Urban Legend PropagationIan Dennis Miller
University of Toronto
My name is Ian Dennis Miller from the University of Toronto.
Urban Legend Propagation
In this presentation, I’m going to discuss Urban Legend propagation and the computational modeling of psychological phenomena.
This work begins with a finding from the literature: Urban legends propagate farther when they are disgusting. This is an example of a phenomenon known as emotional selection, in which the emotion evoked by a message can influence how the message is treated. The current work involves the construction of a computational model of urban legend transmission based on the emotional selection literature. Then, we extend the model to generalize beyond the laboratory in an effort to achieve better ecological validity.
This work offers two major contributions. First, I’m going to discuss urban legends and the effect of emotional selection on the propagation of different kinds of stories. I will also discuss a fundamental science of computational modeling for social psychological phenomena - which I needed to articulate in order to perform my research on urban legends.
I make the observation that experimental control is typically inversely proportional to ecological validity. In the experimental sciences, we strive to control as much as possible, but in doing so, we sometimes create experiments that don’t look like the real world anymore. Ecological validity is a common and sometimes valid criticism in the psychological sciences. We want both validity and control but maybe it’s not even possible to have both of these at the same time. I have come to think of these as separate, complementary steps in my own epistemology of science.
This presentation is structured according to the outline of an academic article. There’s a background section, in which I discuss urban legends and computational modeling. For the remainder of the presentation, I describe a series of 3 studies over the course of the methods, results, discussion, and conclusion.
Background: Urban Legends
Regarding Urban Legends.
Background: Urban Legends
By way of background, I’m going to describe the serial reproduction task, which is one way researchers have approached the study of urban legends. I also need to talk briefly about theories of communication and networks. Then, I introduce contagion and epidemiology so that we can characterize the way things propagate. Next, I discuss one of my favorite topics: memes, those wily replicators. Memes are actually what got me into this research in the first place. The scientific study of Social Transmission has lead to new theories about the propagation of words and ideas. The term “Cascade” is used in the literature to characterize social transmission from a media-centric perspective. Cascades are time-lines of media propagation in which many people can participate, often using social media. Then, I’m going to directly discuss urban legends, as a phenomenon unto itself. Finally, I discuss the some ways emotions can influence the propagation of ideas, which is especially important for urban legends.
Serial Reproduction Task
The serial reproduction task first appears in Bartlett’s 1932 book entitled Remembering: A Study in Experimental and Social Psychology. In this work, Bartlett also conducted a study of images, which has renewed relevance to online phenomena in modern times.
Serial reproduction works like a kid’s game called “telephone operator.” In the telephone operator game, kids line up and whisper a message to one another, quietly enough that nobody but the intended listener can hear the message. The message then propagates down the line, one kid at a time. Person A tells person B, Person B tells person C - and so on. What usually happens is that by the end of the game, the starting message has become mutated and distorted beyond recognition.
The task can be conducted using paper or another medium that has greater fidelity than whispering, which eliminates the mistakes as a source of noise. Bartlett discovered that serial reproduction can be a useful research method for studying word of mouth phenomena with rigorous scientific controls.
The canonical theory of communication can be traced back to Claude Shannon in the 1940s. At the time, they were working on telephone networks. Shannon characterized communication in terms of these essential components: a transmitter and a receiver, the channel that connects them, and the message that travels along that channel.
This is also the gateway to information theory. Shannon’s model includes a noise source, which manifests in the children’s telephone operator to produce unexpected results. The signal to noise ratio is proportional to the uncertainty of accurately receiving a message from a transmitter.
A branch of discrete math called Graph Theory deals with networks, including the telephone networks of the 1900s and the social networks of the 2000s. The serial reproduction task is an example of a very simple communication network. Graph theory provides a vocabulary for saying exactly what a serial reproduction experiment is.
In graph theoretical terms: a serial reproduction task consists of a directed, acyclic graph; “directed,” meaning all the graph edges feed forward in a specific direction; and “acyclic,” meaning there are no loops. Every person is represented as a node in the graph. If there are 20 people linked together in a serial reproduction chain, then k = 20. For each person, there is only one other person sending them messages and only one other person receiving messages from them. In graph theoretical terms: we say that nodes have an in-degree of one and an out-degree of one.
For the present purposes, let’s say contagion is the process of transmission by mere contact. This definition for contagion removes the choice dimension entirely. Contagion is what happens; not what someone does.
Nobody chooses to be sick with influenza. To the contrary, most people will make an effort to avoid contracting the flu. Nevertheless, the flu persists because it is contagious - and despite our best efforts, the flu lives on because it is transmissible. This type of pathogenic contagion is studied in the field of epidemiology, which has produced a robust literature on the computational modeling of epidemics. Epidemiological modeling can be applied to more than contagion, including processes that include choices and decisions. The underlying mathematics are generalizable.
Other forms of contagion also exist. Emotional contagion occurs when person A “catches” the emotions of person B. Fear spreads contagiously through social animal groups - not because the animals simultaneously choose to feel fear but through the process of contagion. Emotional contagion could lie at the heart of more sophisticated social processes like sympathy, perspective-taking, and - according to some - altruism.
