Rethinking digital trace inference in political communication: evidence from a simulated platform experiment

Digital trace data such as likes, shares, and comments are widely used to infer what captures attention in digital political communication. Yet these visible signals may not map closely onto underlying attention, especially in political contexts where users may monitor content without publicly engaging with it. We examine this attention-engagement discrepancy in a 2 (content domain: political vs. nonpolitical) × 3 (feed curation: engagement-based, chronological, randomized) experiment on a simulated social media platform. We measured attention (dwell time) and engagement (likes, shares, and comments) across 100 posts. Results reveal a political exception: although attention and engagement are only weakly coupled overall, the discrepancy is significantly larger for political than for nonpolitical content. Political posts attract sustained attention while eliciting fewer visible signals. This pattern remains consistent across different feed curation logics, suggesting that the gap reflects a systematic feature of user behavior in feed-based social media rather than a product of any single curation logic. We further show that engagement-based sampling can misrepresent what users attend to by overrepresenting publicly expressive behavior and underrepresenting attentive but less visible forms of political consumption. These findings have implications for research that uses digital trace data to infer attention, issue salience, and public opinion.