Estimating Social Network Models with Link Misclassification

2025-09-08

AI Summary

Binary links in social networks are frequently misclassified due to respondent recall bias or data entry errors, inducing endogeneity in peer variables and yielding severe estimation bias in conventional approaches (e.g., OLS or standard 2SLS). This paper proposes a nonparametric method to estimate misclassification rates without modeling network formation mechanisms. Leveraging these estimates, we correct the peer variable and construct novel instrumental variables, enabling a modified two-stage least squares (2SLS) estimator that consistently identifies peer effects. Unlike existing literature, our approach avoids strong assumptions about misclassification structure or network generative models, enhancing both robustness and practical applicability. In an empirical application to microcredit participation decisions in Indian villages—and across multiple simulation scenarios—the proposed method effectively eliminates bias arising from unaddressed misclassification, reducing average estimation bias by over 70%.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Causality
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Responsible Web: Social and technical mechanisms of refusal for web technologies and applications

Abstract

We propose an adjusted 2SLS estimator for social network models when reported binary network links are misclassified (some zeros reported as ones and vice versa) due, e.g., to survey respondents' recall errors, or lapses in data input. We show misclassification adds new sources of correlation between the regressors and errors, which makes all covariates endogenous and invalidates conventional estimators. We resolve these issues by constructing a novel estimator of misclassification rates and using those estimates to both adjust endogenous peer outcomes and construct new instruments for 2SLS estimation. A distinctive feature of our method is that it does not require structural modeling of link formation. Simulation results confirm our adjusted 2SLS estimator corrects the bias from a naive, unadjusted 2SLS estimator which ignores misclassification and uses conventional instruments. We apply our method to study peer effects in household decisions to participate in a microfinance program in Indian villages.

Problem

• Estimating social network models with misclassified binary links
• Correcting endogeneity from misclassification in network regressors
• Developing adjusted 2SLS estimator without link formation modeling

Innovation

• Adjusted 2SLS estimator for misclassified binary network links
• Novel estimator of misclassification rates to adjust peer outcomes
• Constructs new instruments for 2SLS without modeling link formation