Expressions of emotions and social behavior around COVID19 on Twitter the USA

Methods

• We analysed 4.5 billion tweets from users located in the USA by Crimson Hexagon. Only original tweets (no retweets) in English and from users with a number of followers between 100 and 100.000 (to remove spam and media) were included.
• For anxiety, tweet text is matched for anxiety terms from the LIWC lexicon. LIWC is a standard methodology in psychology for text analysis that includes validated lexica in English.
• Similarly, tweet text was matched to a list of prosocial terms used in previous research, including for example words related to helping, empathy, cooperating, sharing, volunteering, and donating. We excluded words referring to (1) health, (2) service (mostly comments about the public services and availability of other services during the Corona-crisis), and (3) sharing words. We did this to avoid confounds with non-prosocial tweets about Covid-19 related health- and other services, and about sharing of links/tweets/videos, etc.
• Marked dates during the Covid-19 outbreak are taken from Wikipedia.

Anxiety

In both figures below, the lines indicate the following events:

• Red:
  • Terrorist attack in Christchurch, New Zealand on 2019-03-05
  • Mass shootings in Ohio and El Paso on 2019-08-04
• Yellow
  • Thanksgiving on 2019-11-28
  • Christmas on 2019-12-24, shortly followed by another low for New Years eve
• Violet: USA kills Iraqi military commander Soleimani on 2020-01-03
• Green: First Covid-19 case on 2020-01-20
• Light red: Super Bowl annual chamiponship of the National Football League (NFL) on 2020-02-03. Large sports events are usually linked to an increase of various emotions, not just anxiety, so it is likely that we will see a similar increase when analyzing other emotions.
• Orange: First death from Covid-19 on 2020-02-29
• Turquois: Measures
  • Light: On 2020-03-05, American Congress votes an 8.3 million emergency plan to fight against the coronavirus. During the next days, the Governor of New York declares state of emergency on 2020-03-07, Wall Street collapses on a black Monday on 2020-03-09, and President Trump closes the borders to European citizens on 2020-03-11.
  • Dark: Trump urges people to practice social distancing on 2020-03-16. The first social distancing measures (transition to online courses) were implemented by schools and universities in New York, Massachusetts and Washington on 2020-03-10, other institutions and states follow every day from now on.

Percent of tweets containing anxiety-related terms

Percent change in anxiety-related tweets relative to the average level in 2019

The figure below shows the difference to the average amount of anxiety-related tweets in 2019 (i.e. the baseline) on a logarithmic scale. Taking into account the baseline level makes emotion levels comparable across countries, and shows how much emotions change from their usual level expressed in percent. A value of zero corresponds to the average level in 2019, and everything above or below to an increase or decrease in comparison to this level.

Observations for anxiety

• A typical day in the USA shows a frequency of anxiety terms in Twitter of about 3.4%.
• The events around Covid-19 led to the highest values so far, with 4.8% of tweets conttaining anxiety related words.
• This corresponds to an increase of 44% compared to an average day in 2019.

Prosocial behavior since the outbreak of Covid-19

In most countries in which expressions of anxiety have risen since the outbreak of Covid-19, expressions of empathy, willingness to support each other and cooperate (in short, prosocial behavior) have increased as well. The increase begins at different moments in different countries, probably depending on which information people had about Covid-19 in other countries, and on which measures were communicated in what way by their governments.

In the plot below, the lines indicate the following events:

• Green: First Covid-19 case on 2020-01-20
• Orange: First death from Covid-19 on 2020-02-29
• Turquois: Measures
  • Light: emergency plan on 2020-03-05
  • Dark: Social distancing measures start in some states from 2020-03-16 on.
• Grey: 12 March 2020: Since the outbreak of Covid-19, the day on which levels of prosociality in tweets increase above the average level from 2019 and remain high for several days in a row.

Percent change in prosocial tweets relative to the average level in 2019

In the figure below, a value of zero corresponds to the average level in 2019, and everything above or below to an increase or decrease in comparison to this level. The figure on the left displays the evolution over time since 1 January 2020. The small figure on the right shows the mean difference from the 2019 baseline before and after the day since which the level of prosocial terms on Twitter lies and remains above the baseline.

Observations for prosocial behavior

• Expressions of prosocial behavior and feelings in tweets were already slightly higher than average in the second half of February, and showed a peak on 5 March 2020, when congress voted for an emergency plan against Covid-19.
• After falling back to the baseline level (the average form 2019), they sharply increased by 8% shortly before the first states started implementing social distancing measures.

Conclusions

• It is clear that the events around the outbreak of Covid-19 and the communication of measures taken in response have increased levels of anxiety expressed in tweets. It seems like this increase may be lasting longer than those observed for other major events, such as mass shootings or the killing of the Iraqi commander Soleimani by the US military.
• Tweets that refer to some form of supporting others or empathy for others also showed an increase, although not as strong as the one for anxiety. The increase is clearest since social distancing measures are being implemented in several states across the US. However, since the US government reacted later than other countries, the development will still have to be observed in the next days.

Back to main page