Modeling, Estimation and Interpretation of Peer Effects in Social Networks

Shin, Jaekyeong

Graduate Thesis Or Dissertation

Modeling, Estimation and Interpretation of Peer Effects in Social Networks Public Deposited

Analytics

Download PDF

Citations

Citeable URL: https://scholar.colorado.edu/concern/graduate_thesis_or_dissertations/hm50tt57h

Abstract

In this dissertation, I propose a model that estimates the underlying motives of peer effect. With Monte Carlo simulations and actual data, I apply the model and interpret the results by relating to the motives. Additionally, I suggest an algorithm to capture network links with panel data and binary outcome variables for the cases where network data are unavailable.

In the first chapter, a generalized spatial autoregressive model (GSAR) estimation model is presented. In addition to the intensity of peer effect, which is the sole focus of the standard SAR model, I introduce a ``conformity parameter'' to capture the underlying motives of peer effect: complementarity and conformity. These motives lead to different policy implications as only the former generates the social multiplier effect. I perform Monte Carlo simulations and demonstrate that the GSAR model is robust to model selection among different motives of peer effect. Also, I characterize the threshold for a positive social multiplier effect.

In the second chapter, I examine the microfinance data collected by Bharatha Swamukti Samsthe (BSS) in Karnataka, India. I extend the GSAR model for binary outcome variables and higher-order networks and apply it to the data. As a result, I find strong evidence of the villagers' peer effect with a significant conformity motive. Using the estimated parameters, I conduct a counterfactual analysis with alternative target groups. It is shown that the target group chosen based on the GSAR model is more effective than the original group designated by BSS.

The final chapter studies a case where network data is unavailable. I estimate a network structure using panel data of repeatedly observed samples with binary outcome variables. I present an algorithm that combines the binary particle swarm optimization process with the shrinkage estimator. Simulation results for different types of networks are presented, along with a comparison of the estimation performance with varying numbers of particles.

Creator