💬 NLP

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

By analyzing large-scale social media data with a specialized anxiety lexicon, this paper reveals that anxiety levels follow systematic temporal patterns—peaking at 8 a.m. and mid-week while dipping on weekends—and are significantly influenced by grammatical tense and pronoun usage, offering insights into how temporal focus and perspective shape human anxiety.

Original authors: Saif M. Mohammad

Published 2026-02-12

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Saif M. Mohammad

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

When does the human 'worry engine' run at full throttle?

Have you ever wondered why you suddenly feel a knot in your stomach on a Tuesday morning, while on Saturday afternoon everything seems just fine? Scientists have investigated when we as humans feel most anxious or restless using a kind of 'digital thermometer'. They did this not by asking people "how do you feel?", but by looking at the enormous mountain of messages on social media (such as X/Twitter).

Imagine that the collective emotion of millions of people is a vast ocean. The researchers used a special 'filter' that recognizes words we associate with anxiety (such as stress, scared, uncertain) and contrasts them with words that radiate calm (such as calm, relaxed, peaceful).

This is what they discovered about the tides of our emotions:

1. The daily heartbeat of anxiety

You can view the anxiety of a day as a wave going up and down:

The morning peak (8:00 AM): This is the moment when the 'worry engine' kicks in. It's as if the engine starts cold and immediately gives it full throttle. This is likely due to our body (cortisol, the stress hormone) shaking us awake for the workday.
The midday dip (around 12:00 PM): Around noon, the sea is calmest. We are in a rhythm, we are eating, and the biggest morning stress is over.
The evening wave: As the evening progresses, anxiety rises again. It is the moment when we start 'reworking' things in our heads: Did I do everything? What went wrong?

2. The week: The Tuesday dip and the weekend calm

If we look at the entire week, we see a pattern:

The weekend is a safe harbor: Anxiety is lowest here. We are free and the pressure is off.
The Wednesday peak: Surprisingly, anxiety is not highest on Monday (the notorious 'Monday Blues'), but on Wednesday. Why? Because the weekend freedom is long gone, but the new weekend still seems far away. It is the middle of the mountain we have to climb.

3. The time travel: Why we get stuck in the past

The researchers also looked at the time in which we speak. You might think anxiety is mainly about the future (what if...?), but the data show something else:

Anxiety sits in the rearview mirror: We express most anxiety in the past tense. It is as if we constantly ruminate on what has already happened (worrying about mistakes or traumas).
The future is a little sun: In the future tense, we use strikingly many calming words. Despite everything, we are often a little optimistic about what is yet to come.

4. The 'I' versus the 'Others'

Finally, they looked at how we talk about ourselves and others:

The spotlight on the other: We are often more anxious when we talk about others (he, she, they) than when we talk about ourselves. It is as if we see the world around us as an unpredictable, exciting place.
The active role: We are also more anxious when we are the 'subject' of a sentence (e.g., "I feel...") than when we are the 'object' (e.g., "It touches me...").

The conclusion in one sentence:

Our anxiety is not constant noise, but a rhythm that moves with our biological clock, our workweek, and the way we view time and the people around us.

Technical Summary: "When are We Worried? Temporal Trends of Anxiety and What They Reveal about Us"

Problem Statement

While anxiety is a fundamental human emotion with significant evolutionary and psychological implications, much remains unknown about its large-scale temporal patterns in the digital age. Traditional psychological research relies on self-reported data, which is often prone to biases such as recency bias, social desirability bias, and temporal aggregation bias. Furthermore, there is a growing debate regarding whether modern technological environments (like social media) create a mismatch with the biological mechanisms that evolved to manage anxiety. The paper seeks to address these gaps by using social media language as a high-volume, real-time window into collective anxiety levels.

Methodology

The researcher employs a computational linguistics approach to analyze large-scale social media data from North America. The methodology consists of three primary components:

Data Sources:
- WorryWords: A recently developed lexicon containing approximately 44,500 English words. It categorizes words based on their association with either anxiety (~~26%) or calmness (~~13%).
- TUSC (Twitter User Sentiment Corpus): A large corpus of lemmatized, lower-cased, and geo-located tweets from the US and Canada spanning 2015 to 2021.
Metric Calculation:
- The study uses an aggregate anxiety score calculated as:
  $\text{Anxiety Score} = (\% \text{ of anxiety-associated words}) - (\% \text{ of calmness-associated words})$
- A negative score indicates a higher prevalence of calmness-associated words.
Experimental Dimensions:
- Temporal Analysis: Mapping tweets to local time (0–23 hours) and day of the week (7 days).
- Tense Analysis: Partitioning tweets into past, present, and future tenses based on verb usage and modal/signal words (e.g., "will," "tomorrow").
- Pronoun Analysis: Examining the relationship between anxiety and different types of focus (self, social, or outward) by analyzing the use of 1st/2nd/3rd person and subject/object pronouns.

Key Contributions

Validation of Social Media as a Proxy: The paper provides empirical evidence that aggregate-level anxiety expressed in social media posts aligns with established psychological findings on self-reported anxiety.
Multi-dimensional Mapping of Anxiety: It moves beyond simple "sentiment analysis" to map anxiety against specific cognitive focuses: Existential/Task/Social Focus (via time), Temporal Focus (via tense), and Relational Focus (via pronouns).
Baseline Establishment: It establishes a linguistic baseline for "normal" anxiety levels, which can be used to detect anomalies in specific contexts (e.g., anxiety during climate change discussions or vaccine debates).

Results

Temporal Patterns:
- Daily: Anxiety peaks at 8:00 AM (consistent with biological cortisol spikes) and reaches its lowest point around noon. It rises again in the evening.
- Weekly: Anxiety is lowest on weekends and highest mid-week (Wednesday). Interestingly, the study found no evidence of "Sunday Scaries" in the aggregate data, suggesting weekend calmness outweighs Sunday evening apprehension.
Tense Patterns:
- Anxiety is highest in the past tense and lowest in the future tense. This suggests that while anxiety is conceptually about future uncertainty, it is linguistically expressed through rumination on the past and present.
Pronoun Patterns:
- Social Focus: 3rd-person pronouns (he, they) are associated with higher anxiety than 1st or 2nd person.
- Grammatical Focus: Subject pronouns (I, he, she, they) exhibit higher anxiety scores than their corresponding object pronouns (me, him, her, them).

Significance

This research is significant because it bridges the gap between computational linguistics and behavioral psychology. By demonstrating that linguistic patterns on social media reflect biological and psychological realities, the study provides a scalable method for monitoring public mental health trends. The findings offer insights into how human focus (on the self, the past, or others) correlates with emotional states, providing a foundation for future research into how different cultures and socio-economic groups experience and express anxiety in a digital-first world.