EEG vs. Functional Connectivity

Your Brain Has No Solo Artists

Here's a fact that should bother you more than it probably does.

For most of the history of brain science, we studied the brain by looking at individual locations. This region lights up when you see a face. That region activates when you feel fear. Another region handles language. We carved the brain into a map of specialized territories, like a geopolitical map of a very wrinkly country, and declared the job mostly done.

There's just one problem. The brain doesn't work that way. Not even close.

Think about it like this. If you wanted to understand how the internet works, you could study individual computers. You could catalog every server, describe its hardware, measure its processing speed. You'd learn a lot. But you would completely miss the thing that makes the internet the internet: the connections. The network. The fact that a computer in Virginia and a computer in Tokyo are exchanging packets of information thousands of times per second, coordinating to deliver the video you're watching or the page you're reading right now.

The brain is the same. Except instead of fiber optic cables, it uses synchronized electrical oscillations. And instead of billions of computers, it has 86 billion neurons. And instead of sending cat videos, it's generating your entire conscious experience.

For decades, EEG analysis has been the equivalent of studying individual computers. You measure the power at a single electrode location. How much alpha is the left frontal cortex producing? What about beta at the parietal site? This is useful information, genuinely useful, but it's only half the story.

The other half is functional connectivity: the study of how brain regions talk to each other. And when you start looking at the brain this way, as a network instead of a collection of spots, the picture changes so dramatically that it feels like putting on glasses for the first time.

The Traditional Approach: Power at a Point

Before we can appreciate what connectivity reveals, we need to understand what traditional EEG analysis actually does and why it's been the default approach for nearly a century.

When Hans Berger recorded the first human EEG in 1924, he noticed something immediately: the electrical signals from the brain oscillated at regular rhythms. He identified what we now call the alpha rhythm, those 8 to 13 Hz waves that appear prominently when you close your eyes and relax. This was a single-channel observation. One electrode, one location, one signal.

From that starting point, the entire field of clinical and research EEG grew around the same basic idea: place an electrode somewhere on the scalp, measure the electrical activity underneath it, and break that activity down by frequency.

This is spectral power analysis, and it works like this. The raw EEG signal at any electrode is a messy, chaotic-looking squiggle. But hidden inside that squiggle are overlapping waves at different frequencies, like multiple radio stations broadcasting simultaneously. A mathematical technique called the Fast Fourier Transform (FFT analysis) separates these overlapping frequencies, telling you exactly how much power (amplitude) exists at each frequency band.

The standard frequency bands you'll see everywhere in EEG research:

• Delta (0.5 to 4 Hz): Deep sleep, unconscious processing
• Theta (4 to 8 Hz): Memory encoding, drowsiness, meditative states
• Alpha (8 to 13 Hz): Relaxed wakefulness, sensory idling, "the brain on screensaver"
• Beta (13 to 30 Hz): Active thinking, concentration, alertness
• Gamma (30 to 100 Hz): Information binding, high-level processing, consciousness itself (maybe)

Traditional EEG analysis measures these power values at each electrode site independently. You end up with a topographic map: a bird's-eye view of the brain showing hotspots and cold zones of activity at different frequencies. It tells you things like "the right frontal region is producing more high-beta than the left," which can indicate anxiety or rumination. Or "parieto-occipital alpha is elevated," which means the visual cortex is idling, probably because the person has their eyes closed.

This approach has been enormously productive. It's the foundation of neurofeedback. It's how we diagnose certain forms of epilepsy. It's how consumer EEG devices like the Neurosity Crown generate focus and calm scores. It works.

But there's something it fundamentally cannot tell you.

The Missing Piece: Relationships

Here's where it gets interesting. Imagine you're watching an orchestra perform. Traditional EEG power analysis is like measuring the volume of each instrument. You can tell that the violins are loud and the flute is quiet. You can track how the trumpet's volume changes over time. Useful? Sure.

But you're completely missing whether the violins and cellos are playing the same melody in sync. You're missing whether the percussion is locked to the rhythm of the brass section. You're missing, in other words, the thing that turns individual instruments into music: coordination.

This gap isn't just an academic inconvenience. It has real consequences for understanding brain function and dysfunction.

Consider two people who both show elevated frontal beta power. Traditional EEG analysis says they look similar. But what if Person A shows high beta that's synchronized across frontal sites (coordinated activation suggesting intense focused thought), while Person B shows high beta that's completely desynchronized across the same sites (fragmented activation suggesting anxious rumination)? Same power. Completely different brain states. You'd never know the difference without looking at the connectivity.

