Randomized experiments to detect and estimate social influence in networks

TL;DR

This paper reviews experimental designs for measuring social influence in networks, emphasizing how randomized interventions can credibly estimate causal effects and inform social science, marketing, and policy.

Contribution

It systematically explores the design space of network experiments, defining key components and evaluating various strategies for causal inference in social influence studies.

Findings

Randomized experiments can effectively estimate social influence effects.

Design tradeoffs depend on the specific causal questions and network structure.

Randomization provides a foundation for statistical inference in social network experiments.

Abstract

Estimation of social influence in networks can be substantially biased in observational studies due to homophily and network correlation in exposure to exogenous events. Randomized experiments, in which the researcher intervenes in the social system and uses randomization to determine how to do so, provide a methodology for credibly estimating of causal effects of social behaviors. In addition to addressing questions central to the social sciences, these estimates can form the basis for effective marketing and public policy. In this review, we discuss the design space of experiments to measure social influence through combinations of interventions and randomizations. We define an experiment as combination of (1) a target population of individuals connected by an observed interaction network, (2) a set of treatments whereby the researcher will intervene in the social system, (3) a randomization strategy which maps individuals or edges to treatments, and (4) a measurement of an outcome of interest after treatment has been assigned. We review experiments that demonstrate potential experimental designs and we evaluate their advantages and tradeoffs for answering different types of causal questions about social influence. We show how randomization also provides a basis for statistical inference when analyzing these experiments.