third period that included 12889 households in 23 provinces in order to have a better representation of the greater part of rural China. Since the income information was at household level, we need to convert household income into per capita income. We mainly use average income that equals the net income over the number of household members. 2 To compare the income in different years, we transform nominal income to real income based on year 2003 using rural consumer price index of each province. 2.2 Methods Income mobility reflects the degree of opportunity equality among people. Our analysis focuses on income mobility at the aggregate and regional levels. There are many methods to measure income mobility, such as transition matrix (Yin et al, 2006), 2 In previous studies, such as Woolard and Klasen (2005), May, Carter and Posel (1995) and Roberts (2000), Adult Equivalence (AE) income is also used to replace average income. Considering that household income is affected by various types of household members, AE income equation is given as eq _ inc it tot _ incit = adult + minor /2 it it, where i denotes household, t is year, eq_inc is AE income of household, tot_inc is the total net income of household, adult is the number of adult and minor is the number of minor member. We calculated income mobility using average income and AE income respectively. We report the results using average income because there is no significant difference between the results of these two measures.
absolute mobility (Fields and Ok, 1996, 1999), correlation coefficients (Khor and Pencavel, 2006), and Shorrock s rigidity index(woolard and Klasen, 2005). We focus on the widely used income transition matrix and indices derived from the matrix (see Khor and Pencavel, 2006; Yin et al, 2006) to measure income mobility. At the aggregate level, all households in the data set are pooled together and are divided into 5 equally sized income groups (or quintiles) endogenously determined for each year. The income transition matrix measures the probability for individuals in each quintile group at the base year to move to any of the five quintiles at the end year of the studied period. Let P be a matrix of 5 by 5 transitions, the ij-th element of P, P ij, is the percentage of those in the i-th income quintile at the base year that moved to the j-th quintile at the end year of the studied period. The households who have moved from one income group to another between the two periods will be considered "mobile". Those who remain in their original income group will be considered "immobile". If every cell of the matrix is 0.2, it is completely mobile, which indicates that each group in the initial stage has an equal opportunity to move to any of the five ranking groups in the end stage. The advantage of the transition matrix is that it can nicely summarize mobility at various points in the income distribution, which is harder to gauge with a single index. For comparisons of general mobility, we use some derived indices. These indices include Average Quintile Move, Quintile Immobility Ratio, and the fraction of people who move by one quintile. The Average Quintile Move (AQM) is defined as 1 5 5 5 j= 1 k= 1 ( j k ) p jk. This index reflects the general mobility level. The larger the
AQM, the higher the mobility level is. Another index, Quintile Immobility Ratio (QIR), is the fraction of households who remain in the same income quintile (1/ 5) ( p ). This index reflects the tendency for individuals in one group to j= 1,...,5 jj stay in their original income level. Therefore, the larger the QIR, the lower the mobility level is. The last derived index is the fraction of those who remain in the same quintile plus the fraction of those who move only by one quintile. This measurement is similar to QIR and reflects the degree of immobility. To analyze the impact of migration on income mobility, we also compute conditional transition matrix with its elements defined as follows: Pijm PT ( ij mi ) = (1) Pm ( ) i Where PT ( ij m i ) represents the probability for a household to move from level i to level j, given that the household has migrant workers in any of the years between 2003 and 2006. To estimate the magnitude of the impact of migration on income mobility, we performed Multinomial Logit and OLS analysis with income ranking change and income growth rate as the dependent variable respectively. 3. Summary of statistics In this section, we describe some basic statistics of the dataset, especially on migration trend and income inequality in rural China. These are presented in Table 1, 2, 3, 4 and Figure 1, 2, 3. To study the impact of migration on income mobility in rural China, we need
representative samples and detailed information about migration. We choose to focus more on the years between 2003 and 2006 for the analysis. Descriptive statistics of the whole sample from 2003 to 2006 are summarized in Table 1. These households stayed in all four years so that the inequality indices to be computed later on are comparable in different years, and we can compute income mobility. During this period, farmer s net income rapidly increased from 13,161 RMB in 2003 to 18,060 RMB in 2006 with a 11.13% annual growth rate. The three years growth rates are 12.91%, 7.70% and 12.85% respectively. Per capita net income shows a similar trend with the household income. From 2003 to 2006, the per capita net income increased 35.67% with annual growth rates of 13.01%, 9.18% and 9.95%. Other demographic variables, household size, labor force proportion, male labor proportion, proportion of skilled labor, official household and household heads education levels are stable during this period. 3.1 Migration trend It is worthwhile to note that labor income of migrant workers significantly increased from 4,271 RMB in 2003 to 9,938 RMB in 2006 (see Table 1). The proportion of net income that can be accounted for by migration is rapidly increased from 32.45% in 2003 to 55.03% in 2006. 3 According to Table 2, the proportion of households with migrant worker grew from 57.97% in 2003 to 63.46% in 2006. If we 3 The definition for migrant worker in the data set is different before and after 2002. Before 2002, it refers to someone who has worked out of town year-round for a long time. But after 2003, it refers to all who have worked as a migrant worker at any time of the year. Due to this reason, percentages of households with migrant workers before 2002 are all under 20%, much lower than the percentages after 2003. Therefore, we do not make comparisons between the numbers before and after 2002.