This is the fundamental limitation of treating each electrode as an independent observation. In a system defined by connections, measuring individual nodes in isolation is like trying to understand a conversation by measuring how loud each person is talking, without listening to whether they're actually responding to each other.

Functional Connectivity: Now We're Talking (Literally)

Functional connectivity analysis starts from a radically different premise. Instead of asking "how active is this brain region?", it asks "are these two brain regions communicating?"

The word "functional" is important. It distinguishes this from structural connectivity, which refers to the physical white matter tracts that wire brain regions together (think of those gorgeous images from diffusion tensor imaging, the ones that look like a rainbow explosion inside a skull). Structural connectivity is the hardware. Functional connectivity is the software. The physical wires don't change from moment to moment, but the patterns of communication running over those wires change constantly, reconfiguring in real time as your brain shifts from task to task.

Here's the key insight: when two brain regions are communicating, their electrical activity becomes statistically related. Their oscillations sync up, like two pendulum clocks on the same wall that gradually fall into rhythm. This synchronization isn't random. It's the mechanism by which the brain coordinates distributed processing. The visual cortex and the frontal cortex need to sync up when you're trying to find your friend in a crowded room. The motor cortex and the parietal cortex need to sync up when you're reaching for your coffee. The prefrontal cortex and the hippocampus need to sync up when you're committing something to memory.

And here's the beautiful part: EEG can see this synchronization. The same raw data that traditional power analysis uses contains all the information you need to study connectivity. You just have to look at the relationships between channels instead of looking at each channel alone.

The Toolbox: How Connectivity Gets Measured

There are several ways to quantify how two EEG signals relate to each other. Each captures a slightly different aspect of the relationship, and each has its strengths and quirks.

Coherence is the most established measure. Think of it as a correlation coefficient, but in the frequency domain. It tells you how consistently two signals share a common frequency pattern over time. Coherence ranges from 0 (the two signals have nothing in common at that frequency) to 1 (they're perfectly synchronized). High alpha coherence between the frontal and parietal electrodes, for example, suggests those regions are coordinating their alpha rhythms, possibly to maintain attention or manage a working memory task.

Phase-locking value (PLV) is more precise about timing. Two oscillations can have the same frequency but different timing. Imagine two violinists playing the same note but starting at slightly different moments. PLV asks: is the phase relationship between these two signals consistent over time? If the frontal signal always peaks 50 milliseconds before the parietal signal, that's a strong, stable phase relationship, suggesting a directed communication pathway. PLV strips out amplitude information entirely and focuses purely on timing consistency.

Granger causality goes a step further and asks about direction. If the past values of Signal A help predict the current values of Signal B (beyond what Signal B's own past values predict), then A is said to "Granger-cause" B. This isn't true causation in the philosophical sense, but it reveals the direction of information flow between regions. Knowing that frontal activity predicts parietal activity (but not the reverse) during a working memory task tells you something profound about how the brain organizes top-down control.

Phase lag index (PLI) deserves special mention because it solves one of the biggest headaches in EEG connectivity. Remember volume conduction, the way electrical signals smear across the skull? Volume conduction creates artificial correlations between nearby electrodes. If electrode A and electrode B both pick up the same underlying source because of spatial smearing, they'll look highly coherent even though no actual communication is happening. PLI only counts phase relationships with a nonzero time lag, filtering out the instantaneous correlations that volume conduction creates. It's less sensitive, but what it finds is more likely to be real.

The Graph Theory Revelation: Your Brain Is a Small World

Here's where the "I had no idea" moment hits.

Once you've computed connectivity between all your electrode pairs, you have a connectivity matrix: a table showing how strongly each pair of electrodes is connected. With 8 electrodes, that's 28 unique pairs. With 64 electrodes, it's 2,016 pairs.

Now, you can treat that matrix as a network graph. Each electrode is a node. Each significant connectivity value is an edge. And suddenly, the entire mathematical framework of graph theory, the same mathematics used to study social networks, airline routes, and the internet, becomes available to study your brain.

This is where things get genuinely wild.

In 1998, mathematicians Duncan Watts and Steven Strogatz published a paper describing "small-world" networks. These are networks that have two seemingly contradictory properties: they're highly clustered locally (your neighbors are connected to each other) and yet any two nodes can reach each other through a surprisingly small number of steps. Think of your social network. Your close friends mostly know each other (high clustering). But through a chain of maybe five or six acquaintances, you could reach almost anyone on the planet (short path length). That's a small-world network.

