Does Local Social Capital Deter Labor Migration? Evidence from Rural China

Does Local Social Capital Deter Labor Migration? Evidence from Rural China Liqiu Zhao 1 and Xianguo Yao 2 1 School of Labor and Human Resources, Renmin University of China, Beijing, China 2 School of Public Affairs, Zhejiang University, China Abstract This paper empirically investigates the effect of local social capital on jobrelated migration in rural China. A household s social ties in the region of origin, which we refer to as local social capital, may deter migration, because local social capital is location-specific and the returns to it will decrease or disappear if an individual migrates. In view of Chinese gift-giving culture, we use household expenses on wedding gifts for family members outside household, relatives and friends as a proxy for local social networks. Based on the data from China Health and Nutrition Survey (CHNS), we find that in rural China local social networks have a significantly negative effect on labor migration after controlling for the characteristics of individual, household and village. The IV results suggest that 10 percent increase in wedding gifts expenses results in roughly a 1.5 percentage point decrease in migration probability. Interestingly, both the negative effect of local social networks and the positive effect of migration networks are only significant for female. Key words: Migration; Local social networks; Gift exchange JEL Classification: R23, O15, Z13 We would like to thank Joep Konings, Damiaan Persyn, Ulbricht Robert, Wouter Torfs, Joachim Winter, Yaohui Zhao, Linke Zhu and seminar participants at K.U.Leuven and the 6th CWE International Workshop, for helpful comments and suggestions. We also gratefully acknowledge the financial support from the Key Project of National Natural Science Foundation of China (Grant No. 70933001). All errors are our own. 1

1 Introduction A large scale of labor migration from rural to urban areas has occurred in China since the mid-1980s 1, which has created the largest labor flow in world history. This phenomenon attracted much attention from economists (e.g. Yang, 1993; Liang and White, 1996; Zhao, 1999a,b; Scott et al., 1999; Chen et al., 2010; Zhao, 2002, 2003, etc.). Following the classical theories which argue that an individual will migrate if the present value of gains from migration exceeds the costs of migration (Sjaastad, 1962), most of previous research on internal migration in China focus on the determinants of rural to urban migration, and the key variables include age, gender, education, marital status, per capita land allocation, per capita production assets, urban-rural income gap, residence registration and etc. Recently, some studies have attempted to identify the role of social networks in job-related migration. The existing line of research focuses on the facilitating effect of migration networks at points of destination on the propensity to migrate (Knight and Song, 2003; Zhao, 2003; Chen et al., 2010). They find that migration networks, which reduce the costs and risks associated with migration and job search, have a significantly positive effect on the probability of migration 2. However, some research reveal that although migration yields a large monetary premium, rural people generally choose rural non-farm work over migration(e.g. Zhao, 1999a), which suggests that the decision to migrate by Chi- 1 1% National Population Sample Survey in China show that the inter-county migration population grew from 20 million in 1990 to 45 million in 1995, 79 million in 2000 and 147 million in 2005. 2 Migration networks can reduce information cost by providing job information to potential migrants, reduce psychological cost by providing supportive relationship to migrants in destinations and reduce the probability of unemployment by providing direct job search (Zhao, 2003). 2

nese rural people may also be affected by some non-pecuniary forces in the source region, for example, social networks in the source region, which we refer to as local social networks. Nevertheless, none of these studies shed light on the role of local social networks, which may play an important role in the migration decision in rural China. The importance of social networks have been emphasized in the analysis of Chinese economy. A number of studies show that individuals who enjoy more local social networks have higher probability to find a non-farm job in a village (Zhang and Li, 2003), obtain higher income (Knight and Yueh, 2008) and are healthier and happier (Yip et al., 2007). Furthermore, local social capital is often location-specific suggesting that once an individual leaves his original region, the returns to local social capital decrease or even disappear. Thereby, local social networks are a part of the opportunity cost of migration. All these factors indicate that local social networks may have a negative impact on migration. This paper aims to investigate the impact of local social networks on job-related migration in rural China. Does local social capital deter migration? This is hardly a new research question in migration literature. Sjaastad (1962) argues that migration involves a psychic cost, because people are genuinely reluctant to leave familiar surroundings, family and friends. Toney (1976) shows that the duration of residence is longer in localities where an in-migrant has previous social and kinship ties, which helps to explain why some individuals continue to live in economically depressed areas. Speare et al. (1982) present evidence that individuals with friends and relatives in their current location are less likely to move than their counterparts. Using the availability of assistance offered by someone nearby when there is a serious emergency as a surrogate of local 3

social capital, Kan (2007) finds that the availability of emergency assistance to a household from someone living nearby deters a household from moving. Michaelides (2011) shows that workers with strong local ties (i.e. locally born workers and married workers whose spouse is locally born) are significantly less likely to move. David et al. (2010) formalize Kan (2007) s intuition into a model and test it using crosscountry data. However, most of the existing studies apply poor measures of local social networks, for example, a dummy variable to capture the availability of assistance or whether a worker is locally born. In this paper, a better measure of local social networks, i.e., household expenses on wedding gifts for family members outside household, relatives and friends, is used which captures both the quantity and quality of local social networks in the context of Chinese gift-giving culture 3. As it has rarely been explored in the previous literature, a better measure of local social capital is our first contribution to the literature. Secondly, to the best of our knowledge, this paper is the first study examining the role of local social networks in migration in the context of China. The findings from China can also offer insights to other societies with similar socioeconomic conditions and cultural backgrounds. In this paper, we investigate the impact of local social networks on migration decision in rural China. To do so, we use the data from China Health and Nutrition Survey (CHNS), which enables us to measure the local social networks by household expenses on wedding gifts. By addressing potential simultaneity and measurement error problems using IV method, we find that in rural China local social networks have a significantly negative effect on migration after controlling for the characteristics of 3 In Section 2, we provide further arguments for the use of this measure. 4

individual, household and village. Our IV results suggest that 10 percent increase in household expenses on wedding gifts results in a 1.5 percentage point decrease in migration probability. Interestingly, the negative effect of local social capital on migration is only significant for female. Finally, as a robustness check, we apply an alternative measure of local social ties, that is, the number of family members who are local cadres, and the results hold. The rest of this paper is organized as follows. Section 2 briefly introduces social networks in rural China. Section 3 outlines the empirical specification. Section 4 describes the dataset used and presents summary statistics for the sample. Section 5 examines the effect of local social networks on migration by controlling for the characteristics of individual, household and village. Section 6 concludes. 2 Social Networks in Rural China One of the most significant features of Chinese culture is its emphasis on a harmonious society and the appropriate arrangement of interpersonal relationships. Fei (1948) makes a comparison between the West and China and concludes that, structurally, Chinese society is composed of numerous personal networks. Thus, Chinese society is neither individual based nor group based but relation based. Fei (1948) also emphasizes the importance of interpersonal relationships within a village and distinguishes them from market relations outside the village. Villagers get pecuniary and non-pecuniary benefits from local social networks. Local social networks can raise earnings (Knight and Yueh, 2008), promote employment (Bian, 1997; Zhang and Li, 5

