Online Social Influence
Online Social Influence
Files
Publication or External Link
Date
2014
Authors
Zhang, Yuchi
Advisor
Moe, Wendy W
Godes, David
Godes, David
Citation
DRUM DOI
Abstract
This dissertation studies the behaviors of consumers in an online, social context. In the first essay, we jointly model the drivers of social media rebroadcasting behavior. Our goal in this research is to propose a framework and model of social media rebroadcasting behavior that integrates the various factors shown to influence rebroadcasting behavior. These include the role of message content and influence, factors that have been studied separately with very little integration. The results from our proposed model show that not only does rebroadcasting activity vary with the content of the original message but also that individuals are more likely to rebroadcast content that closely fits with their own interests.
In the second essay, we ask whether online opinions impact consumers' decision quality and assess whether this impact occurs immediately or requires one to undergo learning first. We focus on a setting where consumers have multiple learning episodes based on their experiences with opinions from both uni- and bi-directional ties (i.e. weak and strong ties). We find that the dynamic effects are dependent on the strength of the tie. Additional strong ties (operationalized as bi-directional links) lead to immediate positive effects on decision quality. In contrast, additional weak ties (uni-directional, follower relationships) as a source of information lead initially to lower decision quality. However, highly-experienced consumers receive, ultimately, higher positive effects on decision quality from weak ties as compared with strong ties.
Finally, in the third essay, we propose a new framework and model for identifying dimension specific influentials. We explicitly model individuals' preferences by estimating their locations on a market map to disentangle purchase behavior due to homophily from that due to influence. Our results show that it is important to estimate dimension specific influence, based on a comparison of model fit with a baseline model that measures influence along one dimension. We also show that individuals have varying levels of influence across dimensions, and an influencer for one dimension is not always influential on all dimensions.