Collective Memory Decay in Science: Patterns and Determinants of Forgetting Retracted Research

One of the most distinctive features of science is its capacity for self-correction. That correction can take many forms — from gradual revisions of results and paradigm shifts, to formal mechanisms that remove works from the scientific record when they are found to be erroneous, irreproducible, or fraudulent. Retraction is one such mechanism.

But formal correction does not guarantee forgetting. We have known for years that the paper linking autism with vaccines was retracted. Yet in public discourse it continues to be cited to support certain narratives. In academia, it keeps circulating — in 2025 alone it accumulated more than a hundred citations. Many of them do not validate it; they analyze it as a case of scientific misinformation. This introduces a key difficulty: not every post-retraction citation reflects endorsement, but it does indicate that the retracted knowledge remains active. Distinguishing between these forms of persistence is part of the challenge.

This tension between formal correction and collective persistence is at the center of this Fondecyt Regular. The question is simple to formulate and hard to answer: how do scientific systems forget, and why do certain contents remain active even after being invalidated? The consequences are not minor — that persistence can continue to shape research agendas, conceptual frameworks, editorial decisions, and public debates, weakening science’s effective capacity to self-correct.

Answering this question demands a genuine cross-disciplinary approach, because the problem is simultaneously social, dynamic, and causal. The notion of collective memory, developed in sociology and cultural studies, provides the framework for posing the problem. But observing memory means observing trajectories over time, where individual decisions accumulate into collective patterns. To describe that passage from the micro to the aggregate level, the project draws on first-principles dynamic models expressed through master equations from statistical physics, which formalize how the distribution of attention and forgetting evolves within the scientific system.

Capturing the dynamics alone is not enough. The fact that a paper continues to be cited after retraction does not necessarily imply validation. Some citations express endorsement, others criticism, others simple institutional inertia. For this reason, the project incorporates causal inference tools — developed in statistics and economics — to distinguish between structural persistence of the system and effects attributable to retraction itself.