2003) and improve well-being (Yip et al., 2007) in China. Gift exchange, while existing in all societies, appears to be essential to Chinese culture throughout its long history. Courtesy demands reciprocity, The Classic of Rites indicates that the essence of relationship in China is social exchange and the principle of gift exchange is reciprocity 4. In rural China, villagers exchange gifts and offer assistance in order to maintain interpersonal relations, enhance the sense of belonging in the community and accordingly keep and develop various social networks in a village. Gift-giving follows the notion of reciprocity 5. This reciprocity is characterized by the obligation of giving, receiving and returning gifts in the long run. In Chinese society, when one has either happy occasions or difficulties, all one s relatives, friends and acquaintances are supposed to offer a gift or render some substantial assistance (Hwang, 1987). Once a benefactor generates a relationship by giving a benefit to another, the receiver is obligated to repay if circumstances permit in order to restore balance (Hwang, 1987). The obligatory give and take maintains, strengthens and creates various social ties. In addition, traditional Chinese tend to adopt multiple standards of behavior for interacting with people of different types of relationships or different degrees of intimacy around them (Fei, 1948; Hsu, 1953), which suggests that not only quantity (the number of contacts in a social network) but also quality (the intimacy of the relationships) matters for the stock of social capital. Wedding and funeral are the two most important gift-giving occasions in rural 4 The Classic of Rites was one of the Five Classics of the Confucian canon, which described the social forms, ancient rites, and court ceremonies of the Zhou Dynasty (1122 BC to 256 BC). 5 Calculation of balances in social exchange is very important in social relations among villagers, for a feeling of indebtedness from past gifts compels villagers to reciprocate. 6

China. People in the social networks are supposed to give a red packet which is a monetary gift contained in a red envelope to a new couple in a Chinese wedding. The amount of money in the red envelope depends on the extent of intimacy with the new couple. Therefore, the expenses on wedding gifts, to a great extent, reflect the social capital of a household, i.e., the expenses on wedding gifts measure not only the quantity but also the quality of the social networks. In this paper, household expenses on wedding, including gifts for family members outside household, relatives and friends (excluding dowry or bride price), during the past 12 months is used to measure the local social capital of a household. Since social capital is normally homogeneous among household members, we can use household level measure of local social capital as a proxy for an individual s social networks 6. In addition, household expenses on funerals, an alternative measure of local social networks, is also applied in this paper. We discuss it in more detail in Section 3. 3 The Empirical Specification The unit of analysis here is the individual. We define Mig it = 1 if an individual migrates in year t, otherwise Mig it = 0. For an individual i in household h village v, the binary choice of whether to migrate or not can be represented generally by: Mig it = γs ht + β 1 X it + β 2 X ht + β 3 X vt + Cohort + δ p + θ t + ɛ it (1) 6 Family members share household s social networks. For example, you can benefit from the social networks your brother possesses. 7

where s ht refers to local social capital measured by household expenses on wedding gifts in the last 12 months, X it denotes i s individual characteristics such as gender, age, marital status, educational attainment and household registration type. Since migration decisions are probably made by the household as a whole instead of by each individual separately, we control for household characteristics X ht, containing age of household head, household income per capita, cultivated land per capita, household size, number of kids under 6 years, number of kids 6 to 12 years and number of elderly 60 years or older. Note that both the household income per capita and household expenses on wedding gifts are inflated to 2006. As individuals in desirable locations are less likely to migrate, village characteristics X vt, for instance, the percentage of the labor force working out of the village, percentage of labor force engaged in agriculture, whether the village has convenient telephone service, whether the village is near a bus or train station, whether the village has enterprises run by the village, and the average household expenses on wedding gifts at village level are controlled for 7. In contrary to local social networks, the percentage of the labor force working out of the village for more than one month last year, which captures the social networks outside the village extended beyond families, is used to control for the migration networks. Migration networks, which reduce migration costs, are supposed to increase migration probability. To control for cohort fixed effect, we also include Cohort dummies in equation (1) 8. Province dummies δ p and year dummies θ t control for both 7 In rural China, the amount of money spent on wedding gifts varies dramatically across regions (see Table 2). Relative amount of money spent on wedding gifts comparing to others in the same village matters, rather than the absolute value. Therefore, we control for the average expenses on wedding gifts at village level. 8 Individuals are divided into five birth cohort groups: born before 1950, 1950-1960, 1960-1970, 1970-1980, after 1980. 8

fixed province effects and time-varying macroeconomic shocks. ɛ it is the error term representing unobserved individual heterogeneity, and it is clustered at household level to allow for arbitrary correlation within families. Our key hypothesis is that an increase in local social capital has a negative effect on migration probability (γ < 0). Though we have longitudinal data, we apply pooled probit model here. The reason is that since local social networks at household level are not likely to change substantially over time, panel analysis might fail to identify the full effect of local social networks. One problem concerning the above specification is that household s expenses on wedding gifts may not be exogenous. The investment in social capital is determined by the expected benefits. If households perceive themselves as being strongly attached to a village, they will invest more in local social capital, because the returns from these social ties are high. However, households with higher propensity to migrate have less incentive to invest in social capital, for the family members in the household who migrate will benefit less from the local social networks. Thus, s ht in equation (1) may be correlated with the error term ɛ it. If we assume that the household expenses on wedding gifts are determined by household characteristics X ht, s ht = X htθ + ɛ ht (2) where X ht is a vector of variables some of which may overlap with X ht. Then the error terms ɛ it and ɛ ht may be correlated. The simultaneity between the social capital investment and the migration gives rise to an identification problem. However, we argue 9

