The internet and social media face growing scrutiny over their potential negative effects on human behaviour, politics and society. This scrutiny makes it crucial to empirically determine the direct consequences of internet usage. However, not all studies convincingly overcome the methodological challenge of determining causation. Media coverage of these studies can also be misleading. Fabio Sabatini reviewed the empirical literature, shedding light on the challenge of identifying causality.
The rise of the internet and social media has revolutionised the way we interact, communicate, and access information. Initially praised for their multiple advantages, these technologies now face growing scrutiny over their potential negative effects on human behaviour, political dynamics and societal structures.
The discussion surrounding the internet and social media frequently spills over into public discourse, often sparking vigorous discussions among non-experts, including journalists and politicians. Media coverage of internet-related studies can sometimes be inaccurate or misleading. Ascertaining the direct consequences of internet usage is fraught with complex methodological hurdles, which not all studies in the field convincingly overcome. Reverse causality is one of the major concerns, as it is hard to determine whether internet use is a cause or a consequence of certain phenomena. For example, socially isolated individuals might seek connection through the internet, while others could become isolated due to intensive internet use.
For non-specialists and the public, it can be quite challenging to discern whether a study is asserting a causal link or merely identifying a correlation as a precursor to more in-depth empirical research. In a recent paper, I reviewed the empirical literature on the behavioural, economic and political outcomes of internet and social media use, with a specific focus on the challenge of identifying causal effects. Instead of merely cataloguing the manifold impacts of the internet, my paper aims to navigate the methodologies employed to uncover these impacts, asking critical questions about the establishment of causality in economic studies and the robustness of the evidence regarding the outcomes of internet use.
I identify two main approaches to assessing the causal impact of the internet and social media. The first approach uses natural experiments, where researchers look at how individuals or areas casually gained access to broadband services or specific online content due to factors beyond personal choice or control. For example, natural occurrences like lightning strikes can disrupt mobile Internet access, affecting some but not others, independent of their actions. Studies using this method typically merge data on these external variations in connectivity with behavioral data to examine the impact on Internet usage or interaction with specific content.
The second approach involves setting up experiments in environments where conditions can be tightly controlled, such as in a laboratory. This allows researchers to carefully observe and measure individuals’ direct responses when specific variables or stimuli are manipulated systematically.
My paper focuses primarily on studies using natural and quasi-experiments to gauge the impacts of broadband and social media. The survey begins by evaluating how high-speed internet access impacts users and then shifts focus to the influence of social media. It reviews studies that take advantage of situations where certain people randomly gained access to broadband or specific online content, while others did not, unintentionally creating a “control group” for comparison. Each of these naturally occurring variations guides researchers in determining a tailored approach for their empirical analysis. In my review of the literature, I’ve detailed each approach in its own subsection, examining the methodological challenges, empirical designs, and key findings of significant studies. To aid readers, these subsections include concise tables summarising the essential details of the research discussed.
Studies on the impact of broadband access primarily use natural and quasi-experiments arising from the supply dynamics of internet services. In contrast, research on social media frequently capitalises on what are presumed to be exogenous variations in user demand for these platforms. For instance, certain trending topics or significant events might draw public interest and lead more people to engage with a social media platform, such as becoming more involved with a viral movement or hashtag. Additionally, access to social media platforms may depend on whether the service is available in a particular country or offered in the local language.
The most established strategy in the literature leverages the architecture of the voice networks, laid down between the World Wars, making it exogenous to modern-day supply decisions and consumer preferences. In the initial phase of broadband expansion, the distance to a voice network node significantly influenced access to high-speed internet, offering a source of exogenous variation in internet connection quality among demographically similar areas. My review also examines policy measures that created quasi-experimental settings for accessing high-speed internet. Additionally, I explore studies leveraging phased introduction of mobile networks, the laying of undersea cables to Africa, and instances of unforeseen technical disruptions affecting internet service availability.
Moving to social media, the challenge of identifying causal effects intensifies due to the less tangible nature of network architectures. As social networks emerged when broadband was already widespread, supply-side factors are less influential in explaining their proliferation. However, recent research has leveraged the nuanced topology of online networks to craft innovative identification methods. These methods rely on the assumption that network nodes possess a specific geographic dimension, with centrally located nodes enhancing platform penetration and content reach within certain areas. In this context, the geographical distribution of these central nodes could prove instrumental for identification, provided it’s feasible to identify the exogenous factors influencing the architecture of the network. The enduring stability of network centrality across time lends credibility to using the historical distribution of these nodes as an instrumental variable for analysing social media’s effects.
After discussing the strategies exploiting the initial geographical distribution of key network nodes as instrumental to a platform’s penetration, I explore studies that harness online movements or trends that drove surges in the demand for social media. Next, the focus turns to studies that use the timing of social media content publication to gauge its impact on real-world outcomes. Finally, I explore how disparities in social media access stemming from supply-side factors are employed to study variations in media influence across different regions.
When researching the impact of high-speed internet or social media, it’s essential for scholars to look for unexpected factors that naturally divide people into users and non-users. Imagine if some towns randomly receive high-speed internet while others don’t, just because of their location. By examining these ‘lucky’ and ‘unlucky’ towns, researchers can truly see the effects of these services. It’s like inadvertently forming a test group and a comparison group, which can provide trustworthy insights into the influence of internet access or social media on communities.
However, retrieving such experimental conditions in nature is challenging. When such natural randomisation is not available, field or laboratory experiments emerge as the most viable alternatives.
Similar observations can be made regarding the emerging field of research on the behavioural and economic impacts of artificial intelligence (AI). While a review of these studies falls beyond the scope of my paper, noteworthy similarities are evident. AI, predominantly used online, can be empirically treated as a form of specific content accessible through high-speed internet. The staggered introduction or occasional interruptions of AI-based services present the opportunity to assess the economic impact of AI. Reflecting the methodologies of many studies covered in my paper, research that capitalises on variations in AI access must adopt designs akin to intention-to-treat analysis. This approach means that, when an AI tool like ChatGPT becomes available in one country but not in another, researchers can study the effects by looking at the first country as if all its residents are using the tool. This assumption allows them to gauge the potential overall impact of AI, even without specific data on each person’s usage, giving a broad picture of how such a tool could affect society. An alternative, yet potentially effective, approach might involve field experiments in which subjects, whether they are firms or individuals, are randomly assigned the chance to use specific forms of AI.
Several previous surveys have summarised the effects of internet content on specific domains of individual or public life. My paper stands apart from earlier reviews by prioritising methodological rigor as the principal criterion for evaluating the empirical literature. This effort aims to serve as a guide for researchers embarking on the empirical exploration of the impact of the internet and social media. While the review primarily encompasses economic and political science research, the insights provided are also pertinent for other disciplines such as sociology and psychology, offering a set of basic principles to assess the reliability and external validity of empirical studies. For the wider audience, including policymakers, journalists, and the public, this paper presents a straightforward set of rules to judge the credibility of studies that often influence public discourse and policy. Understanding the causal impact of information and communication technologies is vital for policymakers to formulate strategies that amplify the positive aspects of the internet and social media, while also recognising and curbing their potential negative implications for democracy and human rights.
- This blog post is based on The Behavioral, Economic, and Political Impact of the Internet and Social Media: Empirical Challenges and Approaches, IZA DP No. 16703.
- The post represents the views of its author(s), not the position of LSE Business Review or the London School of Economics and Political Science.
- Featured image provided by Shutterstock.
- When you leave a comment, you’re agreeing to our Comment Policy.