Meeting Friends of Friends: Scientists, Teenagers, and International Trade

Fafchamps, Goyal and van der Leij (2010) study the formation of new collaborations between academic economists over 20 years, from 1980 to 1999. They find that proximity in the existing coauthorship network has a strong impact on the likelihood to form a new collaboration. For instance, being at a network distance 2 instead of 3 raises the probability of initiating a collaboration by 27%. Building confidence in causal interpretations with observational data is always a challenge; the authors skillfully rise to it in their analysis. In particular, they control for any time-invariant characteristics of pairs of economists and for main time varying confounding factors. They also show that network proximity affects first collaborations but not subsequent collaborations after the first. Overall, their results are strongly suggestive of referral effects at work in scientific networks. A common coauthor may introduce and vouch for a potential new collaborator.

In an influential study, Jackson and Rogers (2007) provide some indirect but substantial empirical evidence about the role played by the existing network in shaping up new links. They develop a neat theoretical model of growing, stochastic network formation. New nodes are born sequentially. They first meet some existing nodes at random and then meet some friends of these first contacts. The authors solve the model by a clever use of analytical techniques imported from physics. They show that the model generates five key empirical features observed in real social networks : low diameter, high clustering, fat tails in degree distribution, positive degree-degree correlation and negative degree-clustering correlation. This type of model, founded on network-based meetings, currently provides the only known parsimonious way to generate all of these realistic features. The authors then investigate some consequences of network-based meetings and find that it tends to generate inequality in connections. Well-connected agents are more easily accessed through the network and hence gain new connections at a faster rate. This mechanism likely plays an important role in explaining the emergence of global winners in network contexts: influential scientists, highly cited papers, very popular teenagers or dominant exporters.

More recently, researchers have started to investigate how network-based meetings may interact with individual characteristics. In particular, we know that homophily is pervasive in network settings: Links are generally much more likely to connect similar agents. How does homophily interact with network-based meetings? Does forming new links through common friends dampen or amplify assortative tendencies? Bramoullé et al. (2012) extend the analysis of Jackson and Rogers (2007) to a setup where agents are of different types and meetings may be type-biased. They show that network-based meetings actually tend to reduce biases in random meetings. Friends' friends generally form a more diverse crowd than direct friends. This typically generates a negative relationship between degree and homophily. As a node ages and gets more connections, it gets a larger share of his new connections through friends of friends and hence from diverse nodes. The authors then look at citation patterns for articles published in journals from the American Physical Society between 1985 and 2003. They find that, indeed, the proportion of citations that an article obtains from other articles in the same field generally decreases as the paper ages and becomes more cited. This is consistent with the theoretical model and with the practice, common in academia, of learning about relevant papers because they are cited in known sources.

In their working paper version, Bramoullé and Rogers (2009) also look at gender-based homophily in friendship nominations among teenagers, based on the first wave of the National Longitudinal Survey of Adolescent Health (Add Health). This survey collected detailed information on about 90 000 teenagers in American middle and high schools in the academic year 1994 – 1995. In particular, most students within schools were sampled and interviewed teenagers were asked to name up to five best male friends and five best female friends. Authors find a strong decreasing relationship between homophily and popularity. For instance, the probability to receive a nomination from a friend of the same gender is equal to 75% for teenagers who get a unique nomination but drops to 51% for teenagers with ten nominations. This pattern holds for boys and girls separately and provides some indirect evidence for the importance of network-based meetings in teenager socialization.

In two recent and far reaching papers, Thomas Chaney provides what may be, to date, the most important economic application of the idea that new connections are formed through the existing network. Chaney (2014) considers international trade and develops a new model to explain how exporters grow and get access to distant markets. His model can be viewed as a spatial variant of the model of Jackson and Rogers (2007). Firms are located at fixed positions in space. To be able to trade, they first must find foreign trading partners and this search takes place in two ways. Some firms meet at random and these random meetings are spatially biased. Then, once a firm has acquired a network of contacts in another location, it can remotely search for new trading partners from these locations. As a firm grows, it thus gets connected with trading partners further away and exports over longer distances. The author shows that the model generates two key predictions on the dynamics of trade and finds strong support for both predictions using data on French firms from 1986 to 1992. First, when a firm exports to more countries in year t, it is more likely to enter a new market in year t+1. This is consistent with the fact that under network-based linking, nodes with more links get new connections at a faster rate. Second, a firm is more likely to enter new markets geographically closer to the countries to which it is currently exporting. These results provide strong indirect evidence for the fact that firms build upon their existing networks of consumers, suppliers and business partners to establish new connections.

In a recent working paper, the author takes these ideas to the next level (Chaney 2013). He shows that such network-based modeling of informational frictions can actually explain the gravity equation, a key empirical regularity of international trade. The gravity equation states that bilateral exports between countries are proportional to economic size and inversely proportional to geographic distance. It has received strong empirical support in the literature on international trade and seems remarkably stable over time. However, economists have been struggling to provide a satisfactory, microfounded explanation for this empirical regularity. Chaney's study may well close this gap in our understanding and will likely generate much further interest into the economics of social networks.

References

• Marcel Fafchamps, Sanjeev Goyal and Marco J. van der Leij. 2010. "Matching and Network Effects" Journal of the European Economic Association, 8 (1): 203-231.
• Yann Bramoullé, Sergio Currarini, Matthew O. Jackson, Paolo Pin and Brian W. Rogers. 2012. "Homophily and Long-Run Integration in Social Networks" Journal of Economic Theory, 147 (5): 1754-1786.
• Yann Bramoullé and Brian W. Rogers. 2009. "Popularity and Diversity in Social Networks" CIRPÉE working paper 09-03.
• Thomas Chaney . 2014. "The Network Structure of International Trade" American Economic Review forthcoming.
• Thomas Chaney. 2013. "The Gravity Equation in International Trade: an Explanation" working paper Toulouse School of Economics.
• Matthew O. Jackson, and Brian W. Rogers. 2007. "Meeting Strangers and Friends of Friends: How Random are Social Networks?" American Economic Review, 97 (3): 890-915.