Spillover Effects in Microeconometrics with Experimental and Observational Data Restricted; Files Only

Abstract

Spillover effects are pervasive in economic and social environments, where individuals behave and interact strategically. Estimating the presence and magnitude of these effects is challenging due to the inherent endogeneity of social network formation. This dissertation develops new microeconometric methods to identify and estimate spillover effects using both observational data and randomized experiments.

When only observational data are available, the dissertation proposes a model-based strategy to estimate peer influence across multiple layers of social connections, assuming a continuous outcome variable. The identification strategy exploits the characteristics of distant peers, who do not directly influence an individual, as instrumental variables to disentangle peer effects from homophily. When the outcome of interest is binary, the dissertation adopts a partial identification approach using moment inequalities. This method directly estimates revealed social preferences while accommodating the possibility of multiple equilibria in agents' decisions. Both methodologies are applied to the National Longitudinal Study of Adolescent Health (Add Health), uncovering positive peer effects on cigarette smoking and marijuana use across diverse networks, including friendships, classmates, and clubmates.

In experimental settings with outcome data available only for a nonrandom subset of individuals, the dissertation introduces a new method to estimate spillover effects under a novel monotonicity assumption. This approach generalizes conventional trimming bounds for attrition to settings with network interference. It allows for covariate adjustments and the incorporation of high-dimensional controls using modern machine learning tools. An empirical application to a randomized intervention on computer usage in public schools reveals meaningful spillovers, even when outcome data are partially observed.

Together, these studies contribute to the econometric analysis of social interactions by addressing key empirical challenges: endogeneity, sample selection, and multiplicity of equilibria. They demonstrate that careful modeling of network structure and behavioral responses can yield credible estimates of spillover effects. The findings highlight the central role of peers in shaping individual behavior and offer practical tools for researchers and policymakers seeking to understand and leverage social influence.

Table of Contents

1. Estimating Peer Influence in Multilayer Networks

1.1 Introduction

1.2 Social Marginal Effects

1.2.1 Social Interactions Model

1.2.2 Heterogeneous Katz-Bonacich Centrality

1.3 Data and Descriptive Statistics

1.3.1 Add Health Data

1.3.2 Descriptive Statistics

1.4 Empirical Framework

1.4.1 Identification Strategy

1.4.2 Estimation Strategy

1.4.3 Social Effects

1.5 Results and Discussion

1.5.1 Friends and Classmates Effects

1.5.2 Identifying Influencers

1.6 Conclusion

2. Spillover Effects with Nonrandom Sample Selection

2.1 Introduction

2.2 Spillover Effects

2.2.1 Setup

2.2.2 A Sample Selection Model with Network Interference

2.3 Partial Identification

2.3.1 Spillover Bounds

2.3.2 Bounds with Covariates

2.4 Estimation and Inference

2.4.1 Estimation

2.4.2 Multiple Treatment

2.4.3 Covariate Adjustment

2.4.4 Inference

2.5 Monte Carlo Simulations

2.6 Field Experiment on Computer Use

2.6.1 Spillover Bounds without Covariates

2.6.2 Including Covariates on the Selection Process

2.7 Conclusion

3. Estimating Peer Effects in the Presence of Multiple Equilibria: A Moment Inequalities Approach

3.1 Introduction

3.2 Multiple Equilibria in Social Interactions

3.3 Partial Identification and Moment Inequalities

3.4 A Model of Peer Effects

3.5 Estimating Moment Inequalities

3.6 Add Health Data

3.7 Results

3.8 Conclusion

Appendix A. Estimating Peer Influence in Multilayer Networks

A.1 Appendix Tables and Figures

A.2 Social Interactions Model

A.3 Identification and Estimation Details

A.3.1 Identification

A.3.2 Average Social Effects

Appendix B. Spillover Effects with Nonrandom Sample Selection

B.1 Appendix Tables and Figures

B.2 Proofs and Estimation Details

Appendix C. Estimating Peer Effects in the Presence of Multiple Equilibria: A Moment Inequalities Approach

C.1 Appendix Tables and Figures

Bibliography

About this Dissertation

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Laney Graduate School
Department	Economics
Degree	Ph.D.
Submission	Dissertation
Language	English
Research Field	Statistics Economics, General Economics, Theory
关键词	Spillover effects Sample selection Partial identification Social networks Machine learning
Committee Chair / Thesis Advisor	David Jacho-Chavez, Emory University
Committee Members	Ian McCarthy, Emory University Weihua An, Emory University Pedro Sant'Anna, Emory University Alejandro Sanchez-Becerra, Emory University

最新修改

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	File download under embargo until 22 May 2027	2025-04-10 10:41:43 -0400	File download under embargo until 22 May 2027

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions