So, what did
we learn?
We’ve recorded the signal, checked for unwanted activity and separated its rhythms. We can now ask two different questions: which rhythms changed, and how did the brain respond to an event? Both need a clear comparison before we interpret them.
1. Which rhythms changed?
Delta, theta, alpha, beta and gamma are frequency ranges. We look at where a rhythm was recorded, when its power changed and what the person was doing.7
Choose a band for a research example and the comparison that helps interpret it.
Why compare? More power has no universal meaning. A change during sleep, movement or an attention task can answer different questions, even if it falls in the same frequency band. Keep recording and processing choices consistent, and check the background beneath a peak, as we did on page 3.2
2. What followed an event?
To study a response to a sound, picture or button press, we mark exactly when it happened, then line up short EEG segments at that moment. The responses tied to those events are called event-related potentials (ERPs).6
On the other page, try a P300: a response often studied when someone detects an occasional target among frequent non-targets. It answers a question about processing that event, rather than assigning a feeling to an entire session.5
Read further: foundations & practical guides
1. Barry et al. (2007). EEG differences between eyes-closed and eyes-open resting conditions. Clinical Neurophysiology.
2. Keil et al. (2022). Recommendations and publication guidelines for studies using frequency domain and time-frequency domain analyses of neural time series. Psychophysiology.
3. Poldrack (2006). Can cognitive processes be inferred from neuroimaging data? Trends in Cognitive Sciences. A general discussion of inferring mental processes from brain measurements, not an EEG-device validation.
4. Varoquaux et al. (2017). Assessing and tuning brain decoders: Cross-validation, caveats, and guidelines. NeuroImage.
5. Start with P300. Polich (2007). Updating P300: An integrative theory of P3a and P3b. Clinical Neurophysiology. Explains the components, tasks and factors affecting them.
6. How to run an ERP study. Picton et al. (2000). Guidelines for using human event-related potentials to study cognition. Psychophysiology. Recording, timing, analysis and reporting standards.
7. Why rhythms matter. Buzsáki & Draguhn (2004). Neuronal oscillations in cortical networks. Science. A foundational overview of network rhythms; not a dictionary of mental states.
8. Sleep rhythms. Steriade, McCormick & Sejnowski (1993). Thalamocortical oscillations in the sleeping and aroused brain. Science.
9. Theta. Cavanagh & Frank (2014). Frontal theta as a mechanism for cognitive control. Trends in Cognitive Sciences.
10. Alpha. Jensen & Mazaheri (2010). Shaping functional architecture by oscillatory alpha activity: Gating by inhibition. Frontiers in Human Neuroscience. An influential interpretation, rather than the only account of alpha.
11. Beta. Engel & Fries (2010). Beta-band oscillations—signalling the status quo? Current Opinion in Neurobiology.
12. Check muscle activity. Whitham et al. (2007). Scalp electrical recording during paralysis: Quantitative evidence that EEG frequencies above 20 Hz are contaminated by EMG. Clinical Neurophysiology.
13. The early P300 finding. Sutton et al. (1965). Evoked-potential correlates of stimulus uncertainty. Science. A landmark original experiment.
All plots and numbers here are invented teaching examples, not participant data or device performance estimates.
7Find a response to a target
You’re shown a series of circles and occasional squares, and asked to count the squares. Each picture presentation is one trial; squares are the targets because they are the pictures you’re asked to notice.
Select a picture to view its recording, then compare the averages. 0 ms marks when the picture appeared, so the horizontal axis tells us how long afterwards a response occurred.
What are we looking for? In the average view, compare the upward response to squares and circles a few hundred milliseconds after the picture. A larger response to targets is often seen in tasks like this and is called the P3b, part of the P300 family. Its peak can occur later than 300 ms.5
What does it tell us? The brain responded differently to the two picture types in this task. It does not tell us whether someone’s attention is “good” or “bad.” To investigate attention, we would also examine task performance and compare well-controlled conditions.5
Why does averaging help here?
Because every segment starts from the same picture onset, responses that occur consistently after it can remain in the average, while unrelated activity tends to shrink. If the picture times are marked incorrectly, or the response arrives at different times across trials, the average can blur it, so we check timing and the individual recordings too.6
When averaging hides a change
Same average, different story
Why it matters: a session average can hide a short burst or a period when a rhythm disappeared. Keep the measurements over time if your question is about brief changes.2
How does this become an app’s “focus” score?
From measurements to predictions
Software may use EEG measurements to estimate “focus.” To understand that score, we need to know what counted as focus during development and how well it predicts that outcome in new recordings.3
Here, the labels are rest and doing an attention task. Each point represents one recording. The model tries to distinguish these conditions using a measurement from EEG; this does not directly measure how focused someone feels.
What would make a score useful?
Test it on recordings kept out of training and model selection. If people will use it at home, it needs testing at home; if it is intended for new users, those users should be excluded from training. Check how often it is wrong, and whether acting on the score helps. If you change a model after seeing its test results, use a fresh test set for the final evaluation.4 Correctly identifying a task doesn’t by itself show that a score improves attention.