The Size of Individual Social Networks and Rural-Urban Migrants’ Wages—Findings from a Survey of Migrants in China’s 6 Provinces and 12 Cities

1 Introduction

Among the literature of labor economics, roles of social capital and social networks in the labor market have grabbed considerable interest. One prominent example of how social networks play a role is that social networks can significantly facilitate job search. Studies over the recent decades show that throughout the world at least 30%~50% of job seekers got their jobs through the help of relatives or friends (Rees, 1966; Addison and Portugal, 2002; Wahba and Zenou, 2005; Zhang et al., 2008; Topa, 2011).

The rural-urban migrants in China is a group in the labor market among which social networks play a particularly important role. The reason is two-folds. First of all, China is traditionally recognized as a “Guanxi society” where social connections and personal relationships are in a central position. Though a comprehensive economic reform aiming at establishing a modern market economy is taking place, such a traditional feature is still embedded in the institutional transmission and plays an unneglectable role in market transactions^[1] (Zhang and Zhao, 2015). On the other hand, compared with natives, the migrant workers as “newcomers” are usually lack of access to job information, and therefore tend to find jobs with the help of relatives, friends and villagers who have been settled in the same city (Borjas, 1992; Munshi, 2003, 2011). These two features of the rural-urban migrants in China justify the importance of social networks for this group. Furthermore, investigating the specific role that social networks play as well as the underlying mechanism is the key for understanding the interaction between the market and the society in the process of economic reform and how the migrant workers make a prominent contribution in the rapid urbanization and economic growth in China.

In this paper, we attempt to identify the effect of individual network size on rural-urban migrants’ wages in China. The proxy we use for the size of social networks is “the number of relatives and friends in the city where the migrant works”, which limits the social network to the working cities and is more likely to be related to the migrant worker’s job seeking and wages. We find that the significant positive correlation between migrant workers’ wages and their individual network size is an overestimate due to endogeneity of social networks. Once the endogeneity problem is corrected by the IV approach, there’s no evidence supporting a positive mean causal effect of social network size on wage. A further exploration in quantile estimations, on the other hand, shows that although network size has no significant impact at most quantiles above 40%, it does benefit the lower end migrants, which confirms the conjecture that lower-income migrants might rely more on their social networks.

This paper has two major contributions to the literature. Firstly, we introduce the provincial-level geopolitical immigration network, the number of samples from the same province in the destination city, as the instrument for individual social networks to correct potential endogeneity problem. As will be discussed in more details later, many instruments used in the existing literature are not plausible enough in the exclusive condition (probably correlated with unobserved disturbance). Our instrument is not related to the characteristics of unobservable capacity at the individual or family level, and thus corrects the problem of omitting variables. Furthermore, we propose a modified version in order to rule out the potential selection problem at the group level, providing robust evidence that the size has no mean causal effect. Secondly, we investigate the quantile effect of social network size on migrants’ wages. To the best of our knowledge there are not many papers on this topic (Grootaert, 1999; Woolcock and Narayan, 2000; Zhou, 2012), and most of the existing ones have not considered the endogeneity problem as we do. Our result based on the control function approach (Lee, 2007) shows that the quantile estimations might be misleading without correcting endogeneity: underestimating the coefficients for lower quantiles and overestimating the coefficients for higher quantiles.

The rest of this paper proceeds as follows. Section 2 reviews related literature in both general contexts and in particular China’s rural-urban migration. Section 3 introduces the data and reports the descriptive statistics of relevant variables. Section 4 discusses our main empirical strategies. Section 5 reports the empirical results and Section 6 concludes.

2 Literature Review

This paper contributes to a large class of literature in the field of social networks and China’s rural-urban migration. Despite the richness of existing literature, there seems not to be a generalized framework, nor a widely accepted agreement on the impact of social networks on labor market outcomes. This is due to three challenges faced by studies in this topic: complexity and conceptual vagueness of the term “social networks” (Portes, 2000; Manski, 2000; Mouw, 2006), richness of underlying mechanisms that’s hard to be characterized in rigorous economic models, and difficulty in causal effect identification given the nature of social network variable and limited data resources (Durlauf, 2002).

First of all, the concept of “social networks” is multi-dimensional. According to Nahapiet and Ghoshal (1998), the impact of social networks can be jointly decided by the structural dimension (overall pattern of connections between actors), relational dimension (interaction and intimacy) and cognitive dimension (reciprocal behavior and common values). It is difficult to make a general analysis and a convincing comparison between different empirical results if social network proxies in different dimensions are used (Durlauf and Fafchamps, 2005). In practice, current empirical literature uses different proxies for social network variables and has not reached an agreement on the network effect on wages, nor a general framework to analyze the possible economic mechanism. Some studies find no significant network effects on workers’ wages (Mouw, 2003; Liu and Zhang, 2007; Zhang et al., 2008; Zhang and Lu, 2009), while some others find a significantly positive effect (Munshi, 2003; Knight and Yueh, 2008; Ye et al., 2012; Hensvik and Nordstrom, 2013; Ye and Wu, 2014; Burks et al., 2015). Even if we restrict the emphasis on the size of individual social network size, the proxy variables are still different case by case. For example, Munshi (2003) finds that the expansion of the migration network from the same source-community could help Mexican immigrants increase the probability of getting non-agricultural jobs in the United States and also get higher wages. Knight and Yueh (2008) on China rural-urban migrants and Amuedo-Dorantes and Mundra (2007) on Mexican-American immigrants both use “the number of relatives and friends of the family” as a proxy for social networks and find significant impacts on both job seeking and wages. On the other hand, Zhang and Lu (2009) investigate the impact of “the number of relatives and friends to whom an individual has presented gifts”, which reflects not only the size of individual or familial social networks but also the willingness to invest in social networks, but find little impact of size on wages.

Secondly, the different dimensions of social networks may have distinct channels to impact outcome variables (Zhang and Li, 2003). In economic literature, most formal theoretical models on the wage effect of social networks focus on three categories: overcoming inefficiency from information asymmetry through job referral, information advantage in job search, and peer effect on productivity. The seminal work by Montgomery (1991, 1992) shows that when heterogeneous worker productivity is private information, the recommenders are willing to act as a screening mechanism for the sake of their own career development, resulting in the employer’s willingness to offer more favorable positions to the referred job seekers. Mortensen and Vishwanath (1994) further demonstrate the information effect: even with identical job seekers, those with a higher probability of obtaining offer information through employed contacts tend to earn more due to more favorable wage information.^[1] Besides the information effect, Delattre and Sabatier (2007), Ye et al. (2012) and Lindquist et al. (2015) investigate the productivity effects of peer pressure, showing that high quality co-worker (or acquaintance) social networks can motivate workers to spend more effort and thus obtain higher wages.^[2] Despite the richness of recent theoretical research on network effect on wages, most of the existing models focus on referral using or quality of co-worker networks. The impact of network size, on the other hand, is hard to formalize. Wahba and Zenou (2005) develop a model to build a link between the size and the use of social networks. They find that conditional on being employed, the probability of finding a job through social networks increases with the size of the networks, and a survey from Egypt provides empirical evidence supporting their conclusions. In a similar manner, Munshi (2003) uses a statistical model involving referrals (analogous to the Montgomery setting) and confirms that the expansion in network size can improve the labor market outcome of Mexican immigrants in the U.S.

Finally, though some of the earlier studies pay little attention to the endogeneity problem of social networks (Lin, 2001; Mouw, 2003), it has soon become a central issue in the empirical literature.^[1] Instrument variable (IV) approach is the most commonly used identification strategy with three types of instruments: natural conditions (Munshi, 2003; Zhang and Zhao, 2015), family political background (Zhang and Lu, 2009) and the social and cultural background (Narayan and Pritchett, 1999; Ye and Zhou, 2010; Ye and Wu, 2014). Munshi (2003) uses the lagged rainfall of the migrants’ origin as an instrument for the Mexican-American community migration networks. Such exogenous shocks effectively satisfy the exclusiveness and correlation conditions while imposing strong data and background limitations on external applications. Zhang and Zhao (2015) use the distance between the capital city of the home province and the working city as the instrument to study the impact of the size of relatives and friends on the migrants’ choices of self-employment, but the instrument is difficult to apply to the study on wages because of the migrating self-selection bias brought by omitting variables.^[2] In terms of the families’ political, social and cultural backgrounds, Zhang and Lu (2009) use the political classes of the spouse family during the Cultural Revolution as an instrument for the number of relatives and friends. But family political classes are often related with economic background, genetic capacity, family education and other unobservable variables. Even the spouse’s family background is also related if we take the assortative mating into account. Ye and Wu (2014) use the Spring Festival greetings as an instrument by arguing that this behavior is more influenced by customs and could been seen as exogenous. However, it is difficult to rule out the possibility of inverse causality: migrants with high income are more likely to give a New Year’s call than those with low income due to vanity concern.

