TL;DR
This paper challenges the traditional information theory view that novelty and surprise drive information sharing. Through Twitter analysis and controlled experiments, they demonstrate that information consistent with pre-existing beliefs is shared more, even though it's less surprising. This suggests that semantic meaning, not just novelty, is a crucial factor in what information people choose to disseminate.
What Problem Does This Paper Address?
The prevailing information theory framework suggests that the "value" of information, and thus its likelihood of being shared, is directly proportional to its novelty or surprisingness. This has led to hypotheses that false news might spread faster than true news because it's often more novel and surprising. However, this perspective largely ignores the semantic content of information and how it aligns with an individual's existing beliefs. This paper addresses this gap by investigating whether information's consistency with prior beliefs, a factor that influences perceived surprise, plays a more significant role in information sharing than novelty alone. They aim to disentangle the effects of novelty, surprise, and belief consistency on sharing behavior.
Methods
The researchers employed a multi-pronged approach:
- Twitter Analysis: They analyzed real-world sharing behavior on Twitter.
- Novelty: Operationalized as the time elapsed since the initial tweet about a discrete event. Tweets posted closer to the event were considered more novel.
- Belief Consistency: Assessed by comparing the ideological slant of tweets (pro-liberal or pro-conservative) with the estimated ideology of the users who retweeted them. Users were considered to have ideologically consistent beliefs if they retweeted tweets aligning with their inferred political leanings.
- Sharing: Measured by the number of retweets.
- Controlled Experiments: Two experiments were conducted to establish causal relationships and control for confounding factors.
- Belief Induction: Participants' initial beliefs were manipulated on unfamiliar topics (e.g., risk-taking and firefighting, country joining the EU).
- Information Manipulation: Participants were presented with "update" information that either supported or contradicted their induced beliefs (belief consistency manipulation) and was either previously encountered or novel (novelty manipulation).
- Measures: Participants rated the novelty and surprise of the information and indicated their intention to share it. Experiment 2 also included a measure of perceived information value.
Datasets
- Twitter Data: A dataset of 223 tweets and their associated metadata (retweet counts, timestamps) from non-partisan news organizations about discrete events was collected. For the belief consistency analysis, retweeter IDs and their followed politicians' data were also gathered, resulting in over sixty million data points.
- Experimental Data: Data from two experiments involving undergraduate students (Experiment 1) and Amazon Mechanical Turk workers (Experiment 2) were collected. Specific details on the number of participants and their demographics are provided.
Key Insights and Ideas
- Belief Consistency Trumps Novelty: The most striking finding is that information aligning with pre-existing beliefs is shared more, even when it's less surprising. This directly challenges the idea that surprise is the primary driver of sharing.
- Surprise is Context-Dependent: Surprise isn't solely about newness. Information that contradicts what someone already believes can be surprising, but this type of surprise doesn't necessarily lead to sharing; in fact, belief-consistent information (less surprising) is shared more.
- Novelty's Role is Questionable in Controlled Settings: While Twitter data showed a correlation between novelty and sharing, the controlled experiments failed to find a significant effect of manipulated novelty on sharing or surprise. This suggests that in real-world scenarios, other factors (like algorithms) might amplify novelty's effect, or that people are less sensitive to objective novelty than assumed.
- Semantic Meaning Matters: The research strongly advocates for considering the semantic content of information and its alignment with individual beliefs, moving beyond purely information-theoretic approaches.
Key Results
- Twitter Analysis:
- Tweets posted closer to an event (more novel) received more retweets.
- Users were more likely to retweet content that aligned with their inferred political ideology, suggesting belief consistency predicts sharing.
- Experimental Results:
- Belief-consistent information was rated as significantly less surprising than belief-inconsistent information in both experiments.
- Novelty manipulation did not significantly impact surprise ratings or sharing intentions.
- Belief-consistent information was shared more than belief-inconsistent information in both experiments.
- In Experiment 2, belief-consistent information was also rated as more valuable.
- Perceived novelty ratings did not differ based on the novelty manipulation, indicating participants did not consistently identify novel information as such.
Limitations and Future Work
- Twitter Algorithm: The authors acknowledge that Twitter's recommendation algorithm is a "black box" and could influence observed sharing patterns, making it difficult to isolate individual preferences from algorithmic effects.
- Proxy Measures: Ideological consistency is an imperfect proxy for actual beliefs. Similarly, using time since the first tweet as a proxy for novelty has limitations.
- Controlled Settings: The experimental findings, while causal, were based on unfamiliar topics and hypothetical sharing decisions. Generalizability to contexts with strong pre-existing beliefs (e.g., politics, health) and real-world consequences needs further investigation.
- Future Directions:
- Investigate other factors that might correlate with information freshness and predict sharing.
- Explore how framing information to be consistent with prior beliefs can increase its likelihood of retransmission, especially in public service announcements or political messaging.
- Develop and utilize agent-based models to simulate information diffusion considering individual-level factors like prior beliefs.
- Conduct more real-world field experiments to assess sharing decisions with actual consequences.
- Examine how self-presentational and affiliative motives influence information sharing in more controlled settings.
Summary updated