When are We Worried? Temporal Trends of Anxiety and What They Reveal about Us

In this short paper, we make use of a recently created lexicon of word-anxiety associations to analyze large amounts of US and Canadian social media data (tweets) to explore when we are anxious and what insights that reveals about us. We show that our levels of anxiety on social media exhibit systematic patterns of rise and fall during the day – highest at 8am (in-line with when we have high cortisol levels in the body) and lowest around noon. Anxiety is lowest on weekends and highest mid-week. We also examine anxiety in past, present, and future tense sentences to show that anxiety is highest in past tense and lowest in future tense. Finally, we examine the use of anxiety and calmness words in posts that contain pronouns to show: more anxiety in 3rd person pronouns (he, they) posts than 1st and 2nd person pronouns and higher anxiety in posts with subject pronouns (I, he, she, they) than object pronouns (me, him, her, them). Overall, these trends provide valuable insights on not just when we are anxious, but also how different types of focus (future, past, self, outward, etc.) are related to anxiety.

💡 Research Summary

This paper investigates when people feel anxious by analyzing large‑scale social‑media text using a newly created anxiety‑lexicon called “WorryWords.” The authors combine the lexicon with the TUSC corpus, a collection of geo‑located tweets from the United States and Canada spanning 2015‑2021. WorryWords contains about 44,500 English terms, of which roughly 26 % are marked as anxiety‑associated and 13 % as calmness‑associated; inter‑annotator reliability is high (Pearson r = 0.89).

The methodology is straightforward: each tweet is assigned to an hour‑of‑day bin (0‑23) and a day‑of‑week bin (Monday‑Sunday) based on its local timestamp. For each bin the authors compute an aggregate anxiety score defined as the percentage of anxiety‑associated words minus the percentage of calmness‑associated words. The overall mean score across the entire dataset is –15.13, indicating that, on average, calmness words are used more often than anxiety words. Statistical significance is assessed with t‑tests (α = 0.05).

Daily patterns. The anxiety score peaks at –13.5 around 8 a.m., the time when cortisol levels rise after waking. It drops to its lowest –17.4 around noon, then climbs again in the late afternoon and reaches a secondary peak near 10 p.m. All pairwise differences between adjacent time windows (e.g., 5‑8 a.m., 8‑10 a.m., 10 a.m‑12 p.m., 12‑2 p.m., 2‑6 p.m.) are statistically significant. This diurnal rhythm mirrors physiological findings that anxiety is highest shortly after awakening, declines during routine activities, and rises again in the evening when people reflect on unfinished tasks.

Weekly patterns. When aggregated by weekday, the highest anxiety occurs on Wednesday (score ≈ –14.95) and the lowest on weekends, especially Sunday (score ≈ –15.32). The difference between Sunday and Wednesday is significant. The magnitude of change across weekdays (≈ 0.4 points) is far smaller than the change across hours of the day (≈ 4 points), suggesting that time‑of‑day exerts a stronger influence on expressed anxiety than day‑of‑week.

Tense analysis. Tweets are classified into three tense categories: past‑tense (≈ 47 % of tweets), present‑tense (≈ 52 %), and future‑tense (≈ 1 %). Past‑tense tweets have the highest anxiety score (–14.1), present‑tense tweets are lower (–16.4), and future‑tense tweets are the lowest (–21.9), well below the overall mean. All pairwise differences are significant. Although anxiety is conceptually linked to future uncertainty, the language data show that people express anxiety mainly when recounting past events or describing present circumstances, while future‑oriented statements tend to contain more calmness‑related vocabulary, possibly reflecting a general optimism about what lies ahead.

Pronoun analysis. The authors examine how pronoun usage relates to anxiety. “I” appears in about 60 % of tweets, “you” in 31 %, and other pronouns much less frequently. Overall, tweets containing any pronoun have a lower anxiety score (–17.77) than the corpus average, perhaps because users avoid overtly negative language when referring to themselves or others. However, specific pronouns differ: third‑person pronouns “he” and “they” show higher anxiety scores than first‑person “I” and second‑person “you,” while subject pronouns (I, he, she, they) are more anxious than their corresponding object forms (me, him, her, them). This pattern suggests that references to others, especially third‑person subjects, are associated with more anxiety‑laden language, whereas self‑references may be moderated by social desirability or self‑presentation concerns.

Implications and limitations. The study demonstrates that large‑scale lexical analysis can capture temporal anxiety trends that align with physiological and psychological literature, offering a complementary perspective to self‑report surveys that suffer from recall and social‑desirability biases. The authors propose that these baseline patterns can serve as reference points for domain‑specific investigations (e.g., climate‑change discourse, vaccine debates, immigration, hate speech). Limitations include the geographic and linguistic scope (US/Canada English), potential cultural bias in the lexicon (annotated primarily by US‑based crowdworkers), and the inability of a static lexicon to capture sense‑disambiguation in specialized contexts. Ethical considerations are addressed: the work aggregates at the population level, does not infer individual mental‑health status, and acknowledges biases inherent in the resources used.

Conclusion. By leveraging the WorryWords lexicon and the TUSC tweet corpus, the authors map when North Americans express anxiety on social media. They find a clear diurnal rhythm (peak at 8 a.m., trough at noon), a weekly rhythm (mid‑week peak, weekend low), higher anxiety in past‑tense language, and pronoun‑dependent variations that reflect self‑ versus other‑focus. These findings provide a data‑driven foundation for future cross‑cultural, multilingual, and topical studies of anxiety in digital communication.