3 Data and Descriptive Statistics

In 2009, one of the authors conducted a large-sample migrants survey in 12 cities across four major urbanized regions: the Yangtze River Delta (Jiangsu and Zhejiang Province), the Pearl River Delta (Guangdong province), the Chengdu-Chongqing region (Sichuan Province and Chongqing Municipality), and the Bohai Bay Area (Hebei and Shandong Province). We randomly select one megalopolis (urban population larger than 2 million), one large city (500000~2 million), and one small-and medium-sized city (<500000) in each of the four urbanized regions. Due to the huge number of migrants in the megalopolis, we only sample in one urban district randomly selected in the megalopolis, while in the large and small- and medium-sized cities we target all urban districts in the city as our sampling frame. Then we randomly select 200 migrants in each city (2400 migrants in total) from the migrant registration list provided either by the local Public Security Bureau or by local government migrant administrative agency.^[1] Migrants are defined as people whose Hukou is not registered in the city in which they lived at the time of survey, and they have left their Hukou registration places for more than three days.^[2] Finally, 2299 valid samples were obtained. The survey collected information about migrants’ various aspects of livelihood, including demography, employment, income, social network, and so on.

The main variables in this study include wage, social network, human capital and other personal characteristics. Table 1 reports the descriptive statistics of the main variables. The dependent variable is the logarithmic hourly wage of the migrant worker in 2008, and the average hourly wage is 7.39 yuan (RMB). 56.6% of the samples are male, the average age is 32 years-old, and 62% of the samples are married. In terms of social networks, including the number of friends/relatives in the same city and the total number, a migrant worker averagely has 8 friends and 4 relatives in the city where they work. In terms of human capital, the migrants’ average years of schooling is 9 years (roughly equivalent to junior high school) and the working experience in the current city is 7 years on average. In addition, the proportion of CPC party members is only 6%, which is consistent with our experience and findings in the literature.

Table 2 reports the statistics of wages and the size of social networks of migrant workers in the four major urbanized regions. It can be seen that the average wage of the migrant workers in the Bohai Bay Area are relatively low and the size of social networks is also smaller, while the migrant workers of the Pearl River Delta have the highest average wage and the largest network of friends and relatives.

4 Empirical Strategy

5 Empirical Results

5.3 Quantile Regression Results

At last we perform the quantile regression results. Figure 1 shows the coefficient estimates of total network size as well as the confidence intervals at 90 quantiles from 0.05 to 0.95. The upper figure is the result from quantile regressions without considering endogeneity problem, while the lower figure takes network size to be endogenous and uses control function approach proposed by Lee (2007). In this section we use the modified instrument to perform the first stage regression, which is no different from the first stage in the last section. Then the predicted residual is used as one of the control variables to perform quantile regressions. The first observation is that at high wage quantiles (above 0.8) estimations in both specifications are highly chaotic. This means that for the most successful rural-urban migrants, individual network size has no explanatory power. The second observation is that the coefficients in the first specification are steady around 0.025, which is quite similar to the OLS estimation. Meanwhile coefficients in the CFA specification are in general decreasing along quantiles from positive values to negative. To illustrate the significance level of the quantile regression, we present coefficients for below-mean quantiles in Table 5. It’s clear that at the lower end (below 25%), social network size plays a significantly important role to help migrant workers get higher wage. In contrast, for above-mean quantiles, we cannot obtain coefficients that are significantly different from 0.