below that the simultaneity problem is probably not a serious problem in the context of rural China. First of all, because of the household registration system (Hukou system) and other institutional barriers, the dominant form of labor migration in China is circular migration (Zhao, 2003) 9. Migrants travel between rural home and urban jobs every year. An average migrant returns home two to three times (for around 5 months) a year during the harvest times and the Spring Festival (Hare, 1999; Knight et al., 1999), suggesting that migrants contact people in the village of origin from time to time through which local social capital is maintained. Thus, contrary to the deteriorative effect of permanent migration on local social ties, circular migration serves to maintain and strengthen local social networks. Secondly, rural migrants mostly move as temporary migrants, without their families to which we refer as individual instead of family migration. Thus, the left-behind family members can maintain the relations with others in the village. Finally, the reciprocity principle of wedding giftgiving implies that the expenses on wedding gifts are determined by the wedding gifts you received before or will receive in the future. The expenses on wedding gifts are balanced in the long run which are probably not affected by temporary migration decisions. Though the problem of reverse causality from migration to household expenses on wedding gifts may not be serious, we apply instrumental-variable method to deal with this potential problem. Measurement error is another potential problem which may arise when household 9 The Hukou system in China is similar to an internal passport system. A person s local citizenship and residence is defined as a birth right. The entitlements and details of a system differ for urban and rural residents. Temporary migrants to large cities have very high priced access to health care, schooling facilities, pension benefits and housing. And migrants have to apply for temporary residential permit to stay in cities. 10

heads under- or over-report their expenses on wedding gifts or household expenses on wedding gifts depend on the probability of the occurrence of weddings. In the classical model of measurement error, s ht = s h + µ 1ht (3) where s h is the unobservable true value of local social capital of household h and µ 1ht is the measurement error capturing the report errors or the randomness of occurrences of weddings, which is uncorrelated with the true value s h. It implies that Cov(s ht, µ 1ht ) 0, then the estimate of γ in equation (1) has an attenuation bias. This paper pursues two strategies for treating the simultaneity problem and measurement error problem. The first approach is to average household s expenses on wedding gifts in equation (1) over several years: Mig it = γ s h + β 1 X it + β 2 X ht + β 3 X vt + Cohort + δ p + θ t + ɛ 1it (4) The average is preferred to a single year value for the following two reasons. First of all, using the average of household s expenses on wedding gifts can weaken the impact of the propensity to migrate on household investment in local social capital and thus mitigate the reverse causality from migration decision to household expenses on wedding gifts. Secondly, since single year expenses on wedding gifts are probably measured with classical measurement error, using the average expenses can decrease the measurement error and then reduce the attenuation bias. 11

The second approach is to apply instrumental variable to estimate equation (1). To address the measurement error problem, an alternative measure of s h whose measurement error is uncorrelated with µ 1ht is a candidate of instrumental variables. In a Chinese funeral, relatives and friends are supposed to send wreath or monetary gifts contained in a white envelope to the family of the deceased. So the amount of money a household spent on funerals is an alternative measure of local social networks. We apply the three- to four-year-lagged household expenses on funerals(period t 1) as an IV for the household expenses on wedding gifts (period t) to eliminate the bias caused by measurement error 10. f ht 1 = s h + µ 2ht 1 (5) where f ht 1 is the household expenses on funerals in period t 1, which should be partially correlated with s ht. Normally, the probability of the occurrence of death in period t 1 is uncorrelated with the probability of the occurrence of weddings of people in the networks in period t. Meanwhile, whether respondents under- or over-report the expenses on funeral in period t 1 is uncorrelated with whether respondents under- or over-report the expenses on wedding gifts in period t. It implies that Cov(µ 1ht, µ 2ht 1 ) = 0. The two arguments above indicate that f ht 1 can be a candidate of IV to deal with the measurement error issue. On the other hand, household expenditure on funerals three or four years ago is probably not directly 10 More precisely, we use household expenses on funerals in 1993 as an instrument for the household expenses on wedding gifts in 1997, and expenses on funerals in 1997 as an instrument for the expenses on wedding gifts in 2000. 12

related to the migration behavior of household members in period t, suggesting that it can mitigate the simultaneity problem at the same time. Therefore, the amount of money spent on funerals three to four years ago can be treated as an IV for the expenses on wedding gifts. As a robust check, we follow Kan (2007) and use six- or seven-year-lagged household expenses on wedding gifts as the instrument for expenses in this period 11. Sixor seven-year-lagged expenses on wedding gifts should be highly correlated with the current expenses on wedding gifts, as the local social networks for a household are stable in the short run. Meanwhile, it is reasonable to assume that the expenses on wedding gifts six- or seven-year ago has no direct effect on the current propensity to migrate. Thus, six- or seven-year-lagged expenses on wedding gifts can be treated as the instrument for the expenses in this period. 4 Data The data used in this study is from China Health and Nutrition Survey (CHNS), which is a longitudinal dataset with seven survey waves (1989, 1991, 1993, 1997, 2000, 2004, 2006) 12. The CHNS is conducted by the Carolina Population Center at the University of North Carolina at Chapel Hill and the National Institute of Nutrition and Food Safety at the Chinese Center for Disease Control and Prevention. The survey covers 9 provinces including Guangxi, Guizhou, Heilongjiang, Henan, Hubei, Hunan, 11 Kan (2007) uses whether or not a household has received assistance offered by people living nearby in the previous five years and whether or not a household has given assistance to someone outside the household in the previous five years as instruments. 12 See http://www.cpc.unc.edu/projects/china for details. 13

