The Research

What can a smartwatch tell us about emotions?

Affective computing and anxiety disorders, explained: what wearables measure, what the research shows, and what a smartwatch cannot tell you.

What is affective computing?

Affective computing is the study and design of technology that can recognise, interpret or respond to human emotions, a field named by Rosalind Picard in the 1990s [1]. In mental health, the appeal is simple to state and hard to deliver: technology that notices emotional patterns over days and weeks, outside the clinic, would give people and their clinicians more to work with than memory alone [2].

What a smartwatch actually measures

Consumer smartwatches such as the Google Pixel Watch 2 carry several sensors used in emotion research:

• Heart rate and heart rate variability, from an optical sensor (photoplethysmography, or PPG).
• Electrodermal activity (EDA): small changes in how well the skin conducts electricity as sweat glands respond to the sympathetic nervous system. The Pixel Watch 2 can measure it continuously (cEDA) [5].
• Skin temperature.
• Movement, from the accelerometer and gyroscope.

None of these measures an emotion. They measure how activated the body is, which rises with fear and with excitement alike, and also with a flight of stairs or a strong coffee.

From body signals to emotions

Emotion research often describes feelings along a few dimensions rather than as a list of labels. The two most common are valence, how pleasant or unpleasant a feeling is, and arousal, how calm or activated the body is [3]. Our review of affective computing in anxiety disorders found valence and arousal to be the most widespread way of quantifying emotion in this field [2].

Body signals say a good deal about arousal and much less about valence. A large meta-analysis found no consistent physiological “fingerprint” for individual emotion categories such as fear or anger [4]. The same racing heart can belong to very different feelings.

What the research says about anxiety disorders

Anxiety disorders are diagnosed conditions, such as generalised anxiety disorder, panic disorder, social anxiety disorder, obsessive-compulsive disorder (OCD) and post-traumatic stress disorder (PTSD), in which anxiety is too intense, lasts too long or disrupts daily life. Our rapid literature review [2] found:

• Most technology detects everyday anxiety, the normal response to a stressful moment. Far fewer studies address anxiety disorders.
• The evidence is uneven. Of 355 search results, 38 studies met our criteria, and PTSD was by far the most studied condition.
• There are promising uses, from tracking progress during behavioural therapies to assessing deep brain stimulation for intractable OCD.
• Combining several signals tends to work better than relying on one.
• Clinical use is still rare. More evidence is needed, especially beyond PTSD, and so is work on how these tools fit into care pathways.

What a smartwatch cannot tell you

• What you are feeling. A change in body signals can be panic, excitement or exercise. Only you can say which.
• Whether you have an anxiety disorder. Diagnosis needs a clinical assessment. No consumer wearable replaces it.
• Anything with certainty in daily life. Movement, a loose strap, heat and cold all add noise, and everyday EDA recordings in particular contain artefacts that need careful handling [6].

MEMoPAD is a research prototype. It is not a medical device, it is not publicly available, and it is not a substitute for professional care.

References

1. Picard, R. W. (1997). Affective computing. MIT Press.
2. Moretti, L. A., Thompson, M., Matthews, P., Loizou, M., & Western, D. (2025). Affective computing in anxiety disorders: A rapid literature review of emotion recognition applications. In Proceedings of BIOSTEC 2025, Volume 2: HEALTHINF (pp. 273–284). SCITEPRESS. Abstract and PDF
3. Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39(6), 1161–1178. doi:10.1037/h0077714
4. Siegel, E. H., Sands, M. K., Van den Noortgate, W., Condon, P., Chang, Y., Dy, J., Quigley, K. S., & Barrett, L. F. (2018). Emotion fingerprints or emotion populations? A meta-analytic investigation of autonomic features of emotion categories. Psychological Bulletin, 144(4), 343–393. doi:10.1037/bul0000128
5. McDuff, D., Thomson, S., Abdel-Ghaffar, S., Galatzer-Levy, I. R., Poh, M.-Z., Sunshine, J., Barakat, A., Heneghan, C., & Sunden, L. (2024). What does large-scale electrodermal sensing reveal? bioRxiv. doi:10.1101/2024.02.22.581472
6. Gashi, S., Di Lascio, E., Stancu, B., Das Swain, V., Mishra, V., Gjoreski, M., & Santini, S. (2020). Detection of artifacts in ambulatory electrodermal activity data. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 4(2). doi:10.1145/3397316