🔬 physics

Modeling the amplification of epidemic spread by individuals exposed to misinformation on social media

This study employs a mobility-informed epidemic model integrated with social media data to quantify how exposure to online vaccine misinformation amplifies disease spread, revealing significant increases in infections under worst-case scenarios to inform public health policy.

Original authors: Matthew R. DeVerna, Francesco Pierri, Yong-Yeol Ahn, Santo Fortunato, Alessandro Flammini, Filippo Menczer

Published 2026-02-09

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Matthew R. DeVerna, Francesco Pierri, Yong-Yeol Ahn, Santo Fortunato, Alessandro Flammini, Filippo Menczer

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the spread of a virus like a wildfire moving through a forest. Usually, scientists try to predict how fast that fire will burn by looking at how close the trees are to each other and how dry the leaves are.

This paper adds a new, invisible layer to that forest: a cloud of smoke that changes how the trees behave.

Here is the story of what the researchers found, broken down into simple concepts:

1. The Two Forests

The researchers realized that people live in two different "worlds" at the same time:

The Physical Forest: This is where people actually meet—walking to work, sitting in a classroom, or grabbing coffee. This is where the virus spreads.
The Digital Forest: This is where people share stories and news on social media (specifically Twitter in this study). This is where misinformation spreads.

The paper argues that these two forests are connected. If someone in the Digital Forest believes a lie (like "vaccines are dangerous" or "masks don't work"), they might act differently in the Physical Forest. They might stop wearing masks or stop social distancing, making it easier for the virus to jump from person to person.

2. The "Resilience" Test

To see how bad this could get, the researchers built a giant computer simulation with about 20 million virtual people. They created two extreme scenarios to test the limits:

The "Super-Resilient" Forest (Best Case): Imagine a forest where the trees are made of steel. Even if a puff of misinformation smoke blows by, the trees ignore it. These people follow all the health rules (masks, vaccines, distancing).
The "Super-Sensitive" Forest (Worst Case): Imagine a forest made of dry kindling. If a single puff of misinformation smoke touches a tree, that tree immediately catches fire (becomes misinformed) and stops following safety rules.

3. The Experiment

The researchers used real data from Twitter to see how many people were sharing "low-credibility" content about vaccines. They then mapped these people onto their giant 20-million-person simulation.

They asked: "If we have a lot of people who believe the lies, how much worse does the virus get?"

4. The Shocking Results

When they ran the simulation with the "Worst Case" scenario (where people believe misinformation after just one exposure), the results were dramatic compared to the "Best Case":

The Fire Spreads Faster: The peak of the infection happened two weeks earlier.
The Fire Gets Bigger: The number of people getting sick at the peak was six times higher.
The Total Damage: Over the course of the epidemic, an additional 14% of the entire U.S. population would get infected.
- To put that in perspective: That's nearly 47 million extra Americans getting sick.
The Cost: The researchers estimated this would cost the U.S. healthcare system over $143 billion.

5. Why This Happens

The model assumes that people who believe misinformation are less likely to take safety precautions. In the simulation, these "misinformed" people act like they are in a normal, non-pandemic world (hugging, not masking, gathering), while the "well-informed" people are acting carefully.

Because the virus doesn't care who is careful and who isn't, the careless behavior of the misinformed group creates a bridge that allows the virus to infect the careful group, too. It's like one group of people leaving a door open for a draft; eventually, the cold air (the virus) gets into the whole house, even the rooms where people tried to keep the windows shut.

Summary

The paper doesn't say that misinformation is the virus. Instead, it shows that misinformation acts like a force multiplier. It doesn't just hurt the people who believe the lies; it makes the whole epidemic worse for everyone, accelerating the spread and infecting millions more people than necessary.

The researchers used real data to show that if we don't address the spread of online lies, we are essentially handing the virus a shortcut to infect the entire population.

Technical Summary: Modeling the Amplification of Epidemic Spread by Individuals Exposed to Misinformation

Problem Statement
Public health outcomes are significantly influenced by social factors, particularly the willingness of populations to adopt mitigation measures. Recent evidence suggests that exposure to misinformation on social media correlates with increased vaccine hesitancy and reduced compliance with public health guidance. However, quantifying the causal impact of misinformation on epidemic dynamics remains challenging due to a lack of holistic, data-informed models that integrate both the spread of information and the spread of disease. Existing studies often rely on simplified experimental settings or lack real-world integration of social media data with physical contact networks. This paper addresses the gap by proposing a framework to estimate the additional harm caused by misinformation during an epidemic, specifically focusing on the COVID-19 pandemic in the United States.

