The first study focused on stated forwarding intentions. A total of 266 Chinese participants read three pairs of health-related messages, each pair consisting of a piece of misinformation and a matched detailed correction. In one condition, participants rated how likely they would be to forward the correction directly. In another, they were instructed to assume that the misinformation had already been forwarded before they saw the correction, and then rated their willingness to forward the corrective message. Under the conditions used, in which labels disclosing veracity were not identical across message types, direct corrections actually elicited a higher stated forwarding likelihood than misinformation, with a mean difference of 1.24 on the rating scale, a 95 percent confidence interval of 0.52 to 1.96, and an effect size of Hedges’ g equal to 0.63. Telling participants that the misinformation had already spread did not significantly reduce their stated willingness to forward the correction; the mean difference was 0.35, with a confidence interval spanning from minus 0.37 to 1.07. The authors are careful to note, however, that these are hypothetical intentions, vulnerable to social desirability bias and to the well-documented gap between what people say they will do and what they actually do. Because the labeling of messages differed across conditions, the design also cannot fully isolate the effect of correction content itself from the effects of disclosed veracity or corrective framing.

The second study translated these measured propensities into a competitive agent-based model of diffusion. The researchers used the survey ratings as a relative propensity profile rather than as calibrated probabilities for any real platform, and ran simulated contests between misinformation and corrections across three classic network structures: Barabási–Albert scale-free networks, Erdős–Rényi random networks, and Watts–Strogatz small-world networks. They systematically varied four factors: transmission intensity, the onset delay before corrections entered, the supply of correction seeds, and the placement of those seeds in the network. The central quantity was the temporal catch-up cost, abbreviated TCC, defined as the smallest tested ratio of correction seeds to misinformation seeds that restores the cumulative-coverage reduction achieved when both message types start seeding simultaneously in equal numbers, minus one. In plain terms, it measures how many extra corrective seeds are needed to make up for a delayed start.

The simulation results were unambiguous about the price of delay. Across three transmission regimes, all 18 equal-seeding delay penalties were positive, and each penalty grew larger as the delay lengthened and as transmission intensity increased. When corrections were allowed to enter after only one generation of misinformation spread, all nine combinations of regime and network topology managed to recover the benchmark coverage within the tested parameter grid, with temporal catch-up costs ranging from 0.80 to 7.60. After two generations of delay, however, outcomes diverged sharply: some combinations still recovered within the grid, while others failed to do so entirely. Most strikingly, for Barabási–Albert networks under the reference regime, the model implied that no tested seed ratio could restore the benchmark at all, a condition the authors describe as model-implied unattainability.

The one encouraging result involved placement strategy: targeting high-degree nodes in the reference regime especially reduced the catch-up requirements on Barabási–Albert networks, indicating that in the model, where corrections are seeded changed how many were needed.

The authors frame their conclusions with deliberate caution. The thresholds they report are conditional comparisons within their model, not estimates of the resources or clock time that any specific platform would need to contain a real rumor outbreak. The survey findings rest on self-reported intentions rather than observed sharing behavior, and the labeling asymmetry across conditions limits what the first study can prove about correction content alone. What the combined approach does establish is a conceptual point: a correction-sharing disadvantage is not necessary for corrections to lose at the network level. Even when people report being more willing to forward corrections than misinformation, the timing of a correction’s entry can generate catch-up costs that grow steeply with delay, and in some simulated regimes become impossible to overcome regardless of how many corrective seeds are committed.

Subject of Research: The timing disadvantage of misinformation corrections in network diffusion, studied through forwarding-intention surveys and agent-based modeling

Article Title: From sharing propensity to diffusion control: temporal catch-up costs of delayed misinformation correction

Article References: Ding, W., Cheng, Z., & Ge, Y. (2026). From sharing propensity to diffusion control: temporal catch-up costs of delayed misinformation correction. Frontiers in Physics, 14, Article 1953841. https://doi.org/10.3389/fphy.2026.1953841

Image Credits: AI Generated

DOI: 10.3389/fphy.2026.1953841

Keywords: misinformation, fact-checking, information diffusion, agent-based modeling, complex networks, social physics, correction timing, temporal catch-up cost, seed placement, forwarding intentions, sharing, propensity