Social contagion occurs when person A “catches” the beliefs of person B - due, again, to a process other than choice. Asch displayed lines of varying lengths on a screen and induced participants to produce incorrect answers on the basis of group pressure. Roediger, et. al. induced false memories, also through social influence, when confederates systematically fed incorrect answers to participants. These are situations in which ideas are transmitted by a social pressure or force.
Memes are the way I characterize the message or content component in Shannon’s communication theory. The word “meme” comes from Dawkins, in the 7th chapter of his book “The Selfish Gene.” Dawkins regards memes and genes as being fairly analogous - but others have taken a broader or narrower conception of memes.
For my purposes, a meme is a media text or a unit of culture. A meme is anything that can be repeated; it doesn’t need to be a visual text; it could be a tune you can hum. A meme can be mutated; it can be changed. In the case of a poem, a stanza can be mis-remembered, mis-communicated, intentionally altered, or otherwise.
Memes rose to prominence in light of the viral social network phenomena of the early 2000s. The fanatical popularity of memes raises the question of whether memes spread as a contagion. According to a strict interpretation of contagion, I think the answer is no; choice is involved. Nevertheless, memes can be subjected to epidemic analysis.
In the online context, a so-called “meme” refers to a picture with a text overlay that is embedded within the image so that it is shared as a single digital object. These image memes are more accurately called “image macros.” The premiere example of an image macro is the “i can haz cheezburger” cat, which produced a prolific cascade of online image sharing. This particular cat image is not the canonical cat; instead, it’s a knock-off image selected from Wikipedia, demonstrating the principles of propagation and mutation.
The meme is the message in the Shannon theory of communication; it’s what is transmitted. The meme is the content or data that will ultimately be subjected to this computational modeling exercise.
“Word of mouth” communication is called “Social Transmission” in the academic literature. Social transmission can occur online, in-person, overheard on the street, via broadcast media, or other ways. Social transmission is like a relaxed definition of social contagion; now choice is allowed. In this framework, messages propagate because they are interesting, popular, gratifying, or otherwise.
This literature has looked at a variety of phenomena, including how emotions affect whether a story is transmitted or not. Does the message evoke strong emotions or does it put people to sleep? My own work has looked at user-generated content, in which the creators of memes are part of the social transmission process.
During the social transmission process, the timeline of a media object’s lifespan is called a “cascade” in the literature. This cascade may consist of a series of actions taken by multiple people - potentially billions of people. The actions related to a cascade may be simple, like clicking a share button, or they may potentially be complex.
There are countless examples of famous cascades. Perhaps the first online viral video was the 3D dancing baby. We already saw an early “image macro” known as “I can haz cheezburger.” Cascades can be political in nature, such as the Arab Spring. Cascades can be massive, like the Gangnam Style music video cascade with over a billion participants. Cascades could have a strong behavioral component, as with the Ice Bucket Challenge fund raiser, which demanded significant action from participants in the cascade.
But cascades have been around much longer than the Internet.
People have been telling stories for a long time - and those stories are propagated through social transmission. Cascades can exist entirely apart from online media, like in the case of chain letters, mythology, and urban legends. An urban legend is a contemporary story that is passed via word of mouth. Frequently, the legend includes a moral or a fitness-related safety message.
Numerous examples of widespread storytelling crazes can be found in the literature. The Phantom Anesthetist was an epidemic of fear documented in rural America in the 1900s. Another “mass hysteria” was contagious laughter, which was treated as a health risk in some regions of the US.
However, urban legends are typically less dramatic than these crowd phenomena. A better example of modern urban legend is the story of the poisoned Halloween candy, although the literature suggests there is no underlying truth to the story. Another classic urban legend is the story of Big Foot, the supposed ape creature that lives in the Pacific Northwest.
The phenomenon of Urban Legends does not particularly resemble the serial reproduction model. The nature of storytelling is usually not one-to-one. Instead, stories are often told to lots of people, perhaps even in a broadcast context. In the so-called “real world,” urban legends don’t have the harsh experimental constraints that a serial reproduction task imposes.
Emotional selection occurs when a person reacts to something they read. When the emotional reaction causes the reader to behave differently, that’s called emotional selection. Sharing decisions are one of the key behaviors that social transmission investigates and emotional selection can influence this decision. What is the likelihood to re-transmit a word of mouth message, when accounting for the emotion that the message evokes?
Heath et al, described emotional selection operating with disgusting stories in a manner that could be counter-intuitive. While you might predict that disgusting stories would lead to less sharing due to the negative emotions they could evoke, the opposite happens: disgusting stories spread farther.
Recap: Urban Legends
To recap, we begin with the serial reproduction task because it provides an experimentally rigorous paradigm for studying word of mouth processes. Communication theory provides a vocabulary for describing the process of sending a message from the transmitter to the receiver. Communication may be transmitted across networks that can be precisely described with the vocabulary of graph theory. Epidemiology and the study of contagion provide a vocabulary of propagation. Memes are the content that is transmitted through these networks. Social transmission is the study of word of mouth phenomena. A cascade is the media-centric lifetime history of viral content. Urban legends are contemporary stories that can be propagated by word of mouth, both online and offline. Emotional selection is the phenomenon by which media propagation can be affected by the emotion that the media elicits.