Methodology
The authors propose a Susceptible Misinformed Infected Recovered (SMIR) model, a multi-level agent-based simulation that couples an information diffusion network with a physical contact network.

Information Network Construction:
- Data Source: The model utilizes a dataset of approximately 26 million English-language tweets regarding COVID-19 vaccines collected between January 4 and September 30, 2021, from over 2 million U.S. Twitter users.
- Geolocation and Filtering: Users were geolocated to 341 U.S. counties (each containing >200 users). Political alignment was inferred using annotated news sources and label-propagation algorithms.
- Misinformation Definition: Accounts sharing links to low-credibility sources (NewsGuard score < 60) were identified as spreading misinformation.
- Diffusion Model: A single-step linear threshold opinion-spreading process models how misinformation converts ordinary users to "misinformed" (M) status. A threshold parameter, $\phi$ $ϕ$ , represents the number of misinformed friends required to convert an ordinary node.
  - Low $\phi$ (e.g., $\phi=1$ ) represents a "worst-case" scenario where a single exposure leads to misinformation (simple contagion).
  - High $\phi$ (e.g., $\phi=20$ ) represents a "best-case" scenario where the population is highly resilient to misinformation.
Contact Network Construction:
- Data Source: A physical contact network of approximately 20 million nodes (representing a 10% sample of the U.S. population) was constructed using SafeGraph cell phone mobility data and 2020 U.S. presidential election voting records.
- Structure: The network captures mobility patterns between and within counties. Node attributes (misinformed vs. ordinary) are assigned based on the distribution derived from the information network, ensuring the ideological split (Republican vs. Democrat) matches empirical voting records.
Epidemic Simulation (SMIR Model):
- The model extends the standard SIR framework. Ordinary individuals (O) are assumed to follow public health guidelines (masking, distancing, vaccination), while misinformed individuals (M) are assumed to refuse these measures.
- Parameters: Transmission probability ( $p$ $p$ ) and contact density ( $\bar{k}$ $\overset{ˉ}{k}$ ) are manipulated to bound scenarios.
  - Ordinary ( $O$ ): $p_O = 0.01$ (high mitigation).
  - Misinformed ( $M$ ): $p_M = 1.0$ (no mitigation, high transmission).
  - Contact density is fixed at $\bar{k} = 25$ (pre-pandemic average).
- Simulations run for 100 days, with recovery rate $\gamma = 0.2$ (5-day quarantine).

Key Contributions

Integration of Real-World Data: The study moves beyond theoretical assumptions by integrating large-scale, empirically derived social media data (Twitter retweets) with physical mobility data (cell phone traces) and demographic data (voting records).
The SMIR Framework: It introduces a specific agent-based model that explicitly tracks a "misinformed" subpopulation and their differential behavior regarding disease transmission.
Quantitative Bounds: By varying the resilience parameter $\phi$ and transmission probabilities, the model provides quantitative bounds on the potential harm of misinformation, rather than a single point estimate.

Results
The simulations compare a worst-case scenario (low resilience, $\phi=1$ ) against a best-case scenario (high resilience, $\phi=20$ ):

Peak Infection: In the worst-case scenario, the peak number of daily infections is approximately six times higher than in the best-case scenario.
Timing: The peak infection occurs approximately two weeks earlier in the worst-case scenario.
Cumulative Impact: Over the course of the simulated epidemic, the worst-case scenario results in an additional 14% of the total population becoming infected compared to the resilient scenario. This represents a 32% relative increase in total infections.
Scale: Based on U.S. Census data, this additional 14% translates to nearly 47 million additional Americans infected.
Economic Impact: The authors estimate the associated healthcare costs for these additional infections to exceed $143 billion.

Significance and Claims
The paper claims that its primary significance lies in providing policymakers and public health authorities with data-informed insights into the potential harms of online vaccine misinformation. By establishing an "alarming bound" on the consequences of misinformation exposure, the work aims to motivate social media platforms and health officials to curb the spread of low-credibility content.

The authors explicitly state that their results represent extreme scenarios (using $p_M=1$ and $p_O=0.01$ ) and acknowledge limitations, including the assumption of a causal link between misinformation and risky behavior, the use of Twitter as a proxy for all information exposure, and the simplification of individual resilience as uniform. However, they argue that even within these constraints, the model demonstrates that misinformation can significantly amplify epidemic spread and delay suppression efforts. The work underscores the necessity of managing the complex relationship between online information networks and physical contact networks to effectively control epidemic crises.