Integrating Persuasion Theory into the Epidemiological Modelling of Health Misinformation Spread on Social Media

In the digital age, false information about health travels with the speed of a virus. Just as a biological pathogen moves from person to person, a misleading claim moves from screen to screen, reshaping how people understand the world and act within it. For decades, scientists have tried to predict these movements using models borrowed from epidemiology, the study of how diseases spread. These models typically divide a population into simple groups: those who can catch the idea, those who are currently spreading it, and those who have stopped. While useful, these traditional tools often treat everyone the same, assuming that a lie spreads at a steady pace regardless of how it feels or how people react to it. In reality, human behavior is messy and emotional. We are more likely to share a story that makes us angry or afraid, and less likely to pass along something that requires too much mental effort to understand. The question facing researchers today is whether they can build a better map—one that accounts for the psychological currents driving the flow of information, rather than just the structural roads it travels on.

A team of researchers at Manchester Metropolitan University has taken a significant step toward answering this question by creating a new way to simulate how health misinformation spreads on social media. They combined the classic disease-spreading models with a well-known theory of human persuasion called the Elaboration Likelihood Model. This theory suggests that people process information in two different ways: sometimes they react quickly to emotional cues on the surface, and other times they slow down to think deeply about the content. The researchers built a computer simulation that treats the spread of misinformation like an epidemic but adds a layer of psychological realism. Instead of a fixed speed for the spread, their model allows the rate to change every day based on real-world signals: how positive or negative the language is, how many people are liking or sharing the posts, and how complex the words are. They tested this new approach, which they call ELM-SIRMMM, against three different collections of real data from Twitter and health forums, covering topics from COVID-19 rumors to general health advice.

The results of their simulations reveal that adding these psychological details makes a tangible difference, but only when the data itself is rich enough to show change. When the researchers applied their model to a dataset of COVID-19 misinformation from Twitter, the new approach proved more accurate than the older, simpler models. It predicted the peak of the misinformation spread with greater precision, shifting the estimated high point from day 150 to day 160 and increasing the predicted peak prevalence from 6 percent to 7 percent of the population. More importantly, the model's error rate dropped by 5.5 percent, meaning its predictions aligned much more closely with what actually happened in the real world. The simulation showed that the misinformation wave was not a smooth, predictable curve but a dynamic event shaped by the ebb and flow of public emotion and attention.

However, the study also uncovered a crucial limitation: the model's ability to capture these human nuances depends entirely on the quality of the data it receives. When the researchers tested the same model on a dataset of general health discussions from an online forum, the results were different. In this forum, the emotional tone and engagement levels of the posts did not vary much over time; the signals were flat and sparse. Because the input data lacked this necessary fluctuation, the sophisticated psychological part of the model essentially went dormant. The simulation reverted to behaving like a standard, rigid model, failing to capture any dynamic shifts in behavior. In this scenario, the model predicted that misinformation would barely take hold, with only 3 percent of users ever reaching the "infected" state, while the majority remained susceptible. This finding suggests that simply building a complex model is not enough; the model needs a rich, changing environment to work properly. Without varied psychological signals, the advanced features cannot activate, and the simulation loses its edge.

The researchers also examined how different types of misinformation behave. In one dataset focused on emotional rumors, the model successfully reproduced a "flash-rumour" pattern, where misinformation exploded rapidly, infecting 38 percent of users within 45 days, before being corrected and fading away. This matched the real-world observation that highly charged content can surge and then be debunked quickly. In contrast, the model showed that in environments where people are less engaged or the content is less emotional, the misinformation struggles to gain traction, leaving a large portion of the population unaware but not necessarily convinced. The study concludes that while the structure of the model is sound, its power to explain real-world events relies on the variability of the human signals feeding into it. The researchers found that the model works best when it can see the shifts in sentiment and engagement that drive people to share or ignore a story.

Ultimately, this work demonstrates that understanding the spread of misinformation requires more than just counting shares or tracking connections. It requires understanding the human mind behind the click. The study suggests that for models to be truly useful in predicting or managing the spread of false health information, they must be fed with data that reflects the changing moods and efforts of the people involved. If the data is flat, the model remains flat. If the data captures the turbulence of human emotion, the model can reveal the hidden patterns of how lies travel and how they might be stopped. This approach offers a clearer path forward for health communicators and platform designers, showing that the key to managing the flow of information lies in recognizing the psychological forces that drive it.