Altogether, these concepts provide the vocabularies we need to talk about the rest of this work.
Reviewing a study on Urban Legends
Let’s review some of the work that already exists on Urban Legends.
Reviewing a study on Urban Legends
In this section, I discuss an emotional selection study that looked at disgusting urban legends and measured actual story sharing in the laboratory. These studies set the stage for my own work.
Eriksson and Coultas published a series of studies in 2014 looking at disgust as a moderator for urban legend sharing. What I especially loved about this article is that they separated urban legend transmission into 3 stages that map onto Shannon’s communication theory. There’s a receive stage in which the participant reads an urban legend. Then, there is the encode/retrieve stage which is like the process of remembering a story. A person has to form memories of a story and then later recall those memories in order to actually share it. Finally, there is transmission, in which the story is shared.
Eriksson and Coultas use a serial reproduction paradigm in which one participant tells the next person, in sequence. One thing they find is that all of their cascades - all of the media lifetimes of all the urban legends in their study - they all die out. Cascades end at different rates depending on how disgusting the urban legend is: low-disgust cascades end much sooner than high-disgust cascades. And yet, we know that in the real world, some urban legends cascades survive for years or centuries. The laboratory results do not appear to resemble a “viral cascade.”
As an aside, I was going to conduct a similar study around the same time. I had just finished my master’s thesis and I found that I really wanted to dissect the process of meme transmission. However, thanks to Eriksson and Coultas, I didn’t have to do that work - and the results they produced are just as useful to me. I love this article and they actually saved me a huge headache.
Eriksson & Coultas Study 2: choose-to-transmit (1/2)
There are four studies in the Eriksson and Coultas article but I’m going to focus on just two of them.
In study 2, Eriksson and Coultas investigated whether ratings of a story would affect its transmission properties. Would disgusting stories be more likely to be transmitted than their non-disgusting counterparts? And further to that question, would the disgust level of a story affect ratings of how funny the story was?
First, Eriksson and Coultas generated four story topics and converted each topic into a high and low disgust variation. The four base topics for story: cake, dog, Nepal, and pizza. These topics were then manipulated to create either a high-disgust version or a low-disgust version of the story. In the case of the dog topic, the low-disgust version involved a dog receiving an unexpectedly delicious meal from a restaurant. In the high disgust version, the restaurant cooked the dog and served it for dinner. The high-disgust version of the story violates numerous expectations about cleanliness and morality that are associated with appraisals of disgust. These four base stories were then presented to 80 mechanical turk participants who were assigned according to counterbalanced disgust conditions. Participants rated the stories along several dimensions, including humor, disgust, and the likelihood they would “pass along” the story.
Eriksson & Coultas Study 2: choose-to-transmit (2/2)
Pass-along ratings correspond to the transmit phase of communication and may be interpreted as “intention to transmit.” Eriksson and Coultas found a positive effect of disgust upon transmission intentions, thereby replicating Heath, et al. In fact, several of the measured dimensions, including humor, were related to transmission - although not as strongly as disgust.
For my purposes, this particular study provides a model of Urban Legend transmission. Table 2, which comes directly from the article, contains one mathematical representation of this model. I interpret these maximum likelihood estimates as slopes and I treat this function as a linear model. This model tells me that on average, most stories are not going to be shared because the intercept is nearly a standard deviation below zero. However, disgusting stories may counteract some of the general bias against sharing, such that disgusting stories could end up being transmitted at a much higher rate.
Eriksson & Coultas Study 3: Receive and Transmit (1/2)
Study 3 from Eriksson and Coultas looked at just the receive and transmission stages; the study was designed to eliminate the encode/retrieve component. Participants were given four stories that were written down on paper so that nobody had to memorize or recall anything. This study was implemented using a serial reproduction paradigm with 40 chains consisting of 2 people, each. Eriksson and Coultas viewed these serial reproduction chains like little cascades of size 2. Therefore, each story could potentially live for 2 generations.
Within each generation, there are 2 steps. First, there is a receive step in which participants look at the titles of the stories they have been given and they decide whether to read each story. Then, there is a transmit step in which participants decide whether to share any stories.
A key difference from the previous study is that instead of predicting how much they would share, participants actually do the sharing activity for stories they wish to transmit. This paradigm realistically simulates the filtering effects present in the propagation of urban legends. The flowchart depicts person 1 receiving four stories, then choosing whether to read any of them, then choosing to transmit the stories to person 2. Presuming person 2 has received any stories, that person may now choose whether read the stories they received and subsequently choose to transmit any.
Eriksson & Coultas Study 3: Receive and Transmit (2/2)
This is a cascade-oriented analysis. Eriksson and Coultas tracked the number of stories that were retained after 2 generations. The plot, taken from the article, demonstrates that after each step and each generation, fewer and fewer stories are retained. Notice that high-disgust stories are retained at a higher rate. At the transmit step of the 2nd generation, there are no low-disgust stories left. This briefly demonstrates evidence for emotional selection both during the receive stage and during the transmit stage.