Econometric models for networks and matching with heterogeneous agents

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

This award funds research that will develop new statistical methods for analyzing data about human behavior in network contexts. A group of people can be described as a social network by explicit description of the links between individuals. For example, your social network includes your family, your neighbors, co-workers, etc, as well as more indirect links such as your link through your children to their friends and teachers, etc. People of course make decisions about their social networks; you may make new friends or spend less time with a neighbor. Recent work in economics has focused on modeling those decisions about network relationships as strategic; each individual in the network forms, maintains, or severs relationships to meet his/her own goals. The PI will develop new methods for testing hypotheses about the effects of networks using statistical methods. He also plans to develop new methods that can be used to analyze data from joint production situations, where multiple people work together to achieve a goal. The results will advance the growing field of network science and will promote the national prosperity because businesses are increasing viewing customer relationships through the lens of network effects. The PI will formulate an empirical model of dynamic network formation that will (i) have a random utility maximization foundation (ii) leave the distribution of unobserved agent heterogeneity flexible, and (iii) leave the process generating the initial network structure unspecified. The project on joint production goes beyond previous work on the econometric methods for the analysis of matchings in situations where the observed matching satisfies strong, random-assignment type assumptions. The new research funded here will develop methods for recovering match output under settings where random assignment does not hold. The project will contribute to our understanding of efficient estimation of semi-parametric models. Software and data produced will be made freely available online.