Figure 1

Coefficients of Network Size at 0.05~0.95 Quantiles

Table 5

Coefficients of Network Size at Below-Median Quantiles

Quantiles	0.05	0.10	0.15	0.20	0.25	0.30	0.35	0.40	0.45	0.50
ln (total number)	0.701* (0.359)	0.684*** (0.233)	0.397* (0.203)	0.278 (0.184)	0.306* (0.167)	0.242 (0.159)	0.186 (0.142)	0.0526 (0.134)	0.00859 (0.130)	−0.0741 (0.132)

The results show that the wage return to network size would be underestimated for lower quantile migrants and overestimated for higher quantile migrants, if we don’t consider endogeneity. This means that lower income migrants might rely more on their social networks than higher income migrants. One possible interpretation is that social networks might be substitutes to human capital and unobservable working abilities. Lower income migrant workers are usually associated with lower unobserved ability and have less capability of collecting job information. It’s more difficult to obtain a higher wage job position by their own means, thus they rely more on the social networks they have. While the case is much different for higher ability migrant workers, even though they are more likely to obtain more and better social capital. Some of them might rely heavily on a larger number of friends or relatives to get more income, especially the successful self-employed migrants, while some others get the high-pay job through their high productivity in a remote city from their hometown where they have few friends or relatives to ask for help.

6 Conclusions

Based on a survey of migrants in China’s 6 provinces and 12 cities, this paper empirically studies the impact of the size of individual social networks on the migrant workers’ wages. Our identification strategy is to use the aggregate migration network as the instrument for individual social networks. Based on the empirical results from IV 2SLS estimations and quantile estimations with endogenous explanatory variable, this paper draws the following conclusions. Firstly, the size of individual social networks and wages are significantly correlated. Secondly, after dealing with the endogeneity problem by IV approach, we cannot find evidence for significant wage effect of network size. Anderson-Rubin robust test further confirms this conclusion, no matter the threat of weak IV exists or not. Our result in general agrees with the conclusion of Zhang and Lu (2009), but differs from most other existing studies that find positive network size effect on wages (Munshi, 2003; Knight and Yueh, 2008). Thirdly, a further exploration in quantile estimations based on control function approach shows that at lower wage quantiles, migrants can benefit from larger individual networks, while this doesn’t hold for migrants at higher wage quantiles.

It’s worth noting that the insignificant results in this paper do not imply that social networks have little effect on the performance of migrants in the urban labor market. Firstly, the key network variables used in this study do not include the migrants’ contacts among urban natives, nor the “neighborhood” and “community” that are valued in the literature of social interaction and peer effects. Secondly, the size is only one dimension of social networks. It’s probable that the strength and quality of social networks are playing an important role in the labor market performance, and whether to use social networks to find a job may also matter. Thirdly, wage is only one labor market outcome. Effects of social networks on other variables such as referral using, turnover rates, social integration and length of unemployment are worthy of further investigation.

So, we believe that further studies (especially in China’s labor market) could focus on the following directions. Firstly, it’s valuable to investigate the impact of social networks on the selection patterns of unobservable ability and observable education in the migration (Mckenzie and Rappoport, 2007). The efforts on this issue will help us better understand the impact of social networks on individuals as well as the income inequality in both rural and urban areas caused by the migration. Secondly, it’s also promising to study the investment choice and accumulation process of social capital as well as their interaction with individual income and economic status. The existing studies, especially in sociology, pay a lot of attention on acts such as gift giving or Spring Festival greetings as ways of social capital investment. Further effort should be integrating these network investment behaviors into formal economic models based on the fruitful area of social interaction literature. Thirdly, it might be investigating mechanism of the network effect in depth making use of solid theory foundation and structural estimation. Theoretical literature such as Montgomery (1991) have developed insightful models in early years, while the empirical breakthrough has not appeared until the past decade (Cingano and Rosolia, 2012; Kramarz and Skans, 2014; Schmutte, 2015; Brown et al., 2016; Hensvik and Nordstrom, 2016). Most of this literature is based on company-employee matching data, and can be helpful in understanding the role of job referrals on the probabilities of job acceptance, wages, turnover rates, productivity, and welfare implications. More works on this type of research with corresponding data-set in China are in need to more precisely and rigorously explain the socio-economic effect of social networks in both urban labor market and rural-urban groups.