Jiangsu, Liaoning and Shandong, which vary substantially in geography, economic development, public resources, and health indicators. A multistage, random cluster process was used to draw the sample surveyed in each of the provinces. The survey was conducted at both the community and household levels. A community refers to a village in a rural area or a neighborhood in an urban area. The household survey contains the detailed demographic, economic, time use, labor force participation, asset ownership, and expenditure information, and the community survey collects information on community infrastructure, services, population, prevailing wages, and related variables. This dataset provides us a unique opportunity to examine the relationship between local social networks and job-related migration. We restrict our analysis to rural sample which accounts for 69.82% of total sample 13. Our analysis pertains to individuals no younger than age 16, which is the legal working age in China, and no older than 60 14. Since it is only from the round of 1997, respondents have been required not only to report whether they live in the household or not but also to report the specific reasons for not living at home(e.g., go to school, military service or seek employment elsewhere), we restrict our analysis to the last four waves of the CHNS survey (1997, 2000, 2004, 2006), which allows us to distinguish migrants who left to seek jobs from individuals who left for reasons such as military services, marriage and schooling. A migrant is defined as someone who is currently not living at home and seeking employment elsewhere. Due to the missing report of marital status for migrants in 2004 and 2006, the analysis in this 13 It is consistent with the population composition in China. 1% National Population Sample Survey in China show that the rural population share dropped from 63.91% in 2000 to 57.01% in 2005. 14 In-school students are dropped from the sample 14

paper relies on the data in 1997 and 2000 15. Figure 1(a) indicates that the rural-to-urban migration rate has increased dramatically from 7.70 percent in 1997 to 25.04 percent in 2006 16. Figure 1(b) illustrates the relationship between age and the probability that a male or female worker will migrate. Women in general are less likely to migrate than men. However, the very youngest women (younger than 22) have a higher probability to migrate than the youngest men. But this probability declines substantially for both male and female over the working life. [Figure 1 about here.] The characteristics of migrants are distinct from non-migrants. Table 1 shows the variables used and summary statistics for the sample that contrast non-migrants and migrants. There are 1,112 migrants in our sample which account for 9.88% of total sample. 58.54% of migrants are male suggesting that men are more likely to migrate. Also, migrants tend to be younger than non-migrants. The average age of migrants is 25.92, while the average age of non-migrants is 38.19. 34.62% of migrants are married while 80.66% of non-migrants are married. Migrants have on average 8.35 years of schooling which is significantly higher than non-migrants (6.82). The expenses on wedding gifts for migrant households are on average 683.83 yuan, which is significantly lower than that of non-migrant households. However, the average household income per capita is significantly higher for migrant households. For the household asset, the cultivated land per capita is 1.43 Mu for migrants, which is significantly lower than 15 Only 1.70% of migrants report their marital status in the waves of 2004 and 2006. 16 The migration rate is defined as the share of labor force in the village who work outside the village to the total labor force (16 age 60). 15

that of non-migrants. In addition, the data also show that migrant households tend to have significantly larger household size, more male laborers, fewer preschool and school-aged children. From Table 1, we also find that migrants are more likely to come from the villages with richer migration networks outside, which is consistent with Zhao (2003) and Chen et al. (2010). In addition, migrants are less likely to come from villages near a train station or from villages having convenient telephone service, paved road or enterprises run by the village. Table 2 presents a clear evidence of the prevalence of social networks in rural China. In the sample, around 71.20 percent households have positive expenses on wedding gifts. And the share of household expenses on wedding gifts in household net income increased from 5.69% in 1989 to 11.12% in 2006. As shown in the first column in Table 2, the household expenses on wedding gifts vary dramatically across provinces, varing from 448.27 yuan in Guangxi to 1497.37 yuan in Liaoning. And the standard deviation in parentheses also indicates a large variation of expenses on wedding gifts within provinces. The CHNS also contains the information about whether a family member is a local cadre in the village, which captures the existence of nepotism. We construct an indicator measuring whether a household has family members as village cadres, which is applied in Zhang and Li (2003), as an alternative measure of local social capital. The second column in Table 2 shows that on average about 5.26 percent of households have access to this measure of social networks, which is having a family member as cadre. This value varies from 3.49 percent in Guangxi province to 6.94 percent in Shandong province. 16

[Table 1 about here.] Given the prevalence of social networks in rural China, we turn to its relation with migration. Figure 3(a) shows plots of the bivariate relationship between the household expenses on wedding gifts and age of household head for households with and without migrants. We find an inverted U-shaped relationship between expenses on wedding gifts and the age of household head. Younger and older families have lower expenses on wedding gifts. The average expenses on wedding gifts are higher for households with migrants at all age of household head. Figure 3(b) plots the migration probability for households with different expenses on wedding gifts 17. The migration probability decreases dramatically as the expenses on wedding gifts increase. [Figure 2 about here.] [Figure 3 about here.] 5 Results In this section, we test whether household expenses on wedding gifts have effects on job-related migration, and measure the magnitude of these effects if they are found. We first show the results using pooled probit model, then present the IV estimation of probit model. Finally, we discuss the validity of the identification by performing some sensitivity tests. 17 Mirant household is defined as households with at least one migrant. 17

5.1 Basic Results The first column in Table 3 reports the results using single-year measure of expenses on wedding gifts. The key coefficient for the effect of household expenses on wedding gifts is significantly negative. The other results confirm what have been found in existing research. Females are much less likely to migrate than males. Being male increases the probability of migration by 4.1 percentage point on average. Marriage reduces the probability of migration. Married people are 7.7 percentage point less likely to migrate compared to unmarried. With no schooling being the reference group, all other categories of education are statistically significant except college school or more, which suggests that education increases the likelihood of migration but the most educated people may prefer local non-farm employment. Age also has nonlinear impact on migration. Household registration type does not have significant impact on migration. In Table 3, most of the household characteristics are statistically significant suggesting that individual s migration decision is probably jointly made by the family as a whole. Household income per capita has significantly positive effect on migration. 10% increase in household income per capita increases migration probability by 4.3 percentage point. Cultivated land per capita which measures the potential income from farming has a negative effect on migration probability. Other things being equal, marginal productivity of agriculture is lower in the household with more family members which implies that the opportunity cost of migration for larger household is lower. Household size is positively and significantly related to the decision to migrate. 18