calculate the proportion of households with migrant worker at any year during this period, this number is 76.37%, which is quite remarkable. There are evident differences between the coastal and interior provinces. Fist, average income level in the rural areas in interior provinces is lower than in the coastal provinces. From 2003 to 2006, the average net income levels are respectively 10810.9, 12423.55, 13461.5 and 15308.78 RMB in interior provinces, and 20473.9, 22439.5, 23913.1 and 26618.9 RMB in coastal provinces. Second, Figure 1 also shows that the proportion of the net income and total income that are accounted by migration income in interior provinces (32.98% in 2003 and 56.54% in 2006) is larger than that in coastal provinces (31.59% in 2003, and 53.26% in 2006), while they have similarly increasing trend during this period. From Table 2, we can see that the proportion of households with migrant workers in the interior provinces is also higher than that in the coastal provinces. 3.2 Trend of income inequality Many studies have calculated annual income inequality in China, including Ravallion and Chen (2004), Lipton and Zhang (2005), Li and Yue (2004), and Wang and Fan (2005). Because they used different datasets, their results are generally different. For example, Wang and Fan (2005) suggested that income inequality trend in rural China matched the Kuznets curve without reaching the peak. This means that inequality had increased with time. Similarly, both Ravallion and Chen (2004), and Lipton and Zhang (2005) pointed out that China already became the most unequal
country with increasing income inequality. Li and Yue (2004) found that income inequality in rural area decreased during 1995-2001. Due to the data limitation, these studies mainly focused on inequality before 2003. In this section, we use a wider range of household survey data to calculate the inequality in rural China in recent years. We first use the Gini coefficient to measure income inequality. While we compute the Gini coefficient for annual income to measure total income inequality, we also compute the Gini coefficient with average income over four or five years to identify permanent income inequality. These results are given in Table 3 and 4 respectively. Per capita net income reported in Table 3 and 4 is the net income of household divided by the total number of family members. Due to the difference in the number of provinces in the data set before and after 2003, we also calculate inequality with the merged data set with the 2095 households who remained in all years. The results are similar to the larger sample. According to Table 3, Table 4, Figure 2, and Figure 3, annual income inequality increased from 0.4121 in 1986 to 0.5612 in 2002, with some fluctuation. After 2003, the Gini coefficient dropped from 0.47 in 2003 to 0.4201 in 2006. This trend also holds true when we use the smaller dataset merged by households. Table 4 and Figure 3 indicate that permanent income inequality is smaller than the corresponding annual income inequality. We can also see that the Gini coefficient decreased during 2003-2006 to 0.3695 (0.4296 if merged data) after an increasing trend during 1998-2002 and reached to 0.4078 (0.4605 if merged data). The decreases of annual
and permanent inequality are good news. 4 These suggest that income mobility might have been high duringthese periods so that permanent inequality was reduced. 4. Income mobility There are different ways to measure income mobility, including correlation coefficients between different years income, income transition matrix within certain period, and the absolute mobility value and decomposition suggested by Fields and Ok (1996, 1999). Following Khor and Pencavel (2006), we first use the correlation coefficient of different years' per capita income levels. As explained earlier, we group the whole data sample into three periods: 1986-1991, 1995-2002, and 2003-2006. The three periods have 5416, 4098 and 12889 households respectively. According to results reported in Table 5, 6 and 7, during first period, the correlation coefficients are larger than those in the next two periods. For example, the coefficient between the income of 1986 and the income of 1987 is 0.494, while it is 0.6487 between 1995 and 1996, and 0.6905 between 2003 and 2004. This means that income mobility may be decreasing with time. The alternative method is to calculate the correlation coefficient of income growth rate and their income ranking. These results are presented in Table 8. Households with low initial income ranking have higher income growth rate. We next focus on the method of income transition matrix. Since correlation coefficient is not completely mapped onto individual or household income ranking 4 We alsoe used alternative indices to measure income inequality, such as standard variance of logarithmic income, and the Coefficient of Variance. We get the same results on inequality trend.
