The Half-Life of Scientific Claims as a Measure of the Pace of Science

Andrew White proposed measuring the pace of science by the lifespan of scientific claims

On 4 August, Andrew White published essay D1-012 and an open dataset on the Diffuse website. He proposes measuring the pace of science by the time between the inclusion of a notable claim in a review and its substantial revision or refutation in a later paper.

The number of new papers measures the rate of publication. White instead proposes tracking when researchers revise a specific conclusion about a disease mechanism, effect size, or material property. A later paper substantially qualifies or refutes the claim, thereby changing its status.

White uses the inclusion of a claim in a review as the starting point, treating this as a practical threshold of significance. The PaperQA3 scientific literature search system then looks for later papers. If a paper substantially changes or refutes the claim, White requires its DOI, the publication's permanent identifier. This makes each change in status traceable through a sequence of events: the claim appeared in a review, a later paper changed its status, and a measurable period elapsed between the two events.

The dataset contains 3 444 claims from 151 reviews. Under White's criteria, 2 915 retained their status, 419 were substantially revised, and 110 were refuted. To estimate the lifespan of claims of different ages, he used a Kaplan-Meier curve, which shows the proportion of items that remain unchanged over time. The model produces a median estimate of 115 years, the point at which the curve reaches one half. When the claims are grouped by decade, the model gives 250 years for claims from the 1950s and 70 years for claims from the 2000s.

A 2023 paper by Michael Park and colleagues measured a different quantity: the extent to which papers and patents alter the network of subsequent citations. White focuses on a different object of measurement, namely the time it takes for later literature to change the status of a specific claim. Publication counts measure the volume of the research output, while his model describes what happens to individual claims within that output.