The number of preschool children and the number of elderly people in the household have a negative effect on migration, because family members should spend more time on taking care of them. Desirable villages, for example, having convenient telephone service or enterprises run by the village or being close to a train station, have significantly negative effect on migration. However, though the sign of the percentage of labor force working out of the village last year, which captures village s migration networks, is positive, it is insignificant. [Table 2 about here.] The results using average measure of expenses on wedding gifts are shown in column (2). We average the expenses over several years. As argued in Section 3, using average expenses can mitigate the endogeneity problem and eliminate the attenuation bias caused by measurement error. Column (2) shows that average expenses on wedding gifts has a significant and negative effect on migration. A 10 percent increase in average wedding gifts expenses results in a 0.12 percentage point decrease of migration probability. The predicted migration probability at the sample mean of the data is 0.037, which implies that a 10 percent increase in wedding expenditure leads to a 3.24 percent decrease of migration probability. The negative effect of local social networks may differ across gender groups. We estimate the determinants of migration for male and female workers separately and report the results in Table 4. Both the local social networks and migration networks outside the village have significant impact on migration for female, but these effects 19

are insignificant for male, which suggests that there are gender differences in the stock and returns of social capital in China. It seems that social networks, either local social networks or migration networks, matter more in the migration decision for female. Cultivated land per capita is more important in reducing male migration. But the number of preschool children and the number of elderly people decrease migration probability of female. Overall, women seem to be more constrained in migration than men do. The constraints may partly come from household joint decision in the allocation of labor. [Table 3 about here.] 5.2 IV Estimation The results of IV estimation are reported in Table 5 for treating simultaneity problem and measurement error problem. Since only fewer families report their expenses on funerals, the number of observations used in Table 5 falls sharply (1,369). To confirm that the results of IV estimation are not driven by sample selection, we also present the results from pooled probit model in column(3) using the same sample. Column (1) in Table 5 shows the results of the first stage of IV estimation. The instrumental variable is highly significant and has expected sign. The F-statistic in the first step regression is 14.62, suggesting that the lagged household expenses on funerals is not a weak instrument 18. The results in the first column also provide some insights into the black box of local social capital in rural China. 43.66% variation 18 Staiger and Stock (1997) suggest that an F-statistic of less than 10 is an indication of weak instruments. 20

in household expenses on wedding gifts can be explained by the variables in column (1). It shows that household expenses on wedding gifts are not significantly affected by individual characteristics. Households with larger size, higher income per capita, more land endowment have higher local social capital. As expected, because of the reciprocity principle of gift giving in a village the household expenses on wedding gifts are positively correlated in a village and the average household expenses on wedding gifts at village level is highly significant. Taking IV approach as the preferred specification, the effect of household expenses on wedding gifts on migration using IV approach is relatively larger in magnitude than that using single-year measure of expenses over several years. An increase in expenses on wedding gifts of 10 percent decreases migration probability by 1.5 percentage point. However, as shown in column (2), IV approach has larger standard error than probit model. The coefficients of other controlled variables are comparable to those using pooled probit model. In column(3), we report the results from pooled probit model using the same sample. The coefficient of the expenses on wedding gifts is significantly negative which rules out the potential sample selection problem. [Table 4 about here.] 5.3 Robustness Checks In this section, we perform several sensitivity tests (see Table 6). To test whether the results are robust to different measures of local social networks, we apply an 21

alternative measure of local social networks, that is, the number of village cadres in a family. Having family member as village cadres captures the existence of nepotism (Zhang and Li, 2003). Due to the social networks established through the village cadre position, cadre households have a greater likelihood of getting a non-farm job, working in town and village enterprises (TVEs), obtaining higher income (Walder, 2002; Zhang and Li, 2003) 19. To address the endogeneity of the number of village cadres in a family, we use six- or seven-year-lagged number of village cadres as an instrument. The results of the first stage regression show that the six- or seven-yearlagged variable is economically and statistically significant. The IV results provide evidence that individuals from families having village cadres are less likely to migrate. [Table 5 about here.] Similarly, we follow Kan (2007) and apply six- or seven-year-lagged household expenses on wedding gifts as an instrument for expenses in period t. And the results are shown in Column (3) and (4) in Table 6, providing further evidence that local social networks deter job-related migration in rural China. 6 Conclusion Using data from China Health and Nutrition Survey, this paper analyzes the determinants of job-related migration in rural China with special attention paid to the effect of local social networks, i.e., social ties in the source region. By using household expenses on wedding gifts as a measure of local social networks, we find that in 19 Local cadres often use their power by granting key positions in TVEs to their relatives and friends (Wu, 1994; Nee, 1996). 22

rural China local social networks have a significantly negative effect on labor migration after controlling for the characteristics of individual, household and village. And the negative effect of local social networks on migration is mostly driven by female sample. This paper provides new insights into the role of local social networks in migration, which helps to explain why rural people choose non-farm jobs over migration despite a large monetary premium from migration in rural China. Different types of local social networks may have different effects on the choice of migration, the distance of migration as well as the duration of stay in destination areas. Investigating the role of different types of local social networks in migration deserves further research. 23

References Yanjie Bian. Bringing strong ties back in: Indirect ties, network bridges, and job searches in china. American Sociological Review, 62(3):366 385, 1997. Yuyu Chen, Ginger Zhe Jin, and Yang Yue. Peer migration in china. NBER Working Paper No. 15671, January 2010. Quentin David, Alexandre Janiak, and Etienne Wasmer. Local social capital and geographical mobility. Journal of Urban Economics, 68(2):191 204, 2010. Xiaotong Fei. Peasant Life in China. London: Routledge & Kegan, 1948. Denise Hare. push versus pull factors in migration outflows and returns: Determinants of migration status and spell duration among china s rural population. Journal of Development Studies, 35(3):45 72, 1999. Francis L.K. Hsu. Americans and Chinese: Two Ways of Life. New York: Henry Schuman, 1953. Kwangkuo Hwang. Face and favor: the chinese power game. The American Journal of Sociology, 92(4):944 974, 1987. Kamhon Kan. Residential mobility and social capital. Journal of Urban Economics, 61(3):436 457, 2007. John Knight and Lina Song. Chinese peasant choices : Migration, rural industry or farming. Oxford Development Studies, 2(3):185 206, 2003. John Knight and Linda Yueh. The role of social capital in the labor market in china. Economics of Transition, 16(3):389 414, 2008. John Knight, Lina Song, and Huaibin Jia. Chinese rural migrants in enterprises: Three perspectives. Journal of Development Studies, 35(3):73 104, 1999. Zai Liang and Michael J. White. Internal migration in china, 1950-1988. Demography, 33(3):375 384, 1996. 24