The human brain, it turns out, is a small-world network. And not just a little bit. It's one of the most perfectly small-world networks ever measured in nature.

Graph theory gives us mathematical tools to quantify these network properties:

• Clustering coefficient: How densely interconnected your brain's local neighborhoods are. High clustering means brain regions that work together are tightly coupled.
• Characteristic path length: How many "hops" it takes, on average, to get from any node to any other node. Short path length means efficient global communication.
• Hub identification: Some brain regions serve as major relay stations, connecting to many other regions. Damage a hub and the whole network suffers.
• Modularity: The degree to which the brain organizes into distinct functional communities, like specialized departments in a company that also cooperate across divisions.

And here's the thing that keeps neuroscientists up at night: these network properties change in predictable ways in brain disorders. Alzheimer's disease disrupts the brain's small-world architecture. Schizophrenia alters the brain's hub structure. ADHD brain patterns shows altered connectivity patterns between frontal and parietal network modules. Depression shifts the balance between network integration and segregation.

Traditional EEG power analysis can detect some of these conditions. But connectivity analysis detects them earlier, more specifically, and with better differentiation between conditions that look similar in power spectra but have very different network signatures.

The Head-to-Head: Power vs. Connectivity

Let's put these two approaches side by side and be explicit about what each one reveals across real use cases.

Brainwave data, captured at 256Hz across 8 channels, processed on-device. The Crown's open SDKs let developers build brain-responsive applications.

Explore the Crown

The pattern is consistent. Power analysis provides a useful first approximation. Connectivity analysis provides a deeper, more specific, and often more clinically relevant picture. But notice that the answer isn't always "connectivity wins." For many applications, the two approaches complement each other. The best analyses use both.

The Complexity Trade-Off

I'd be dishonest if I didn't acknowledge the elephant in the room. Connectivity analysis is harder. Significantly harder.

Traditional power analysis is computationally simple. Take a window of EEG data, run an FFT, look at the power values. A first-year grad student can learn it in an afternoon. A developer can implement it in a few dozen lines of Python.

Connectivity analysis involves choosing between multiple metrics (each with its own assumptions), dealing with volume conduction artifacts, setting statistical thresholds for "significant" connections, handling multiple comparisons (28 pairs from 8 channels means 28 statistical tests), and interpreting the resulting network in a way that's meaningful.

Graph theory analysis adds another layer on top of that: binarizing the connectivity matrix, choosing threshold criteria, computing global and local metrics, and comparing against null models to confirm that the network properties you're seeing aren't just noise.

This complexity is why traditional power analysis still dominates consumer EEG applications. It's not that power analysis is better. It's that it's simpler. But the gap is closing fast.

Eight Channels, Twenty-Eight Conversations

Now, here's where this becomes personally relevant if you're someone who owns or is considering a consumer EEG device.

The Neurosity Crown has 8 channels. For traditional power analysis, that gives you 8 independent measurements across the scalp. Useful, and already enough for focus tracking, calm monitoring, neurofeedback, and basic BCI.

But for connectivity analysis, 8 channels gives you something more: 28 unique electrode pairs. That's 28 separate "conversations" you can eavesdrop on. Frontal-to-parietal coherence. Left-to-right hemisphere synchrony. Central-to-occipital phase coupling. The network of relationships between those 8 locations contains far more information than the 8 locations measured independently.

The Crown's electrode positions (CP3, C3, F5, PO3, PO4, F6, C4, CP4) are distributed across both hemispheres and all four lobes. This wasn't just designed for good spatial coverage of power. It was designed so that the pairwise relationships between electrodes span meaningful network dimensions: fronto-parietal (attention), inter-hemispheric (integration), and centro-occipital (sensory-motor coordination).

With the Crown's raw EEG data accessible at 256Hz through its JavaScript and Python SDKs, developers can compute coherence, phase-locking value, and even basic graph metrics in real time. You're not limited to the focus and calm scores that the on-device processing provides (though those are great starting points). You can go deeper. You can build applications that respond not just to how active your brain is, but to how connected it is.

This is the difference between knowing "your frontal cortex is active" and knowing "your frontal cortex is actively coordinating with your parietal cortex." One tells you a brain region is working. The other tells you your attention network is engaged.

Clinical Connectivity: Where the Real Impact Lives

The clinical implications of connectivity analysis are staggering, and they represent one of the strongest arguments for why this approach matters beyond academic curiosity.

