Pilot Projects | Project Supplements

Principal Investigator: Ethan Ligon

There is a burgeoning interest in social networks among economists. A quick summary of the theoretical literature suggests that the structure of a social network can matter for a wide variety of outcomes; a quick summary of the empirical literature suggests that the structure of networks does matter for these outcomes.

While the empirical literature on social networks shows the importance of social networks (in that they seem to have an effect on outcomes), it makes no effort to measure the value of networks, nor does it offer much guidance as to how (or whether) changes in the network itself might produce changes in outcomes. Much of the empirical literature uses extremely crude measures of the impact of an individual’s network on outcomes, often focusing on the number of “links” each individual has to others, to the exclusion of other sorts of data regarding network architecture. At the same time, much of the theoretical literature focuses on the importance of network architecture, and predicts that sometimes even apparently
modest differences in architecture can have important effects on outcomes.

In our research, we are interested in the effect that the architecture of a given network has on sharing, and in particular in the effects of network architecture on the sharing of various sorts of risk. The standard model of risk-sharing in economics (Townsend, 1994) assumes that exchange and sharing among individuals is immediate and costless, and further, that it is not relational in the sense of Levin (2003). A shock to any one individual in a set of connected individuals, say an income shock, should be distributed across the entire network. Sharing in this standard model is not dyadic; rather, each individual shares with the entire community defined by the network. The structure of the network (beyond being connected) simply does not matter.

Of course, the prediction that networks do not matter for sharing depends very much on the absence of frictions within the network. Transaction costs, private information, limited commitment, or the other sorts of impediments to trade that we encounter in the real world void the prediction that network structure will not matter. Some researchers have sought to test whether or not observable network structure has a measurable effect on patterns of sharing. Examples in economics include Dercon and De Weerdt (2008), and in anthropology Ensminger (1992).

From our perspective, studies of this sort have two principal shortcomings. First, they rely on a complete description of the entire network—having a random sample of nodes in a network can’t be relied upon to generate valid inferences regarding sharing within the network, even as the sample grows large. Second, there’s typically not a simple characterization of the effect that a given network structure will have on sharing, and so researchers are often reduced to using quite ad hoc measures of network characteristics and then relating these to outcomes in a manner which isn’t endorsed by any plausible theory.