Understanding Social Media: Misinformation, Attention, and Digital Advertising

Online platforms have fundamentally changed the dynamics of social interactions and information transmission. In this thesis, I explore recent trends in social media through models and experiments of user behavior, platform algorithms and incentives, and policy initiatives. I focus on the social consequences of new communication technologies, their intended and unintended societal consequences, and how to steer them in more socially beneficial directions. The analysis is divided roughly into three parts. First, I discuss the role of social media in the propagation of misinformation, how latent platform algorithms may exacerbate its influence, and analyze various policies to correct misinformation spread. Second, to characterize the landscape of digital content, I propose a model of content creation and consumption on digital platforms where users have limited attention, and discuss related experiments on the role of algorithmic ranking in user engagement. Lastly, I study the main business model of social media platforms, digital advertising, to evaluate the welfare implications of social media use. In recent years, social media has become a breeding ground for misinformation, but the reasons misinformation spreads are still imperfectly understood. First, I present a network model of online content sharing where users must decide whether to share an article (of unknown veracity) with others. The model establishes that the impact of homophily on content virality can be non-monotone, but social media platforms interested in maximizing engagement tend to design their algorithms to create more homophilic communication patterns (“filter bubbles”). These incentives to amplify misinformation are particularly pronounced for low-reliability content likely to contain misinformation and when there is greater political polarization and more divisive content. Using this framework, I investigate various regulatory solutions suggested in the public sphere, and show that if not carefully designed, these policies can “backfire” by exacerbating the prevalence of misinformation. I lastly conclude with an experiment intended to identify the key drivers of misinformation sharing in real social media environments. Technological advances stemming from social media have enabled users to systemically access a deluge of information, yet, it is unclear to what extent this technology has actually helped to better inform. Second, I present an attention-based model of social media content creation and consumption on a digital news feed. Contrary to expectation, I show that information overflow and competition for user attention can lead to overall less available information. To show this, I first establish a benchmark level of information provision in a game where a monopolistic content provider invests in creating a single article, and the consumer can strategically choose how much time to invest in reading it. I then provide conditions under which the information provided by a monopolist on a digital platform, who can supply many such articles at the same time, is richer or poorer than this benchmark. Lastly, I find that competition across many content providers further degrades the richness of information in equilibrium, as providers often resort to low-quality content such as catchy click-bait. I conclude with an ongoing experiment studying user attention on an online news feed under different ranking algorithms chosen by a mock social media platform. Business models of online platforms drive much of the content creation and algorithmic choices of platforms, and ultimately impact human-machine interactions. The final part of the thesis discusses the various business models of media platforms and their implications for consumer welfare. Unlike other media platforms such as Netflix or YouTube (who offer subscription-based plans), social media overwhelmingly relies on digital advertising as their main source of revenue (98% of Facebook’s revenue from 2017-2019 was from digital advertising). I find that consumer welfare unambiguously decreases with targeted digital advertising, despite the information consumers gain from watching ads. Using this as motivation, I consider the impact of a digital advertising tax, meant to encourage the platform to switch from an ad-based business model to a subscription-based one. While potentially effective, such a tax can have unintended negative consequences, because it might intensify platform incentives to target ads at susceptible populations who are most influenced. I provide a characterization of a tax policy that necessarily leads to higher consumer welfare. I conclude by generalizing our insights and policy recommendations in the presence of product-level and platform-level competition.

Committee:

Asuman Ozdaglar, EECS (advisor)

Daron Acemoglu, Economics (advisor)

Daniel Huttenlocher, EECS

Adam Berinsky, Political Science