Modelling the Spread of Fake News

Fake news has become very prominent and influential in today’s world, especially with the many forms it can now spread in, whether that be in-person, through the media, or online. Fake news can be dangerous and detrimental for many different reasons. It can influence behaviors, beliefs, and actions. For example, there has been a recent increase in people ingesting household cleaning liquids after President Trump made suggestion that doing so may be a good way to protect oneself against the Coronavirus (“KDHE: Man Drinks Cleaning Product”, 2020).

Different social scientific theories are relevant to the spread of fake news. One is the diffusion of information theory which states that over time, information gains momentum and spreads through a specific population or social system following curvilinear trajectory (Rogers, 1976). As such, a few people will believe the fake news, which will gradually spread over time. However, there will be people who will not ever believe the fake news. Another social scientific concept is the Big 5 Model of personality (Goldberg, 1992). This model identifies five key individual differences that make up or personality. A factor of particular interest is extraversion. Extraversion is characterized by sociability, talkativeness, assertiveness, and excitability. Therefore, people who are high in extraversion tend to seek out social stimulation and opportunities to engage with others. This would most likely lead them to spread fake news faster and to more people than those who are less extraverted. Lastly, people may be influenced to believe fake news due to the availability heuristic, which states that people tend to use readily available information to make judgements (Tversky & Kahneman, 1973). Therefore, the more times they hear the fake news, the more likely they are to believe it.

We hope that our model is able to accurately simulate the flow of fake news and highlight the general components that impact both the pattern and speed of fake news spread. By highlighting factors that affect both the speed and the pattern of the spread of fake news, it will help us identify aspects that we must address in order to combat the spread of fake news. For example, the more an individual hears about the fake news, the more likely they are to believe it and therefore spread the fake news. Therefore, in order to combat the spread of fake news, it is important to reduce the amount of potential sources spreading the news whether that be individuals, media outlets, or online websites.

Our agents represent people. The number of agents will be set at a specific number and remain constant over the course of the simulation. Agents are coded with different colors depending on if they are unaware of the fake news (white), aware but do not believe the fake news (purple), or aware and believe the fake news (yellow). Furthermore, while the simulation is running, it is graphing those three elements.

The rules the agents follow to transmit the fake news is that one may become aware via broadcast or via those they interact with. People will not spread the news unless they believe the news to be true. The more times the person hears the fake news, the more likely they will believe it. However, just like with any news, real or fake, it will eventually not be the hot topic of discussion, so after 2 weeks the upon believing the fake news, people will stop spreading it.

Method

NetLogo Initial Setup

We used NetLogo to run our simulation (Wilensky, 1997). For our simulation, clicking the Setup button will populate the NetLogo view with turtles in the shape of people in random locations. The number of people is determined by a slider (10 – 200; default is 100). Some of the turtles take the shape “person student”, which represent introverts. The percentage of introverts that are present is controlled by a slider (0 – 100%; default is 40%). Further, a slider is present that determines how many people originally heard and believe the fake news (0 – half of the total number of people; default is five people). Each person has their own network, which is shown by a link to everyone in the network. Each person’s network size is determined by a slider that creates an average number of links per person (0 – 10; default is 5), which allows for every individual’s network to vary slightly.

There are four news-related parameters. Three of these (news relevance, source reliability, and news believability) are sliders that are used to affect the likelihood of someone believing the news. Each of them ranges from 0 - .01, with a default value of .005. The last parameter is broadcast influence. This determines the likelihood of people who are unaware of the news to hear it from the original source rather than another person. It is also represented as a slider with values from 0 – .1 (default is .05). After seven ticks (a tick represents a day) broadcast influence becomes 0.

Simulation Procedure

We will sweep the following parameters: average network size (low/med/high), the number of people who initially heard the news (low/high). Because they each function similarly, the news-related parameters (news relevance, source reliability, and news believability) will be manipulated simultaneously (low/med/high) and will always have the same value for each simulation. For average network size, low = 1, med = 5, high = 10. For people who initially heard and believe the news, low = 5, high = 50. For overall news believability, low = .001, med = .005, high = .01 for all three parameters. The remaining parameters (number of people, introvert percent, and broadcast influence) will be kept constant at their default values.

We are interested in measuring the percentage of people who believe the news and the percentage of people who are unaware of the news. For each simulation, there is a plot for how many people are: aware of the news, unaware of the news, and believe the news over time.

A simulation ran with the default settings:

Results

Our results showing the percent of people who believe and the percent of people who are unaware of the news are summarized in Table 1. From this table, a couple things are apparent. First is that network size and overall news believability together make a large impact on the spread of the news, both in terms of those who are just aware of it, and those who believe it. With low believability, the awareness of the news increases, but there is little change for how many people actually believe the news as network size increases. With low network size, there is little change for how many people are aware of and believe the news as believability increases. Thus, an increased network size will help spread awareness of the news, but for more people to believe the news, the overall news believability needs to be higher. Interestingly, a combination of moderate network size and moderate believability is where the first big change in numbers is. That is, a dramatic increase of those who believe or are aware of the news does not happen unless overall news believability or network size is at least at moderate levels.

The other observation one can make from Table 1 is how little the number of people who initially hear and believe the news is a factor in proportional spread of the news. Similar patterns of those who believe and are unaware are visible across high and low levels of initial fake news believers. The obvious difference is the starting point. The impact of network size and overall news believability is the same. Thus, even when half of the population initially hears and believes the news, the spread is minimal across believability levels if people have small networks. Similarly, if the believability is low, more people become aware of the news but not too many more believe it as network size increases. The most interesting observation may be that even with 10 times as many people initially hearing the news, at high levels of believability and network size, there is no difference in the amount of people who believe the news in the end. While the vast majority of people believe the news in both cases, it is important to note that not everyone believes the news.

Conclusions

Our findings suggest that the spread of fake news is mostly impacted by the average network size of individuals, the news relevance and believability, and the source reliability. Our findings are consistent with existing theories of information spread. There tends to be a fast rate of agents who first begin to believe the news in the beginning that slows down over time, consistent with the diffusion of innovation theory (Rogers, 1976). Furthermore, with at least moderate levels of network size and overall news believability, most (but not all) people tend to believe the news which is also consistent with that model. Although we do not report it in the results, the news spreads further and faster when there are less introverts in the population. This suggests that extraverts are more likely to let others know about the news they’ve heard. Because more people believed the news when network sizes are larger, it appears that our results are consistent with the availability heuristic. Larger networks increases the chances of being connected with people who believe the news. This increases the amount of times someone heard the news from a friend, making it more likely for someone to believe the news themselves (i.e. more available information increased the likelihood of judging the information to be true).

Real-world issues regarding the spread of fake news can be taken advantage of by this model by understanding the impact each one of these components have on the pattern and speed of fake news spread. This will help give insight on what components interventions should target in order to combat the spread of fake news. Since source reliability was a factor that greatly impacted the spread of fake news, we highly recommend individuals to do proper fact checking, and vet the sources where the news originated from.

References

Goldberg, L. R. (1990). An alternative" description of personality": the big-five factor structure. Journal of personality and social psychology, 59(6), 1216.

KDHE: Man drinks cleaning product ‘because of advice he received’. (2020, April 27). KWCH. Retrieved from: https://www.kwch.com/content/news/Kansas- health-officials-see-increase-in-people-misusing-disinfectant-569992881.html

Rogers, E. M. (1976). New product adoption and diffusion. Journal of consumer Research, 2(4), 290-301.

Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive psychology, 5(2), 207-232.

Wilensky, U. (1999). NetLogo. http://ccl.northwestern.edu/netlogo/. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL