The reflection problem in network effect estimation

• DOI: http://doi.org/10.1111/jems.12301
• Published date: 01 January 2019
• Author: Marc Rysman
• Date: 01 January 2019

Received: 25 August 2018

|

Accepted: 28 August 2018

DOI: 10.1111/jems.12301

The reflection problem in network effect estimation

Marc Rysman

Department of Economics, Boston

University, Boston, Massachusetts

Correspondence

Marc Rysman, Department of Economics,

Boston University, Boston, MA.

Email: mrysman@bu.edu

Abstract

This paper discusses the empirical identification of network effects in light of

the reflection problem of Manski. I argue that models of indirect network effects

present reasonable exclusion restrictions to address the challenges of the

reflection problem.

KEYWORDS

identification, network effects, reflection problem

1

|

INTRODUCTION

The literature on platform economics is among the most interesting developments in industrial organization over the

last 20 years. The literature describes some of the most important and dynamic firms in the world and is central to

understanding significant policy issues, such as net neutrality, financial crises, and standard setting. Research on

platforms is far from complete. Indeed, basic questions such as how to evaluate market power in two‐sided markets are

still up for debate (for different perspectives, see Evans & Noel, 2005; Katz & Sallet, 2018; Weyl, 2010).

A crucial input into the empirical study of platform problems is the estimation of the strength of the positive

spillover from one set of agents to the other. In this paper, I discuss the empirical identification of this spillover. I orient

my discussion around the reflection problem of Manski (1993, 1995). The reflection problem is well known in public and

labor economics, leading to an impossibility of identification in the context of local spillovers between agents. Although

it is always difficult to separate between causal neighborhood spillovers and local correlation driven by unobserved

neighborhood effects, Manksi argues that if we could control for all of these variables, then the model would be

unidentified. This result forces researchers in this area to make strong assumptions on how these effects enter a

problem (if at all) to estimate successfully. The problem should be equally important in the context of identifying

network effects, but I am unaware of a paper that discusses this. For example, Angrist (2014) describes this general

problem in a number of contexts, but does not mention network effects.

I argue that indirect network effects provide a much more natural way to address the reflection problem than direct

network effects. Direct network effects occur when a product is valued directly on how many other consumers use the

product. Communication networks are a standard example, such as fax machines, email, and telephone networks.

Examples of empirical work on these topics are Tucker (2008) and Björkegren (forthcoming). Indirect network effects

exist when consumers care about howmuch of a complementary good is available, and the amount of the complementary

good is determined by the number of consumers using or purchasing the product. For example, consumers care about

how much software is available for an operating system, which is determined by how many consumers adopt the

operating system.Similar examples are consumer andmerchant use of card networks and consumer and advertiseruse of

media, such as Yellow Pages, search engines, and newspapers (for examples, see Rysman, 2004; Fan, 2013; Lee, 2013).

Note that a number of papers study indirect network effect contexts but apply a model of direct network effects, often

because of data limitations. For instance, some early papers on video cassette recorder (VCR) adoption modeled demand

as a function of installed base of consumer adoption rather than on the availability of movie titles. Of course, we would

expect movie title availability to be strongly correlated with installed base.

J Econ Manage Strat. 2019;28:153–158. wileyonlinelibrary.com/journal/jems © 2019 Wiley Periodicals, Inc.

|

153

ORIGINAL ARTICLE