A New SPIN on Misinformation - The Decision Lab (2024)

After reviewing the literature, we gathered an initial list of misinformation types from two of the most comprehensive taxonomies we could find by Kapantai et al and Kozyreva et al. Although both papers originally cut out any terms irrelevant to their respective contexts, we decided to add these terms back in to make our list as exhaustive as possible. Afterward, we turned to ML-GPT to fill in any missing gaps, resulting in 63 terms in total.

We then removed 12 terms based on the following three rules (along with examples of how they were applied).

1. Remove terms encompassed by other, broader terms. We took out “highly partisan news site” since it was already implied by “biased/one-sided.”
2. Remove terms that don’t describe content. We got rid of “bullying” from our list since it doesn’t indicate what’s being shared, just how it’s being shared.
3. Combine similar terms into one, keeping the more popular title. We condensed “cherrypicking” and “bias in fact selection” to be just “cherrypicking” since this is what it’s most commonly called.

This left us with a finalized list of 51 terms, as you can see below.

Now that we had our list, it was time to organize them. From the start, we wanted to create categories that would be most helpful and relevant for informing future interventions. For example, if designing an intervention to encourage participants to verify whether a fact is correct or not, it’s important to know how easily “verifiable” each type of misinformation is first.

We came up with three broader categories—psychological, content, and source—each with its own three subdimensions.

Then, we sorted each of our 51 types of information into these subdimensions. This is what the first five on our list looked like. (Be sure to check out the full taxonomy here!)

To make sense of all of these categories and subdimension, let’s take a closer look at one row in the chart: clickbait, or misleading headlines designed to encourage users to click on a link.

Psychological

This category addresses an individual’s motivation for sharing the information—including their thoughts, emotions, and attitudes. The three psychological dimensions we decided upon were intentionality, profit, and ideological.

1. Intentionality: Is the person purposely lying? Terms could be categorized as either intentional or unintentional.

Since clickbait is deliberately exaggerated to try to get you to click on it, we categorized it as intentional.

1. Profit: Is the person sharing this information to make money? Terms could be categorized as either yes or no.

News websites use clickbait to drive traffic and, in turn, make a profit, leading us to categorize this term as yes.

1. Ideological: Is the person sharing this information motivated by political views or personal values? Terms could be categorized as either yes or no.

Although clickbait might use controversial opinions to catch your attention, the underlying motivation is not always politically driven, so we categorized this term as no.

Content

This category addresses the characteristics of the content being shared. The three content-based dimensions we decided upon were format, manipulation, and facticity.

1. Format: How much information is being provided? Terms could be categorized as either rich or shallow.

Since clickbait only includes the title or thumbnail of an article or video, we categorized this term as shallow.

1. Manipulation: How is the person manipulating the content? Terms could be subcategorized as created, distorted, or reshared.

Clickbait is neither created from scratch nor simply “reshared.” Instead, entities typically twist a story, so we categorized this term as distorted.

3. Facticity: How true do the statements tend to be? Terms could be categorized as false, mostly true, or mixed.

Since clickbait combines true aspects of a story with exaggerated or even fabricated details, we categorized this term as mixed.

Source

This category describes the players at hand, including both those sharing the information and those receiving it. The three source-based dimensions we settled on were audience, verifiability, and agent.

1. Audience: How large does the audience tend to be? Terms could be categorized as either one-to-many or one-to-one.

Articles and videos are shared on the internet with thousands of potential viewers, so we categorized clickbait as one-to-many.

1. Verifiability: Can someone quickly and easily check how truthful the statement is? Terms could be categorized as either yes or no.

After clicking on the article or video, it’s usually pretty clear that it’s loosely based on reality. With this in mind, we categorized clickbait as yes.

1. Agent: Who is the information coming from? Terms could be categorized as either institutions or individuals.

Since clickbait is usually published by an organization, we identified the agent as institutions.

Any time a dimension wasn’t applicable to a particular term, we answered “NA” or “not applicable.”

What went well

We achieved our original objective of creating an exhaustive list of as many types of false or misleading information as possible. This overcame the shortcomings of previous taxonomies by including a wide range of terms spanning several sectors. For instance, “discredited research” might be a term more relevant to academia, while “fake news” might have more political implications. We also accounted for both traditional sources of misinformation, such as “urban legends,” while incorporating novel sources from technological advancements, such as “deepfakes.”

What to work on

Of course, this is not to say that our taxonomy is perfect. The unfortunate reality is that it was impossible for us to include every single type of misinformation due to the sheer volume of words out there. It’s also worth mentioning that each term we included has a wide variety of definitions and examples. This made it difficult to categorize broader terms—like “bogus,” “false information,” or “manipulation”—where various instances may even contradict each other. However, this problem is not specific to our research but is inherent to any misinformation taxonomy, making constant updating and refining even more important.

Where to go from here

As we previously discussed, the ultimate purpose of this taxonomy is to help design future interventions to help participants identify types of misinformation. This could include both pre-existing interventions—such as the bad news game that teaches participants how to spot fake news—as well as designing new interventions to directly combat specific types of misinformation.

Until then, we hope to make this taxonomy as accessible as possible so that anyone can use it as a tool for identifying misinformation. This way, we can all start to make more informed decisions in our lives.

Are you passionate about combating misinformation through behavioral science? The Decision Lab is eager to collaborate with researchers and practitioners to further develop and refine interventions using our SPIN taxonomy. Get in touch today to help us in our mission to create a more informed world!