change, it is not suitable for analyzing the determinants of income mobility. Income transition matrix measures the probability for different individuals or households to move up or down to certain income level. Table 9 reports the income mobility measured by per capita income transition matrices computed for the three periods of 1986-1991, 1995-2002, and 2003-2006. To get a reasonable comparison, we compute the transition matrix within four years in each period (Table 9). As mentioned before, we classify these households by income quintiles from 1 (the bottom quintile and the lowest income class) to 5 (the top quintile and the highest income class) in all three periods, with an equal number of households in each quintile. Table 9 reports the five-by-five income transition matrices in which each element p ij is the fraction of households in quintile i in the base year who occupy quintile j in the end year. As shown in Table 9, 54.1% of the poorest fifth of the households in 1986 remained in the same quintile in 1989. 51.9% of the richest fifth in 1986 remained in the same quintile in 1989. In comparison, the two indices are 51.8% and 49.6% for 1988-1991, 53.8% and 58.3% for 1995-1998, 59.3% and 63.5% for 1999-2002 and 52.4% and 56.9% for 2003-2006. To help compare income mobility across different periods, we calculate three summary indices of the income mobility derived from the transition matrices: AQM, QIR, and the proportion that remained in the same quintile plus the proportion that move upward by only one quintile. These indices are given in Table 10. On the whole, the three indices suggest that income mobility between 1999 and 2002 is noticeably smaller than in other periods. For example, AQM and QIR for 1999-2002 are 0.74 and
0.469. During 1986-1989 and 2003-2006, these indices are 0.926 and 0.385, 0.883 and 0.405. It should be noted that the above transition matrices are created for three time periods with different sample sizes. The results hold if and only if the sample is randomly selected. To make more accurate cross-period comparisons, we need to merge the data across periods by households. Table 11 and Table 12 report the income transition matrices and the mobility indices respectively for the smaller sample that includes only those households that stayed in all three periods (n=2095). The merge allows us to increase the length of the period to examine income mobility from the earlier four-year time frame to the current six-year time frame. General mobility level in the three periods seems to be stable. For example, the AQMs for the three periods are 0.8315, 0.8248 and 0.8286 respectively. 5. Migration and income mobility 5.1 Impact of migration This section examines the impact of rural migration on rural households income mobility. We only report the result of last period because we have a larger sample and more variables in the data set between 2003 and 2006. We approach this issue from a few aspects. First, we compare all households who had migrant workers with those households who did not have migrant workers to examine their differences in income growth rates and income mobility. Second, we use conditional income transition matrix to measure households income mobility given that their households had migrant workers. Finally, we use Multinomial Logit model to examine the magnitude
of migration s impact on the probability of upward or downward mobility. Table 13 reports the average income levels and income growth rates for households with or without migrant workers in the period from 2003 to 2006. This growth rate is 50% for households with migrant workers, much higher than the 37% for households without migrant workers. In 2003, the average income levels of these two types of households are 3332.87 RMB and 3322.9 RMB respectively with no significant difference (p=0.463). In 2006, however, average income of households with migrant workers (4626.45 RMB) is significantly higher than those without migrant workers (4169.12RMB; the difference is 10.97% with p< 0.01). Table 14 computes the conditional income transition matrix given that household had migrant workers. Comparing Table 14 with the results of unconditional matrices in Tables 9, migration significantly increased the probability of moving up in income ranking. For the poorest fifth households, only 47.80% of the households with migrant workers remained in the same quintile, compared with 52.38% for the whole sample in Table 9. Still for the poorest fifth households, 24.77%, 14.74%, 7.80% and 4.88% of those with migrant workers who moved up to the second, third, fourth and fifth quintile respectively in the conditional transition matrix, while these fractions are 23.15%, 13.14%, 7.25% and 4.07% for the unconditional matrices in Table 9. For the households ranked in second, third and fourth quintile, migration enhances their income ranking in the same way. The comparison indicates that the poor households with migrant workers have higher opportunity to improve their income situation. For the richest fifth households, we did not find that migrant workers provided an
advantage for them. These results show that migrant workers play an important role in improving the poor households situation. We next conduct a Multinomial Logit regression to estimate the impact of migration on income mobility. Many factors might influence the absolute and relative movement of household income. We focus on migrant workers role while controlling for other variables. Previous studies have suggested some explanatory variables to consider. Zhang, Mi and Huang (2007) considered household size, labor force ratio, human capital of household, physical assets, etc. Woolard and Klasen (2005) introduced time difference terms of demographic variables as independent variables. We take these variables into consideration in our regression. In addition to the dummy variable for whether a household had migrant workers during the studied period, independent variables we controlled for also include initial income ranking, basic household characteristics, population, labor force ratio, human capital, regional dummy variables, and some first difference variables. The first-difference variables include changes in human capital as well as demographic composition. The dependent variables for the Multinomial Logit model are the probability for the households to move up or down by certain quintile steps. Since all households are ranked in five quintile levels, the best mobility is to move from rank 1 to 5, with a change of 4; the worst mobility is from 5 to 1, with a change of -4. There are totally eight different position-changing possibilities. The base outcome is 0, which means that the households are kept in their initial positions. Table 16 presents Multinomial Logit regression results for eight different income
