Semiparametric methods of policy analysis with social and economic network data

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

Relationships between economic agents are everywhere; firms source inputs from and provide outputs to other firms; nations negotiate and ratify treaties with each other; individuals depend upon networks of friends and acquaintances for fun, emotional support, information and advice. Despite this, rigorous methods for analyzing network data are not widely available. Existing methods are either geared toward descriptive analysis or make restrictive assumptions and are difficult to implement. The proposed research will develop: (i) new methods for policy analysis with two-way interaction outcome data and (ii) models of network formation with many and different agents. Frequently analyzed two-way interaction include trade and migration flows across countries, the value of input flows across firms, and friendships. The proposed methods could lead to efficient analyses and inference about several policy questions such as whether preferential trade agreements increase trade or whether democracy reduces inter-state warfare. The research will therefore aid in formulating efficient policies to govern interactions among individuals, groups, and nations. This will increase trade and exchange and thus improve economic efficiency in the US and around the world. The proposed research will develop: (i) nonparametric methods for policy analysis with dyadic outcome data and (ii) semiparametric models of network formation with heterogenous agents. The research will develop a uniform consistency results for a nonparametric dyadic regression estimator, formulate assumptions supporting causal inference using dyadic data (including estimation methods for proposed parameters of interest and their semiparametric efficiency bound analysis), and introduce methods of (efficient) semi- and non-parametric estimation. These methods are vast improvements over and generalizations of existing methods of network estimation. The research will also develop computation methods and provide ways to implement these new analytical methods. The PIs will make available software implementation of the proposed estimation and inference procedures in the form of a Python 3.6 package free of charge on PyPi and Github. In addition, all replication data, computer codes and supplemental research materials will also be made available online to support additional basic research, and to increase take-up of the proposed methods by policy analysts and empirical researchers. The development of these methods will improve analytical methods for dealing with network data. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.