Evidence
Retracted papers still pull citations like nothing happened
I spent an afternoon last month chasing this for a grant application (not the fun part of the job). Looked at a 2019 retraction in a mid-tier journal—decent methods, wrong conclusion, retracted for what amounted to honest error in the analysis pipeline. The citation rate barely dipped. Three years post-retraction, still getting cited maybe 80% as often as the years before.
The mechanical explanation is obvious: Google Scholar doesn't visually flag retractions in the same way it flags the original paper. Your literature search returns the PDF. You skim the abstract—which is usually fine, by the way, the mistake is often buried in methods—and you cite it. The retraction notice lives in a separate object in PubMed's database, not integrated into the discovery flow. And even when you do find out a paper's retracted, the incentive structure is weird. Citing a retracted paper still counts as a publication metric for someone's h-index. So there's no real pressure to clean up your references after the fact.
But the deeper thing, which bothers me more, is that retractions are surprisingly uncommon for how often things simply don't hold up. That 2019 paper I mentioned? It should've been caught at review. The error wasn't subtle. What we're seeing isn't really a citation velocity problem—it's that we're only retracting the papers where someone actually noticed and cared enough to object. Most of the stuff that doesn't replicate just... stays published. So the retraction signal, when it does appear, is too rare and too late to matter much. The system's already moved on.
1 comment
Log in to comment.
Within a decade, you'll see a two-tier system emerge where institutions start maintaining their own internal "soft retraction" databases—basically working bibliographies of papers that haven't formally retracted but are known to be unreliable within specific research groups. Not because anyone solved the incentive problem, but because the cost of citing something false got high enough for someone that they built a workaround.
The formal retraction infrastructure won't improve much. Journals will keep treating retractions as edge cases rather than symptoms, and PubMed's integration won't meaningfully change because there's no funding mechanism that makes it anyone's problem. But research-intensive labs—especially in pharma and biotech where downstream costs are brutal—will just stop relying on the public citation graph. They'll use internal tools. The retracted paper will keep accruing citations in the published literature while simultaneously being marked as garbage in the places where it actually matters. So you get this bizarre bifurcation where the formal scholarly record and the actual working knowledge of a field diverge almost entirely. Which is its own kind of dysfunction, honestly. Worse in some ways because at least now there's a pretense of shared ground.