Alzheimer's disease is perhaps the most dramatic example. Traditional EEG shows generalized slowing of brain rhythms, more delta and theta, less alpha and beta. But this pattern only becomes reliably detectable after significant cognitive decline. Functional connectivity analysis tells a different story. Years before the first symptoms appear, the brain's small-world network begins to degrade. The characteristic path length increases (communication becomes less efficient). Hub regions in the posterior cortex lose their central position. The network becomes more "random" and less organized. These changes are detectable with EEG-based connectivity measures, potentially years before a clinical diagnosis.

ADHD shows a different connectivity fingerprint. The classic EEG finding is an elevated theta/beta ratio at frontal sites. It's useful, but it's not specific enough, lots of conditions show altered theta/beta ratios. Connectivity analysis reveals something more precise: weakened coupling between the frontal control network and the parietal attention network, combined with excessive connectivity within the default mode network (the brain's "daydreaming" system). This pattern is more specific to ADHD and tracks more closely with symptom severity.

Depression disrupts the balance between integration and segregation in brain networks. Depressed brains show hyperconnectivity within the default mode network (the brain can't stop talking to itself about itself) and weakened connectivity between this network and the cognitive control regions that would normally interrupt rumination.

These aren't theoretical findings. They're driving the development of connectivity-based biomarkers that could transform how we screen for, diagnose, and monitor brain health conditions. The same EEG signals that consumer devices already capture contain this information. The question is whether we look for it.

The Near Future Is Connected

We're at an inflection point. For 100 years, EEG analysis has mostly been about power at individual locations. The tools existed for connectivity analysis, but they were confined to specialized research labs with 64-channel or 256-channel systems, expensive software, and PhDs to run them.

That's changing. And it's changing because of three converging trends.

First, the computational tools have matured. Open-source libraries like MNE-Python, the Brain Connectivity Toolbox, and NetworkX make connectivity and graph analysis accessible to anyone who can write basic code. What required custom MATLAB scripts and a neuroscience PhD 10 years ago now requires a pip install and a tutorial.

Second, consumer EEG hardware has reached the quality threshold for meaningful connectivity analysis. The Crown's 256Hz sample rate provides clean frequency resolution up to 128Hz. Its 8-channel spread covers the major network nodes. Its open SDK gives developers raw data access without vendor lock-in.

Third, and this is the big one, AI is transforming what's possible with limited channel counts. Machine learning models trained on high-density EEG data can learn to estimate network properties from sparse electrode arrays. A model trained on 64-channel connectivity data can learn which features from 8 channels best predict the full network state. This means that an 8-channel system doesn't have to compute everything from scratch. It can use learned patterns to infer network properties that go beyond what the raw channel count would suggest.

The Neurosity MCP integration, which connects the Crown's brain data directly to AI systems like Claude, opens a particularly interesting door here. Imagine an AI that doesn't just receive your EEG power values, but tracks how your brain's connectivity patterns shift throughout the day, learning your personal network signatures for focus, fatigue, creative flow, and cognitive overload.

The Question We've Been Avoiding

We started with a seemingly technical comparison: traditional EEG power analysis versus functional connectivity analysis. But the real question underneath all of this is more fundamental, and a little unsettling.

If the brain is a network, and its network properties determine everything from your ability to focus to your vulnerability to neurological disease, then measuring individual brain regions in isolation isn't just incomplete. It's measuring the wrong thing.

For a century, we've been measuring the volume of individual instruments and calling it music analysis. We've been studying individual neurons and brain regions while the thing that makes a brain a mind, the connections, the synchronization, the network, has been hiding in plain sight. Hiding in the relationships between the very signals we were already recording.

Every EEG ever recorded contains connectivity information. Every brainwave dataset ever collected has network architecture encoded in it. We just weren't asking the right questions.

Now we are.

And here's what keeps me up at night. Right now, while you're reading this, your brain's network is reconfiguring itself moment by moment. Your fronto-parietal attention network is coupling up to process these words. Your default mode network is occasionally breaking through when your mind wanders. The phase relationships between your frontal and posterior cortex are shifting with every paragraph, every new idea, every moment of confusion or clarity.

All of this is happening in electrical signals on the surface of your scalp. Signals that an 8-channel device on your head could detect. The question isn't whether we have the technology to listen to the brain's network. We do. The question is what we'll build once we start listening to the conversations instead of the voices.