Emotions & Misinformation: A Causal Analysis of Twitter News Discussions

75th ICA 2025 in Denver, USA

Jula Luehring, David Garcia, Annie Waldherr, Jana Lasser, Apeksha Shetty & Hannah Metzler

Emotions and misinformation

1. Arousing emotions¹ reinforce partisan-based information processing

Russel, 1980

2. Misinformation² thrives in partisan information environments

Mosleh & Rand, 2022

Misinformation on social media

• Moralizing & arousing content grabs attention

• Misinformation: conflict, negative, moral outrage

    -   only 0.3-6% in 5 studies from 2016-2021,
    -   spread by elite & partisan superspreaders,
    -   creating a self-reinforcing feedback loop with algorithmic systems.

\(\rightarrow\) Misinformation plays into emotional inter-group dynamics

\(\rightarrow\) What are secondary effects?

Nikolov et al., 2021

• On social media, arousing emotions attract attention and higher engagement, reflecting a function of emotions, namely grabbing attention

• Content that is highly arousing and negative gets higher engagement, typically, this is moralizing and conflictual content

• Therefore, misinformation is now understood as a symptom of a clustered information ecosystem where few people are exposed to misinformation, while most people have partisan information diets, including decontextualized but not completely inaccurate information

• So we would also expect misinformation to be embedded in such emotional dynamics, anger, and intergroup hate

• Where a major concern is not just how many people believe in it but secondary effects, such as loss in trust, growing affective polarization, and so on.

What are effects of misinformation on online interactions?

• Therefore, we were wondering what the effects of misinformation are on online discussions, do they trigger conflict and anger, and does this then lead to higher engagement?

There are problems

1. Misinformation is measured as clearly true or false instances,

   -   misrepresenting the broader spectrum of news, 
   
   -   reflecting distinct properties of extreme types. 

2. Different effects of emotions are overlooked by

   -   mixing up timings of emotions,
   
   -   ignoring the function of emotions,
   
   -   measuring positive and negative sentiment only.  

We identified two problems that we wanted to tackle in this study

• First, misinformation is often measured as clearly false, fact-checked instances, which ignores less extreme and more influential types of misinformation, for instance biased news, and it makes it hard to isolate the effects from real news

  • So the question is, are the effects that we are observing unique to misinformation?

• Second, emotions are functional; and the function of emotions hinges on the interaction of prior beliefs and content. Therefore, measuring only positive and negative sentiment, or mixing up emotional state with emotional reactions overlooks the contextual effects of different emotions.

Our objectives

1. Collecting a systematic and large-scale data set for the German-speaking context

Continuous trustworthiness ratings by NewsGuard (#1)

2. Approximating causal inference to test the effects of misinformation in the wild

Non-parametric matching strategy (#2)

So, we derived 2 major objectives:

• First, we wanted to collect a systematic, large-scale and long-term dataset that relies on continuous trustworthiness ratings for sources, including biased but relatively trustworthy sources so that it reflects the whole spectrum of news trustworthiness

• Second, we tried a matching approach to approximate causal inference so that we could isolate the effects of untrustworthy sources

Data collection steps

Big question: How do we measure misinformation?

++ NewsGuard is the most comprehensive list of source ratings

— it’s not reproducible, and we can’t validate it

\(\rightarrow\) today at 12:00!

\(\rightarrow\) ratings and coverage are stable and complete for DE

Emotion detection in tweets

pol_emo_mDeBERTa2

\(\rightarrow\) probability of each tweet containing each emotion (0-1)

• For each tweet, we then classified 8 distinct emotions from text
• This plot shows the covered time frame from October 2020 to March 2022 and we can already see that anger is overall much higher

About populations and covariates

\(\rightarrow\) robust 6% of untrustworthy news

\(\rightarrow\) matching mitigates carryover effects

Non-parametric matching

Matching evaluation

Matching with Nearest Neighbor and Mahalanobis distance

Effects on engagement

Zero-inflated negative binomial models (bootstrapped CIs)

\(\uparrow\) more retweets for tweets with untrustworthy links

\(\downarrow\) but fewer likes & replies

Effects on emotions

OLS regression models (bootstrapped CIs)

\(\uparrow\) more anger, disgust and fear

\(\downarrow\) and less joy in response to untrustworthy sources

• most users replying to untrustworthy news posts also reply to trustworthy news but not vice versa. While 9.6% of users wrote 80% of the replies to trustworthy news, only 2.1% wrote 80% of the replies to untrustworthy news.

Okay, so emotions and misinformation?

• Sources with lower trustworthiness get unique engagement

    - but reach is overall limited

  
  

• Sources with low trustworthiness predict anger & other negative emotions

    - but emotions are functional
  
    - and largely reflect user preferences for such content

Team work <3

Thank you!

Email: jula.luehring@univie.ac.at

Bluesky: @julaluehring.bsky.social

Github: github.com/julaluehring