ranking change. Our main focus is on the impact of migration. In general, controlling for other factors, migration significantly increases upward mobility (with significant 99% confidence level). However, migration does not reduce downward mobility. This means that in general poor households benefit more from migration job opportunity. We compute the marginal effect of migration on income mobility. Controlling for other factors, migration increases 6.7%the probability to move up by one quintile level. It increases 1.82%, 0.28% and 0.01% the probability to move up by two, three and four ranking steps. We also report other determinants that may have affected income mobility. The first explanatory variable in Table 16 is the households initial income ranking during the base year, which is the households income ranking percentile in 2003. The regression results indicate that the higher the initial income ranking, the lower the household current ranking is. Next, households with village officials in a household have no impact in income mobility. Party members, the party member status has no effect in income ranking change in most regressions but it reduces the probability to be in the lowest ranking group (p<0.01). Minority families have a higher probability of belonging to the lowest ranking group. Being a minority increases downward mobility and decreases upward mobility. The sizes of the households and the growth of their sizes also increase downward mobility. This result is significant in most regressions with high significance levels. The growth of labor force proportion of family members, the proportion of skilled
labors and its growth, the average education level of the household and its growth, and residence in the coastal provinces all increase upward mobility. Notably, women as household heads significantly increase downward mobility in income ranking. In addition, even though we find that households with members of poor health tend to have migrant workers later on in section 5.2, they still increase downward mobility in income ranking. We also use Ordered Logit to do the same regressions as above to test robustness of the model. The results are presented in Table 17 and they support the conclusion of Table 16. Furthermore, we use OLS to estimate the determinants of income growth rate. The results are reported in Table 15. Migration increases income growth rate by 8% while other factors are controlled for. The results of Table 16 and Table 15 validate each other. All results suggest that migrant workers bring more upward mobility opportunities to the poor households. As shown in our earlier summary of statistics for the data set, income from migrant workers has accounted for a bigger share of the household s net income in recent years. Therefore, we believe that the migration work opportunity is playing an important role in improving income distribution in rural China. 5.2 Characteristics of households with migrant workers To understand why migration affects income mobility, we conduct regional comparisons in characteristics of households with or without migrant workers. The interior provinces have a higher proportion of households with migrant workers
(77.57%) than the coastal provinces (72.61%). It is reasonable to predict that interior provinces have a higher income mobility. We will also examine where the migrant workers tend to choose to work. Table 18 reports the results of income growth in coastal and interior provinces. If the two types of households are pooled together, in all time periods, farmers income levels in the coastal provinces (5379.35 RMB in 2003 and 6984.40 RMB in 2006) are significantly higher than those in interior provinces (2671.78 RMB in 2003 and 3725.73 RMB in 2006) ( p<0.001). Income growth, however, is faster in the interior provinces than in the coastal provinces. The growth rates are 29.84% in coastal provinces and 39.45% in interior provinces respectively. This implies that the gap between income levels of interior and coastal provinces has narrowed. Table 18 shows that income growth rate of households with migrant workers (31.81% and 43.17% in coastal and interior provinces respectively) is much higher than that of households without migrant workers (24.23% and 26.27% in coastal and interior provinces) during the same period. The growth rate of households with migrant workers is also higher in interior provinces than in coastal provinces, while the growth rate of households without migrant workers are not distinctively different between the coastal and the interior provinces. Table 19 and 20 present the transition matrices and their derived indicators of interior and coastal provinces. The results show higher income mobility in interior provinces. For example, in the coastal provinces, 52.31% of the poorest fifth of the households in 2003 remained in the same quintile in 2006; 55.82% of the richest fifth
in 2003 remained in the same quintile in 2006. In the interior provinces, these two indices are reduced to 49.08% and 47.98% respectively. This supports our conjecture that migrant workers have greater income mobility and experience higher income growth in the interior provinces. We next examine what types of households have a tendency to have migrant workers. In Tables 21, we adopt a Logit model to analyze this. We conduct the same regression for the whole sample and the types of provinces separately. The dependent variable is a dummy variable that takes 1 when a household had migrant worker during 2003-2006 and 0 otherwise. We introduce some demographic variables as independent variables as suggested in Deng and Gustafson (2007). In addition, we produce two new variables: the household s income ranking in the total sample and its income ranking in their village. 5 The reason we choose these variables is that people may tend to compare themselves to others and decide whether they need to make more money by working more so they can consume more conspicuous goods (Luttmer, 2005; Brown, Bulte and Zhang, 2009). We also include coastal province as a dummy independent variable. According to Table 21, household size and its growth, labor force ratio and its increase, and male labor force ratio all have positive effect on migration. It seems counter-intuitive that households with members of poor health have a higher probability to have migrant workers. This might be because these households are poorer and need to increase their income through migrant workers to cover their 5 Income ranking in total sample is defined as the ranking number over the number of sample size and income ranking in their village is defined as the household income over the maximum household income in the village.