Marios Michaelides. The effect of local ties, wages, and housing costs on migration decisions. Journal of Socio-Economics, 40(2):132 140, 2011. Victor Nee. The emergence of a market society: changing mechanisms of stratification in china. American Journal of Sociology, 101(4):908949, 1996. Rozelle Scott, Taylor J. Edward, and debrauw Alan. Migration, remittances, and agricultural productivity in china. The American Economic Review, 89(2):287 291, 1999. Larry A. Sjaastad. The costs and returns of human migration. The Journal of Political Economy, 70(5):80 93, 1962. Alden Jr. Speare, Frances Kobrin, and Ward Kingkade. The influence of socioeconomic bonds and satisfaction on interstate migration. Social Forces, 61(2):551 574, 1982. Douglas Staiger and James H. Stock. Instrumental variables regression with weak instruments. Econometrica, 65(3):557 586, 1997. Michael B. Toney. Length of residence, social ties, and economic opportunities. Demography, 13(3):297 309, 1976. Andrew G. Walder. Income determination and market opportunity in rural china, 19781996. Journal of Comparative Economics, 30(2):354 375, 2002. Harry X. Wu. Rural Enterprises in China, chapter The rural industrial enterprise workforce, pages 117 147. St. Martins Press, New York, 1994. Xiushi Yang. Household registration, economic reform and migration. International Migration Review, 27(4):796 818, 1993. W. Yip, S.V. Subramanian, A.D. Mitchell, D.T.S. Lee, J. Wang, and I. Kawachi. Does social capital enhance health and well-being? evidence from rural china. Social Science & Medicine, 64(1):35 49, 2007. 25

Xiaobo Zhang and Guo Li. Does guanxi matter to nonfarm employment? Journal of Comparative Economics, 31(2):315 331, 2003. Yaohui Zhao. Labor migration and earnings differences: The case of rural china. Economic Development and Cultural Change, 47(4):767 782, 1999a. Yaohui Zhao. Leaving the countryside: Rural-to-urban migration decisions in china. American Economic Review, 89(2):281 286, 1999b. Yaohui Zhao. Causes and consequences of return migration: Recent evidence from china. Journal of Comparative Economics, 30(2):376 394, 2002. Yaohui Zhao. The role of migrant networks in labor migration: The case of china. Contemporary Economic Policy, 21(4):500 511, 2003. 26

Table 1: Variables and Summary Statistics Definition Obs. Non-migrants Migrants Test of Difference No. of workers 11,259 10,147 1,112 % of total labors 100 90.12 9.88 Personal Characteristics Male(%)+ 1=male;0=female 11,259 49.79 58.54 *** Married(%)+ 1=married; 0= never married, divorced, widowed or separated 11,089 80.66 34.62 *** Age Current age in years 11,259 38.19 25.92 *** Primary school(%)+ completed primary school 11,124 27.29 24.32 ** Junior High school(%)+ completed junior high school 11,124 36.51 56.15 *** Senior High school(%)+ completed senior high school or technical school 11,124 14.57 14.01 College(%)+ College or higher 11,124 1.59 1.45 Education Years of schooling 10,711 6.82 8.35 *** Household registration(%)+ 1=urban;0=rural 11,105 24.72 14.18 *** Household Characteristics Social Capital Expenses on wedding during last 12 months(inflated to 2006) 8,195 803.15 683.83 *** Income per capita Adjusted household income per capita last year(inflated to 2006) 10,810 3697.73 4357.11 *** Household head age Age of household head 10,921 46.54 49.64 *** Cultivated land per capita Land cultivated per capita last year (Unit: Mu/capita) 7,050 2.04 1.43 *** Household size No. of household members 10,992 4.22 4.77 *** # Male laborers No. of male (16<age<60) 11,259 1.63 1.92 *** # kids under 6 No. of kids under 6 years 11,259 0.28 0.24 ** # kids 6 to 12 years old No. of kids 6 to 12 years 11,259 0.39 0.27 *** # elderly 60 or older No. of elders 60 years or older 11,259 0.26 0.28 Village Characteristics Migration Rate % of the labor working out of town for more than 1 month last year in this village 10,736 25.43 28.43 *** Telephone service(%)+ 1=The village has convenient telephone service; 0=otherwise 11,045 84.03 79.02 *** Agriculture % work force engated in agriculture 10,882 53.07 61.04 *** Bus(%)+ 1= The village has a bus stop (or long distance bus stop); 0=otherwise 10,981 61.83 59.91 Train(%)+ 1= The village is near a train station; 0=otherwise 10,979 16.52 9.61 *** Stone road(%)+ Stone, gravel or mixed material road 10,094 26.51 28.24 Paved road(%)+ Paved road 10,094 48.49 44.97 ** Non-farm opportunity(%)+ 1= village has enterprises run by the village; 0=otherwise 10,871 41.14 35.89 *** Note: An individual is defined as a migrant if s/he is currently not living at home and seeking employment elsewhere. t-test of difference between non-migrants and migrants. + dummy variable. Variables are summarized using data in 1997 and 2000. *, **, and *** respectively represent statistically significant at the 10%, 5% and 1% level. 27

Table 2: The prevalence of local social networks Province (1) (2) Household Expenses on Having family member as wedding gifts offical cadre or village cadre (%) Guangxi 448.27 (929.49) 3.49 Guizhou 896.77 (1130.95) 4.20 Heilongjiang 1477.97 (1410.02) 5.17 Henan 778.54 (847.25) 5.73 Hubei 819.22 (722.79) 4.69 Hunan 972.40 (1251.05) 6.31 Jiangsu 793.78 (957.01) 5.31 Liaoning 1497.37 (1349.15) 5.85 Shandong 723.53 (1135.59) 6.94 OVERALL 938.74 (1145.28) 5.26 Note: (1)unit: yuan. Standard deviation in parentheses. (2) % families having family member as official cadre or village cadre. 28

