Beyond Engagement: How Attention-Based Algorithms Influence Political Opinion
Partner institution: New York University (NYU)
Funded by the 2024/25 LSE-NYU Joint Research Seed Fund
Start Date: 31 August 2025
End Date: 1 January 2027
Research team
- Dr Christopher Barrie, Sociology, NYU
- Dr Ben Tappin, Psychological and Behavioural Science, LSE
- Dr Jennifer Allen, NYU Stern School of Business
Abstract
As algorithmically-curated social media like TikTok, YouTube, and Instagram grow in popularity, understanding how these platforms shape our news diets and influence our political attitudes is a pressing concern. However, even though our social media is often referred to as an “attention economy”, relatively little work has explicitly focused on the relationship between social media attention and public opinion formation.
This is particularly relevant as newer platforms like TikTok rely more on implicit attention signals like view time for ranking, compared to Facebook which traditionally relied on explicit engagement signals such as likes and views. This could lead to substantial algorithmic differences if, as has been shown, private patterns of attentional consumption differ from patterns of public interaction.
So what content captures our attention on social media? Further, does optimizing for attention (compared to engagement or chronological-based paradigms) change the way people interact with social media and/or affect how social media influences their opinions?
Prior work by the author team introduced a new tool for measuring attention on social media via a custom platform that mimics a social media feed in a controlled environment. They found that users’ patterns of attention differed greatly from their patterns of engagement (see Barrie et al. 2025). Specifically, they find that we pay attention to a lot of political content we don’t necessarily agree with––even if we don’t engage with it (by liking or sharing).
In this project, we propose leveraging this novel paradigm to experimentally explore the effect of attention-based vs. engagement-based vs. chronological ranking on social media behavior and political attitudes. In particular, we will measure how changes in ranking objectives – e.g. attention-focused TikTok vs. engagement-focused Facebook – can affect users’ engagement with cross-cutting content, perceptions of counter partisans, and partisan animosity.
Barrie, C., Atalay, A., & ElKattan, A. (2025, March 26). Information diets are more diverse in attention than in engagement. https://doi.org/10.31219/osf.io/kvbzd_v2
Keywords: social media, attention, methods, political attitudes
Activities
Funded through the LSE-NYU Joint Research Seed Fund, this project has involved a range of activities including:
- Pulling approximately 200,000 political tweets from an existing NYU CSMaP Lab collection and classifying them as left-leaning, right-leaning or politically neutral
- Manually validating a subset of classifications and independently coding 2,000 tweets for inclusion in the experimental platform
- Using an existing “social media clone” platform to present participants with right-leaning, left-leaning and non-political content
- Measuring attention through “Read More” clicks and explicit engagement through likes and shares
- Recruiting participants through CloudConnect
- Conducting a non-experimental pilot with 100 participants to test categorisation, user interface and engagement/attention patterns
- Conducting an experimental pilot with 200 participants to refine the intervention, validate measures and determine the required sample size
- Randomising participants between attention-driven, engagement-driven and chronological-feed conditions
- Conducting the main experimental study, with sample size determined by the pilot power analysis
- Analysing behavioural and attitudinal outcomes including time spent, attention, likes, shares, affective polarisation, anger, political efficacy and perceptions of partisan division
- Writing up the study findings for submission to a general science journal
Outputs
The activities above have contributed to the following outputs:
- Further research using the experimental paradigm to investigate observability, strategic interaction with social-media algorithms and the downstream consequences of online information consumption
- Policy-relevant evidence concerning misinformation, polarisation and digital-communication regulation
- Exploration of funding and collaboration with private foundations and industry partners including Meta and BlueSky
- Strengthened interdisciplinary collaboration between LSE and NYU in political behaviour, psychology, sociology and computational social science
Outcomes
Recorded long-term outcomes, such as follow-on external funding and evidence of long-term collaboration, will be added here.