health expenses. The above results are consistent in four regression models in Table 21. Households with village official and party members have a lower probability to have migrant workers. In coastal provinces, the poor households in 2003 have a higher tendency to have migrant workers, In interior provinces, households with a higher relative income ranking within their village have a higher probability to migrate, but this is opposite in the coastal provinces. Finally, we examine the destinations of migrant workers in households in coastal and interior provinces from 2003 to 2006. The results are summarized in Table 22. Two important features stand out. First, the proportion of labor force accounted for by migrant workers is larger in the interior provinces than in the coastal provinces for all the years. This proportion also increases faster in the interior provinces than in the coastal provinces. For example, the proportion for the coastal provinces is 29.61% in 2003 and 32.26% in 2006, while they are 33.13% in 2003 and 37.68% 2006 for the interior provinces. The difference between coastal and interior provinces in the average number of migrant workers of within a household might be reason that income inequality has decreased in the rural areas. Second, households in coastal provinces are more likely to migrate to their home province. In contrast, migrant workers in interior provinces are more likely to migrate to coastal provinces. For example, the proportion of labor force in coastal provinces that have workers who have migrated to other provinces is 3.36% in 2003 and 4.03% in 2006, while this proportion is 14.87% in 2003 and 16.82% in 2006 for the interior provinces. This might be another contribution to increased income mobility.
The above findings help to understand why migration increase income mobility and reduce annual and permanent income inequality from 2003 to 2006. There are at least four aspects to this issue: First, there is higher proportion of households with migrant workers in the relatively poorer interior provinces than in the coastal provinces. Second, poorer households of coastal provinces have a higher probability to have migrant workers than the richer households. Third, migrant workers accounted for a higher proportion of the labor force in the interior provinces than in the coastal provinces. Finally, the migrant workers in interior provinces tend to work in locations where they can make higher a income, such as coastal cities rather than their hometown. 6. Conclusion In this paper, we examined the dynamics of households income mobility in rural China between 1986 and 2006, and the determinants of income mobility with a focus on the impact of migration from 2003 to 2006. Our results have strong implications for understanding China s rural economic development, the dynamics of income distribution, and the role of migrant workers during this developing process. Firstly, the earlier upward trend of income inequality peaked in 2002 and then decreased from 2003 to 2006. Secondly, income mobility remained at a high level and permanent income inequality decreased. These indicate that the situation of the poor in rural China has been improved. Our analyses suggest that migrant workers have played an important role in this process. The Multinomial Logit regressions find that poor
households with migrant workers had a higher probability of moving upward on the income ranking during 2003-2006. Further analyses provide the evidence on how migrant workers have affected income mobility. In general, the households of interior provinces and the poor households of coastal provinces tend to have a higher tendency to have migrant workers. The direction of migration is toward regions with higher income levels, which directly explains the positive impact of rural migration on upward income mobility. References Benjamin, Dwayne; Brandt, Loren; Giles, John, 2005, The Evolution of Income Inequality in Rural China, Economic Development and Cultural Change, 53(4):769-824. Benjamin, Dwayne; Brandt, Loren; Giles, John, 2006, Inequality and Growth in Rural China: Does Higher Inequality Impede Growth? University of Toronto, Department of Economics, Working Papers. Björklund, Anders, Laura Chadwick, 2003, Intergenerational Income Mobility in Permanent and Separated Families, Economics Letters, 80(2), 239-246. Brown,Philip H., Erwin Bulte and Xiaobo Zhang, 2009, Positional Spending and Status Seeking in Rural China, working Paper. Cai, Fang, 2007, Growth and Structural Changes in Employment in Transitional China, Economic Research Journal, 7, 4-14. (in Chinese) Cai, Fang, Yang Du, Meiyan Wang, 2001, Household Registration System and Labor Market Protection. Economic Research Journal, 12,41-49. (in Chinese) Chadwick, Laura, Gary Solon, 2002, Intergenerational Income Mobility among Daughters, American Economic Review, 92, pp. 335-344. Deng, Quheng, Gustufusson, 2007, China s Permanent Migration. Economic Research Journal, 4, 137-147. (in Chinese)
Dercon, S. and P. Krishnan, 2000. Vulnerability, Seasonality, and Poverty in Ethiopia. Journal of Development Studies, 36, 25-53. Ding, Ning, Yougui Wang, 2008, household income mobility in China and its decomposition, China Economic Review, 19, 373-380. Fields Gary S, Ok Efe A. The Measurement of Income Mobility: An Introduction to the Literature, C.V. Starr Center for Applied Economics, New York University, Working Papers. 1996. Fields, G. S., & Ok, E. A. 1999. Measuring movement of income. Economica, 66, 455 471. Friedman, Milton, 1962, Capitalism and Freedom, Chicago: University of Chicago Press, P172. Gang Ira N, Yun Myeong-Su, Landon-Lane John. Gender Differences in German Upward Income Mobility. Working Papers. 2002. Giles, John. Is Life More Risky in the Open? Household Risk-Coping and the Opening of China s Labor Markets. Journal of Development Economics, 2006(10):25-60. Khor, Niny and John Pencavel, 2006, Income Mobility of Individuals in China and The United States, Economics of Transition, 14(3), 417-58. Li, Shi, Yue Ximing, 2004, Investigation of Disparity between Rural and Urban China, Tribune of Villages and Townships, 4, 21-22. (in Chinese) Lipton, Michael, Qi Zhang, 2006, Reducing inequality and poverty during liberalization in China: rural and agricultural experiences and policy options, FED working paper. Luttmer, Erzo F. P. 2005, neighbors as negatives: relative earnings and well-being, The Quarterly Journal of Economics, August, 963-1002. Olga, Canto, 2000, Income Mobility in Spain: How Much Is There? Review of Income and Wealth, 46(1), 85-102. Ravallion, M. and S. Chen 2004, China s (uneven) progress against poverty, Policy Research Working Paper 3408, World Bank. Solon, Gary. 1992, Intergenerational Income Mobility in the United States.