Table 3: Basic Results Dependent variable: migration (1) Single (2) Average log(household expenses on wedding gifts) -0.011 (0.0050) log(average household expenses on wedding gifts) -0.012 (0.0057) Male(+) 0.041 0.037 (0.0082) (0.0075) Married(+) -0.077-0.079 (0.014) (0.013) Age 0.0053 0.0037 (0.0056) (0.0049) Age squared -0.00019-0.00016 (0.000088) (0.000073) Primary school(+) 0.062 0.064 (0.018) (0.016) Junior high school(+) 0.055 0.065 (0.018) (0.016) Senior high school(+) 0.044 0.054 (0.022) (0.020) College(+) 0.040 0.020 (0.046) (0.046) Household registration type(+) 0.030 0.022 (0.019) (0.017) log(household income per capita) 0.043 0.039 (0.0069) (0.0066) Cultivated land per capita -0.0096-0.0083 (0.0033) (0.0030) Household head age 0.011 0.0095 (0.0040) (0.0037) Household head age squared -0.000088-0.000069 (0.000041) (0.000037) Household size 0.023 0.020 (0.0049) (0.0041) # Male laborers -0.031-0.025 (0.0086) (0.0074) # Kids under 6 years -0.023-0.0090 (0.011) (0.0089) # Kids 6 to 12 years -0.0089-0.0053 (0.0095) (0.0083) # Elderly 60 years or older -0.025-0.020 (0.0099) (0.0090) Telephone service(+) -0.039-0.043 (0.014) (0.011) Migration rate 0.00023 0.00022 (0.00027) (0.00023) Agriculture 0.00040 0.00042 (0.00027) (0.00023) Enterprise(+) -0.023-0.022 (0.011) (0.0099) Bus(+) 0.020 0.023 (0.013) (0.012) Train(+) -0.035-0.037 (0.020) (0.017) Stone road(+) 0.0015-0.0030 (0.012) (0.011) Paved road(+) -0.016-0.022 (0.012) (0.011) log(average expenses on wedding gifts at village level) 0.015 0.0080 (0.010) (0.0087) Provincial fixed effect Yes Yes Year fixed effect Yes Yes Cohort fixed effect Yes Yes Obs. 4,588 5,729 Log-likelihood -1080.68-1379.25 Pseudo R2 0.35 0.33 Note: Coefficients for the Probit model are average marginal effects. Robust standard errors adjusted for clustering at household level are in parentheses. (+) for discrete change of dummy variable from 0 to 1. *** significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. 29

Table 4: Migration by Gender (1) (2) Male Female log(average household expenses on wedding gifts) -0.0092-0.013 (0.0086) (0.0062) Married(+) -0.056-0.089 (0.019) (0.013) Age 0.0065 0.0072 (0.0075) (0.0066) Age squared -0.00022-0.00022 (0.00011) (0.00011) Primary school(+) 0.057 0.035 (0.027) (0.021) Junior high school(+) 0.045 0.047 (0.026) (0.022) Senior high school(+) 0.037 0.035 (0.031) (0.026) College(+) -0.025 0.020 (0.063) (0.069) Household registration type(+) 0.00071 0.038 (0.025) (0.022) log(household income per capita) 0.051 0.024 (0.0088) (0.0080) Cultivated land per capita -0.014-0.0024 (0.0048) (0.0026) Household head age 0.013 0.0044 (0.0052) (0.0042) Household head age squared -0.00010-0.000024 (0.000053) (0.000041) Household size 0.013 0.015 (0.0063) (0.0047) # Male laborers -0.014-0.0063 (0.011) (0.0083) # Kids under 6 years 0.0028-0.017 (0.013) (0.0098) # Kids 6 to 12 years 0.0066-0.012 (0.012) (0.0095) # Elderly 60 years or older -0.0071-0.021 (0.013) (0.010) Telephone service -0.076-0.0090 (0.016) (0.013) Agriculture -0.00039 0.0013 (0.00032) (0.00028) Migration rate -0.000067 0.00059 (0.00032) (0.00027) Enterprise(+) -0.040-0.0035 (0.013) (0.011) Bus(+) 0.045 0.0050 (0.017) (0.013) Train(+) -0.039-0.037 (0.025) (0.017) Stone road(+) 0.0080-0.016 (0.016) (0.013) Paved road(+) -0.018-0.022 (0.016) (0.012) log(average expenses on wedding gifts at village level) 0.013 0.0043 (0.012) (0.0096) Provincial fixed effect Yes Yes Year fixed effect Yes Yes Cohort fixed effect Yes Yes Obs. 2,925 2,804 Log-likelihood -841.54-474.70 Pseudo R2 0.26 0.47 Note: Dependent variable is migration. Coefficients for the Probit model are average marginal effects. Robust standard errors adjusted for clustering at household level are in parentheses. (+) for discrete change of dummy variable from 0 to 1. *** significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. 30

