Models, methods and network topology: Experimental design for the study of interference

Jake Bowers, Bruce A. Desmarais, Mark Frederickson, Nahomi Ichino, Hsuan Wei Lee, Simi Wang

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

How should a network experiment be designed to achieve high statistical power? Experimental treatments on networks may spread. Randomizing assignment of treatment to nodes enhances learning about the counterfactual causal effects of a social network experiment and also requires new methodology (ex. Aronow and Samii, 2017a; Bowers et al., 2013; Toulis and Kao, 2013). In this paper we show that the way in which a treatment propagates across a social network affects the statistical power of an experimental design. As such, prior information regarding treatment propagation should be incorporated into the experimental design. Our findings justify reconsideration of standard practice in circumstances where units are presumed to be independent even in simple experiments: information about treatment effects is not maximized when we assign half the units to treatment and half to control. We also present an example in which statistical power depends on the extent to which the network degree of nodes is correlated with treatment assignment probability. We recommend that researchers think carefully about the underlying treatment propagation model motivating their study in designing an experiment on a network.

Original languageEnglish (US)
Pages (from-to)196-208
Number of pages13
JournalSocial Networks
Volume54
DOIs
StatePublished - Jul 2018

Fingerprint

Social Support
interference
Research Design
experiment
social network
Research Personnel
Learning
methodology
learning

All Science Journal Classification (ASJC) codes

  • Anthropology
  • Sociology and Political Science
  • Social Sciences(all)
  • Psychology(all)

Cite this

Bowers, Jake ; Desmarais, Bruce A. ; Frederickson, Mark ; Ichino, Nahomi ; Lee, Hsuan Wei ; Wang, Simi. / Models, methods and network topology : Experimental design for the study of interference. In: Social Networks. 2018 ; Vol. 54. pp. 196-208.
@article{c009c3ddb71c48008fdff697f95f1167,
title = "Models, methods and network topology: Experimental design for the study of interference",
abstract = "How should a network experiment be designed to achieve high statistical power? Experimental treatments on networks may spread. Randomizing assignment of treatment to nodes enhances learning about the counterfactual causal effects of a social network experiment and also requires new methodology (ex. Aronow and Samii, 2017a; Bowers et al., 2013; Toulis and Kao, 2013). In this paper we show that the way in which a treatment propagates across a social network affects the statistical power of an experimental design. As such, prior information regarding treatment propagation should be incorporated into the experimental design. Our findings justify reconsideration of standard practice in circumstances where units are presumed to be independent even in simple experiments: information about treatment effects is not maximized when we assign half the units to treatment and half to control. We also present an example in which statistical power depends on the extent to which the network degree of nodes is correlated with treatment assignment probability. We recommend that researchers think carefully about the underlying treatment propagation model motivating their study in designing an experiment on a network.",
author = "Jake Bowers and Desmarais, {Bruce A.} and Mark Frederickson and Nahomi Ichino and Lee, {Hsuan Wei} and Simi Wang",
year = "2018",
month = "7",
doi = "10.1016/j.socnet.2018.01.010",
language = "English (US)",
volume = "54",
pages = "196--208",
journal = "Social Networks",
issn = "0378-8733",
publisher = "Elsevier BV",

}

Models, methods and network topology : Experimental design for the study of interference. / Bowers, Jake; Desmarais, Bruce A.; Frederickson, Mark; Ichino, Nahomi; Lee, Hsuan Wei; Wang, Simi.

In: Social Networks, Vol. 54, 07.2018, p. 196-208.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Models, methods and network topology

T2 - Experimental design for the study of interference

AU - Bowers, Jake

AU - Desmarais, Bruce A.

AU - Frederickson, Mark

AU - Ichino, Nahomi

AU - Lee, Hsuan Wei

AU - Wang, Simi

PY - 2018/7

Y1 - 2018/7

N2 - How should a network experiment be designed to achieve high statistical power? Experimental treatments on networks may spread. Randomizing assignment of treatment to nodes enhances learning about the counterfactual causal effects of a social network experiment and also requires new methodology (ex. Aronow and Samii, 2017a; Bowers et al., 2013; Toulis and Kao, 2013). In this paper we show that the way in which a treatment propagates across a social network affects the statistical power of an experimental design. As such, prior information regarding treatment propagation should be incorporated into the experimental design. Our findings justify reconsideration of standard practice in circumstances where units are presumed to be independent even in simple experiments: information about treatment effects is not maximized when we assign half the units to treatment and half to control. We also present an example in which statistical power depends on the extent to which the network degree of nodes is correlated with treatment assignment probability. We recommend that researchers think carefully about the underlying treatment propagation model motivating their study in designing an experiment on a network.

AB - How should a network experiment be designed to achieve high statistical power? Experimental treatments on networks may spread. Randomizing assignment of treatment to nodes enhances learning about the counterfactual causal effects of a social network experiment and also requires new methodology (ex. Aronow and Samii, 2017a; Bowers et al., 2013; Toulis and Kao, 2013). In this paper we show that the way in which a treatment propagates across a social network affects the statistical power of an experimental design. As such, prior information regarding treatment propagation should be incorporated into the experimental design. Our findings justify reconsideration of standard practice in circumstances where units are presumed to be independent even in simple experiments: information about treatment effects is not maximized when we assign half the units to treatment and half to control. We also present an example in which statistical power depends on the extent to which the network degree of nodes is correlated with treatment assignment probability. We recommend that researchers think carefully about the underlying treatment propagation model motivating their study in designing an experiment on a network.

UR - http://www.scopus.com/inward/record.url?scp=85043471711&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85043471711&partnerID=8YFLogxK

U2 - 10.1016/j.socnet.2018.01.010

DO - 10.1016/j.socnet.2018.01.010

M3 - Article

AN - SCOPUS:85043471711

VL - 54

SP - 196

EP - 208

JO - Social Networks

JF - Social Networks

SN - 0378-8733

ER -