Bias reduction of peer influence effects with latent coordinates and community membership

TL;DR

This paper investigates methods to reduce bias in estimating peer influence effects by using latent coordinates and community membership, demonstrating their effectiveness in simulations but not in real website data.

Contribution

It introduces a framework for accounting for latent homophily in peer influence studies and evaluates proxies like latent coordinates and community membership.

Findings

Latent homophily proxies improve bias estimates in simulations.

Proxies show no significant bias reduction in real data.

The approach helps clarify peer influence effects in social networks.

Abstract

The importance of peer influence on consumer actions plays a vital role in marketing efforts. However, peer influence effects are often confounded with latent homophily, which are unobserved commonalities that drive friendship. Understanding causality has become one of the pressing issues of current research. We present an approach to explicitly account for various causal influences. We implement a simulation framework to show the effectiveness of two latent homophily proxies, latent coordinates and community membership, in improving peer influence effect estimates on game downloads in a Japanese social network website. We demonstrate that latent homophily proxies have no significant improvement in peer influence effect bias in the available website data.