American Economic Review, 82(3), 393-408. Sun Wenkai, Jiangyong Lu, Chong-En Bai, 2007, An Analysis of Income Mobility in Rural China. Economic Research Journal, 8, 43-57. (in Chinese) Wang, Xiaolu, Fan Gang, 2005, Trends and Influential Factors of Income Inequality in China, Economic Research Journal, 10, 24-36. (in Chinese) Woolard, Ingrid, Stephan Klasen, 2005, Determinants of Income Mobility and Household Poverty Dynamics in South Africa, Journal of Development Studies, 41(5), 865-97. Yin, Heng, Li Shi, Deng Quheng, 2006, A Study of Income Mobility of Individuals in Urban China, Economic Research Journal, 10, 30-43. (in Chinese) Zhang, Qi, Jianwei Mi, Jikun Huang, 2007, Income moblility and income distribution: evidence from rural China. Economic Research Journal, 11, 123-138. (in Chinese)
Appendix Table 1: Descriptive Statistics of Variables 2003 2004 2005 2006 Per capita income 3330.52 3763.87 4109.43 4518.49 Household net income 13161 14859.67 16004.4 18060.32 Household total income 21341 23008.09 24875.79 27492.95 Migration income 4270.57 7556.52 8573.33 9937.70 Household size 4.145 4.066 4.069 4.075 Labor force 2.678 2.821 2.811 2.806 Male labor force 1.44 1.519 1.521 1.517 Labor force with skill 0.162 0.161 0.163 0.169 Proportion of labor force with skill 0.053 0.054 0.0548 0.0571 Village official 0.0479 0.047 0.0434 0.0425 Education level of household head 6.329 6.302 6.334 6.351 Proportion of household with migration 0.265 0.2741 0.2901 0.3306 Household net income in coastal area 20473.9 22439.5 23913.1 26618.9 Household total income in coastal area 35500.8 35489.5 39395.1 42712.1 Household net income in interior area 10810.9 12423.55 13461.5 15308.78 Household total income in interior area 16792.3 18997.8 20210.2 22602.06 Migration income in coastal area 6467.9 11211.18 12629.44 14177.34 Migration income in interior area 3564.93 6420.43 7299.41 8655.95 Number of Households 12889 12889 12889 12889 Province number 23 23 23 23
Table 2: Proportion of Households with Migrant Workers (1) (2) (3) 1986 18.76% 16.89% 21.00% 1987 20.31% 21.23% 19.21% 1988 22.12% 24.00% 19.86% 1989 17.60% 18.21% 16.86% 1990 19.96% 17.84% 22.50% 1991 16.69% 18.42% 14.62% 1995 18.78% 20.19% 16.79% 1996 18.24% 19.86% 15.93% 1997 18.62% 20.32% 16.19% 1998 17.94% 19.16% 16.19% 1999 16.88% 17.59% 15.86% 2000 15.89% 16.61% 14.87% 2001 16.58% 16.33% 16.92% 2002 17.07% 16.52% 17.85% 2003 57.97% 53.62% 59.37% 2004 58.79% 52.92% 60.67% 2005 60.98% 55.54% 62.73% 2006 63.46% 60.07% 64.56% 1986-1991 49.35% 49.25% 49.47% 1995-2002 38.87% 37.81% 40.38% 2003-2006 76.37% 72.61% 77.57% Notes: (1) Percentage of households with migrant workers;(2) Percentage in coastal provinces;(3) Percentage in interior provinces 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2003 2004 2005 2006 migration income/net income coastal migration income/net income interior migration income/net income Figure 1. The Trend of Proportions of Migration Income Account for Net and Total Income
Table 3: Annual Income Inequality Merged data Gini Unmerged data Gini 1986 0.4121 1986 0.4118 1987 0.3993 1987 0.3841 1988 0.4169 1988 0.4061 1989 0.4115 1989 0.3974 1990 0.4224 1990 0.3811 1991 0.4535 1991 0.4256 1995 0.4424 1995 0.4005 1996 0.462 1996 0.4186 1997 0.4626 1997 0.4154 1998 0.4655 1998 0.4117 1999 0.4724 1999 0.4237 2000 0.49 2000 0.4513 2001 0.4859 2001 0.4383 2002 0.5612 2002 0.4975 2003 0.47 2003 0.4218 2004 0.4758 2004 0.3969 2005 0.4551 2005 0.4053 2006 0.4201 2006 0.4025 Notes: merged sample set contains 2095 households. There are 5416, 4098 and 12889 households in three periods if we don t merge data. Table 4: Permanent Income Inequality Merged data Gini Unmerged data Gini ave86_90 0.3898 ave86_90 0.3565 ave87_91 0.3998 ave87_91 0.3627 ave95_99 0.4241 ave95_99 0.3741 ave96_00 0.438 ave96_00 0.3869 ave97_01 0.4433 ave97_01 0.393 ave98_02 0.4605 ave98_02 0.4078 ave99_03 0.458 ave03_06 0.3695 ave00_04 0.4574 ave01_05 0.445 ave02_06 0.4296
0.58 0.53 0.48 0.43 0.38 1986 1987 1988 1989 1990 1991 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Gini Coefficient Figure 2. Gini Coefficients of Annual Income from 1986 to 2006 0.48 0.46 0.44 0.42 0.4 0.38 ave86_90 ave87_91 ave95_99 ave96_00 ave97_01 ave98_02 ave99_03 ave00_04 ave01_05 ave02_06 Gini Coefficient Figure 3. Gini Coefficients of Average Income of Five Years Table 5: Correlation Coefficients Matrix from 1986 to 1991 aveinc86 aveinc87 aveinc88 aveinc89 aveinc90 aveinc91 aveinc86 1 aveinc87 0.494 1 aveinc88 0.4645 0.6247 1 aveinc89 0.3599 0.4177 0.4976 1 aveinc90 0.3751 0.4457 0.509 0.5778 1 aveinc91 0.2969 0.2885 0.3356 0.3449 0.4238 1
Table 6: Correlation Coefficients Matrix from 1995 to 2002 aveinc95 aveinc96 aveinc97 aveinc98 aveinc99 aveinc00 aveinc01 aveinc02 aveinc95 1 aveinc96 0.6487 1 aveinc97 0.568 0.6191 1 aveinc98 0.4862 0.6148 0.7102 1 aveinc99 0.4826 0.5848 0.6911 0.8886 1 aveinc00 0.4767 0.5901 0.6327 0.8337 0.8481 1 aveinc01 0.4962 0.5242 0.6047 0.5948 0.6447 0.7318 1 aveinc02 0.3861 0.3644 0.3631 0.3795 0.3907 0.4484 0.471 1 Table 7: Correlation Coefficients Matrix from 2003 to 2006 aveinc03 aveinc04 aveinc05 aveinc06 aveinc03 1 aveinc04 0.6905 1 aveinc05 0.7148 0.8154 1 aveinc06 0.5593 0.5989 0.6888 1 Table 8: Correlation of Income Growth Rate and Income Ranking Ln[inc06]-ln[inc03] percentile03 Ln[inc06]-ln[inc03] ln_inc03 Ln[inc06]-ln[inc03] 1 Ln[inc06]-ln[inc03] 1 percentile03-0.4485 1 ln_inc03-0.5050 1 Ln[inc06]-ln[inc03] percentile06 Ln[inc06]-ln[inc03] ln_inc06 Ln[inc06]-ln[inc03] 1 Ln[inc06]-ln[inc03] 1 percentile06 0.3267 1 ln_inc06 0.3696 1 Ln[inc06]-ln[inc03] Percentile03_06 Ln[inc06]-ln[inc03] ln_inc03_06 Ln[inc06]-ln[inc03] 1 Ln[inc06]-ln[inc03] 1 Percentile03_06-0.0233 1 ln_inc03_06-0.0166 1
9869959990031988 10.052 0.143 0.227 0.309 0.269 10.076 0.112 0.256 0.316 0.240 10.049 0.110 0.217 0.409 0.215 20.057 0.144 0.239 0.325 0.235 Table 9: Transition Matrices in the Three Periods for Unmerged Data 1989 (N=5416) 1 2 3 4 5 1 0.541 0.230 0.138 0.057 0.034 2 0.236 0.295 0.250 0.152 0.066 3 0.112 0.261 0.262 0.252 0.113 4 5 0.059 0.071 0.123 0.229 0.519 1991 (N=5416) 1 2 3 4 5 1 0.518 0.254 0.127 0.082 0.019 2 0.201 0.317 0.245 0.173 0.065 3 0.105 0.213 0.307 0.257 0.118 4 0.056 0.107 0.202 0.331 0.303 5 0.119 0.109 0.120 0.157 0.496 1998 (N=4098) 1 2 3 4 5 1 0.538 0.254 0.094 0.087 0.027 2 0.223 0.335 0.249 0.140 0.052 3 0.117 0.248 0.307 0.229 0.098 4 5 0.045 0.052 0.093 0.228 0.583 2002 (N=12889) 1 2 3 4 5 1 0.593 0.250 0.111 0.033 0.014 2 0.240 0.365 0.238 0.110 0.046 3 0.093 0.239 0.344 0.235 0.090 4 5 0.025 0.037 0.090 0.214 0.635 2006 (N=12889) 1 2 3 4 5 1 0.524 0.232 0.131 0.073 0.041 2 0.265 0.315 0.233 0.128 0.059 3 0.118 0.254 0.293 0.239 0.096 4 5 0.035 0.057 0.104 0.235 0.569