Table 5: IV Estimation (1)First stage (2)Second stage (3)Probit Instrumental Variable log(lagged expenses on funerals) 0.087 (0.036) log(household expenses on wedding gifts) -0.15-0.023 (0.085) (0.010) Male(+) -0.027 0.044 0.050 (0.026) (0.015) (0.014) Married(+) 0.027-0.094-0.100 (0.084) (0.027) (0.024) Age 0.016 0.0056 0.0040 (0.027) (0.010) (0.0099) Age squared -0.000034-0.00017-0.00018 (0.00032) (0.00016) (0.00016) Primary school(+) 0.0061 0.13 0.14 (0.060) (0.033) (0.032) Junior high school(+) 0.11 0.11 0.10 (0.072) (0.033) (0.034) Senior high school(+) 0.17 0.15 0.13 (0.13) (0.041) (0.039) College(+) -0.015 0.096 0.10 (0.29) (0.066) (0.078) Household registration type(+) 0.093 0.016 0.0086 (0.15) (0.033) (0.031) log(household income per capita) 0.20 0.074 0.052 (0.059) (0.020) (0.012) Cultivated land per capita 0.027-0.020-0.024 (0.012) (0.0090) (0.0084) Household head age 0.028 0.017 0.014 (0.032) (0.0085) (0.0082) Household head age squared -0.00034-0.00015-0.00011 (0.00033) (0.000088) (0.000082) Household size 0.11 0.030 0.017 (0.036) (0.011) (0.0075) # Male laborers -0.0061-0.041-0.041 (0.065) (0.014) (0.015) # Kids under 6 years -0.054-0.025-0.019 (0.087) (0.021) (0.019) # Kids 6 to 12 years 0.027 0.012 0.0092 (0.072) (0.018) (0.018) # Elderly 60 years or older -0.096-0.060-0.047 (0.088) (0.020) (0.018) Telephone service(+) 0.017-0.036-0.037 (0.11) (0.029) (0.024) Agriculture -0.0022 0.00026 0.00044 (0.0027) (0.00076) (0.00055) Migration rate -0.00097-0.00061-0.00048 (0.0024) (0.00052) (0.00047) Enterprise(+) -0.14-0.026-0.017 (0.093) (0.024) (0.019) Bus(+) 0.14 0.033 0.015 (0.098) (0.025) (0.020) Train(+) -0.0062-0.10-0.11 (0.12) (0.039) (0.035) Stone road(+) 0.015-0.023-0.025 (0.11) (0.026) (0.023) Paved road(+) 0.10-0.010-0.028 (0.12) (0.028) (0.022) log(average expenses on wedding gifts at village level) 0.83 0.095-0.016 (0.075) (0.077) (0.016) Provincial fixed effect Yes Yes Yes Year fixed effect Yes Yes Yes Cohort fixed effect Yes Yes Yes Obs. 1,369 1,369 1,369 Log-likelihood - -1880.97-299.62 Pseudo R2 0.44-0.41 Note: Dependent variable is migration. Coefficients for the Probit model are average marginal effects. Robust standard errors adjusted for clustering at household level are in parentheses. (+) for discrete change of dummy variable from 0 to 1. *** significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. 31

Table 6: Robustness Checks (1) (2) (3) (4) Instrumental variable 6- to 7-year-lagged number of cadres at household level 0.21 log(6- to 7-year-lagged expenses on wedding gifts) No. of village cadres at household level log(household expenses on wedding gifts) (0.053) -0.17 (0.10) 0.16 (0.030) -0.065 (0.040) Male(+) -0.0084 0.030-0.036 0.032 (0.0043) (0.0087) (0.020) (0.012) Married(+) 0.012-0.082 0.073-0.076 (0.012) (0.015) (0.058) (0.019) Age 0.0051 0.0051-0.015 0.0030 (0.0038) (0.0057) (0.021) (0.0074) Age squared -0.000074-0.00018 0.00012-0.00016 (0.000046) (0.000084) (0.00026) (0.00011) Primary school(+) -0.011 0.074 0.080 0.070 (0.0081) (0.019) (0.051) (0.023) Junior high school(+) 0.022 0.079 0.15 0.067 (0.012) (0.019) (0.066) (0.025) Senior high school(+) 0.024 0.076 0.25 0.075 (0.020) (0.023) (0.091) (0.032) College(+) 0.022 0.076 0.15 0.11 (0.060) (0.053) (0.30) (0.063) Household registration type(+) 0.063 0.026 0.21 0.040 (0.042) (0.021) (0.10) (0.027) log(household income per capita) 0.014 0.041 0.12 0.051 (0.0071) (0.0079) (0.039) (0.011) Cultivated land per capita -0.0066-0.0057 0.060-0.0024 (0.0043) (0.0047) (0.022) (0.0075) Household head age 0.0043 0.0089 0.022 0.014 (0.0039) (0.0050) (0.022) (0.0064) Household head age squared -0.000038-0.000061-0.00030-0.00010 (0.000041) (0.000049) (0.00022) (0.000063) Household size 0.0022 0.019 0.076 0.029 (0.0046) (0.0046) (0.028) (0.0068) # Male laborers -0.0045-0.029 0.053-0.042 (0.0075) (0.0085) (0.041) (0.012) # Kids under 6 years 0.0011-0.0059 0.035-0.026 (0.017) (0.011) (0.069) (0.015) # Kids 6 to 12 years 0.0018-0.0091-0.046-0.016 (0.011) (0.010) (0.048) (0.013) # Elderly 60 years or older 0.0082-0.020-0.055-0.041 (0.013) (0.010) (0.055) (0.014) Telephone service(+) 0.0028-0.031 0.080-0.030 (0.017) (0.015) (0.085) (0.021) Agriculture -0.00020 0.00030 0.0029 0.00032 (0.00033) (0.00028) (0.0016) (0.00037) Migration rate -0.00029 0.0000051 0.00076-0.00031 (0.00031) (0.00027) (0.0018) (0.00035) Enterprise(+) -0.025-0.029-0.097-0.026 (0.012) (0.012) (0.064) (0.015) Bus(+) 0.014 0.027-0.052 0.031 (0.018) (0.014) (0.074) (0.016) Train(+) 0.033-0.041-0.0029-0.073 (0.031) (0.019) (0.10) (0.026) Stone road(+) 0.025-0.010 0.16-0.0073 (0.016) (0.014) (0.086) (0.019) Paved road(+) 0.014-0.026-0.0098-0.027 (0.013) (0.013) (0.077) (0.016) log(average expenses on wedding gifts at village level) 0.83 0.051 (0.057) (0.038) Provincial fixed effect Yes Yes Yes Yes Year fixed effect Yes Yes Yes Yes Cohort fixed effect Yes Yes Yes Yes Obs. 4,630 4,630 2,633 2,633 Log-likelihood - -755.08 - -3858.46 Pseudo R2 0.11-0.39-32 Note: Dependent variable is migration. Coefficients for the Probit model are average marginal effects. Robust standard errors adjusted for clustering at household level are in parentheses. (+) for discrete change of dummy variable from 0 to 1. *** significant at the 1% level; ** significant at the 5% level; * significant at the 10% level.

(a) Migration Rate in Rural China (1997-2006) (b) Migration Rate by Age and Gender Figure 1: Migration Rate in Rural China 33

Figure 2: Share of Household Expenses on Wedding Gifts in Household Net Income (%) 34

(a) Expenses on wedding gifts for households with and without migrants (b) Migration probability of household members and expenses on wedding gifts Figure 3: Migration Probability and Expenses on wedding-gifts 35