At social networking (SN) sites, content is almost entirely user-generated. To attract traffic, a SN firm itself cannot do much beyond periodic updates of site features and design elements. The bulk of digital content—the driving force of the site’s vitality and attractiveness—is produced by its users. However, users are not all created equal. Community members differ widely in terms of the frequency, volume, type, and quality of digital content they generate and consume. From a managerial perspective, understanding who keeps the SN site attractive—specifically, identifying users who influence the site activity of others—is vital. Such an understanding enables more precise ad targeting as well as retention efforts aimed at sustaining and/or increasing the activity of influential existing users (and therefore future ad revenue). Firms operating SN sites observe an “overt” network of friends, defined according to who added whom as a friend. Most of the links in this network are “weak” in the sense that the relationships do not significantly affect behavior in the network. It is of interest to identify the “strong” links (i.e., the links corresponding to friends who affect the user’s behavior). Distinguishing weak links from strong links is a difficult problem for two reasons. First, the number of overt links is large. Second, the firm wants to distinguish the links fairly quickly (e.g., in less than three months), so the number of “observations” available is fairly small. This sets up a challenging statistical problem.
To address this problem, the authors develop a methodology for extracting, with limited data, the strong links from a large overt network that has mostly weak links. They test the model on field data provided by an anonymous SN site. As expected, they find that relatively few so-called friends are actually significant influencers of a given user’s behavior (22% is the sample mean), while substantial heterogeneity across users also exists. The authors also find that descriptors from user profiles (e.g., gender, stated dating objectives) lack the power to determine who, per se, is influential. The spirit of this finding is corroborated by Google’s recent efforts to better quantify social influence so that it might extract more revenue from targeted advertising on MySpace. The application to the field data provides a vivid illustration of the set of results that firms could obtain from applying the model in practice. These also have important implications for SN sites as businesses. In addition to the poor performance of profile descriptors in predicting influence, the authors show that friend counts and profile views also fall short of being able to identify influential site members, especially for the most important 5%–10% of users. Examining a user retention scenario, the authors also illustrate the potential for large gaps in financial returns to the firm from using the model-based estimates of influence versus friend count, profile views, or random selection. They also discuss how the approach can be readily applied to other data that might also be available to firms operating SN Web sites.
Michael Trusov is Assistant Professor of Marketing in the Robert H. Smith School of Business at the University of Maryland. He received his PhD from the Anderson School of Management at the University of California, Los Angeles. He also holds a master’s degree in Computer Science and a master’s degree in Business Administration. He is a winner of the Marketing Science Institute’s Alden Clayton Doctoral Dissertation Competition. His research interests include Internet and e-commerce (social networks on the Internet, clickstream analysis, electronic word-of-mouth marketing, electronic customer relationship management, online recommendation systems, paid search, consumer-generated media), discrete choice modeling, eye tracking, and data mining.
Anand V. Bodapati is an associate professor in the Anderson School of Management at the University of California, Los Angeles. His research interests are in the areas of customer relationship management, direct marketing, translation of consumer psychology findings to econometric models, and methodological issues related to assessing customer responsiveness to marketing. He received the 2009 Paul E. Green from the American Marketing Association.
Randolph E. Bucklin is Peter W. Mullin Professor of Marketing in the Anderson School of Management at the University of California, Los Angeles. His research interests are in the development of models of choice behavior using historical records of customer transactions. He has published extensively on customer behavior in variety of settings, including consumer packaged goods, automotive markets, and the Internet. He is the coeditor of Marketing Letters (2006–2010) and serves on the editorial boards of Journal of Marketing Research, Marketing Science, and International Journal of Research in Marketing. Professor Bucklin received his PhD (Business) and MS (Statistics) from Stanford University and an AB (Economics) from Harvard University.
Journal of Marketing Research, Volume 47, Number 4, August 2010
View Table of Contents