986995001Table 10: Summary Indices of Transition Matrix for Unmerged Data (1) (2) (3) 1986-1989 0.926 0.385 0.776 1988-1991 0.964 0.394 0.760 1995-1998 0.865 0.415 0.801 1999-2002 0.740 0.469 0.839 2003-2006 0.883 0.405 0.792 Note: (1): Average Quintile Move; (2): Quintile Immobility Ratio; (3): the fraction that remain in the same quintile plus the fraction who move one quintile. 4 0.043 0.100 0.222 0.360 0.275 14 0.084 0.122 0.220 0.341 0.234 14 0.057 0.160 0.263 0.320 0.201 2Table 11: Transition Matrices in Three Periods for Merged Dataset 1991 (N=2095) 1 2 3 4 5 1 0.563 0.258 0.138 0.029 0.012 2 0.222 0.337 0.255 0.141 0.045 3 0.093 0.229 0.308 0.275 0.096 5 0.079 0.076 0.076 0.196 0.573 2000 (N=2095) 1 2 3 4 5 1 0.535 0.246 0.136 0.067 0.017 2 0.220 0.363 0.227 0.155 0.036 3 0.129 0.212 0.322 0.243 0.093 5 0.033 0.057 0.096 0.193 0.621 2006 (N=2095) 1 2 3 4 5 1 0.573 0.229 0.074 0.067 0.057 2 0.246 0.301 0.222 0.146 0.086 3 0.105 0.270 0.346 0.189 0.091 5 0.019 0.041 0.096 0.279 0.566 Table 12: Summary Indices of Transition Matrix for Merged Data (1) (2) (3) 1986-1991 0.832 0.428 0.814 1995-2000 0.837 0.436 0.795 2001-2006 0.859 0.421 0.800 Note: (1): Average Quintile Move; (2): Quintile Immobility Ratio; (3): the fraction that remain in the same quintile plus the fraction who move one quintile
Table 13: Migrant Workers and Income Growth Households Households with without migrant migrant workers workers during during 2003-2006 2003-2006 Whole sample Per capita income in 2003 3332.87 3322.9 3330.52 Per capita income in 2006 4626.45 4169.12 4518.50 Income growth rate 38.81% 25.47% 35.67% 2003 Table 14: Conditional Transition Matrix during 2003-2006 2006 1 2 3 4 5 1 0.478 0.248 0.147 0.078 0.049 2 0.241 0.311 0.249 0.137 0.062 3 0.113 0.238 0.295 0.255 0.098 4 0.057 0.144 0.229 0.327 0.244 5 0.032 0.059 0.100 0.243 0.566 Table 15: The Determinants of Income Growth Rate, OLS Estimation OLS Robust OLS Coef. P>t Coef. P>t aveinc03-2.88e-05 0.00-2.88E-05 0.00 official03-0.01 0.79-0.01 0.78 party03 3.65E-03 0.83 3.65E-03 0.84 minority03-0.10 0.00-0.10 0.00 pop03 2.35E-03 0.60 2.35E-03 0.61 Labor_ratio03 0.14 0.00 0.14 0.00 Male_ratio03 0.08 0.04 0.08 0.04 Skill_ratio203-0.06 0.13-0.06 0.17 Ave_laboredu03-8.70E-04 0.78-8.70E-04 0.80 Gender03 2.34E-03 0.92 2.34E-03 0.92 Age03-1.36E-03 0.03-1.36E-03 0.03 Edu03-1.93E-03 0.47-1.93E-03 0.46 Health -0.05 0.02-0.05 0.05 Pop_change -0.09 0.00-0.09 0.00 Labor_change 0.39 0.00 0.39 0.00 Male_change 0.15 0.00 0.15 0.00 Skill_change 0.15 0.00 0.15 0.00 Laboredu_change 4.75E-03 0.00 4.75E-03 0.04 Migration 0.08 0.00 0.08 0.00 Coastal 0.03 0.03 0.03 0.19 _Cons 0.30 0.00 0.30 0.00 Sample Size:11609 F-statistic:84.29 R2=0.127 Notes: Dependent variable is the households income growth rate.