In 1888, a French physician named Charles Féré noticed that the electrical resistance of a patient’s skin changed when he showed them a sensory stimulus. A year later, working independently in Russia, the physiologist Ivan Tarchanoff found the same thing and traced it to sweat. Neither man had the tools to know what he’d actually found: a readout of the sympathetic nervous system, delivered through the smallest glands in the body, running on a chemical messenger the rest of the sympathetic system barely uses. That readout became known as the galvanic skin response, then the psychogalvanic reflex, then, once physiologists insisted on precision, electrodermal activity, or EDA.
For more than a century, the field mostly treated it as a single dial. Skin conductance goes up, arousal goes up. Skin conductance is flat, the person is calm. That framework is still how Fitbit, Empatica, and half the wellness industry talk about it. But a run of research over the past two decades, particularly work coming out of psychophysiology labs in Graz, Austria, and Iowa City, has been quietly dismantling the one-dial model. The signal you can see on a graph, it turns out, is a crude average of something much messier underneath: dozens of independent sweat glands, controlled by a nervous system pathway that doesn’t behave like the rest of the body’s stress response, producing bursts that overlap, stack, and sometimes cancel each other out before they ever reach the electrode on your finger.
A Single Spike on the Graph Is Often Several Signals Stacked Together
Here’s the part that surprised even people who’ve spent careers staring at EDA traces. When you look at a skin conductance response, the visible bump you’d mark as “one event,” you’re not looking at one thing. You’re looking at the sum of multiple sudomotor nerve bursts, each firing a slightly different population of sweat glands at a slightly different moment, blurred together because the skin’s electrical response is slow compared to the nerve signals driving it.
Mathias Benedek and Christian Kaernbach, working at the University of Graz, built a method around this problem in 2010 called nonnegative deconvolution, later refined into what they termed continuous decomposition analysis. The technique mathematically unmixes a skin conductance trace back into its component nerve bursts instead of just counting peaks. What they found is that conventional peak-counting, the method nearly every consumer wearable still uses, systematically undercounts the actual sympathetic activity happening underneath. Two people can produce visually identical single-bump graphs while one of them experienced one burst of sympathetic drive and the other experienced four, compressed so tightly together they read as one.
One spike on the graph might be four sweat gland bursts arriving too close together to count separately.
This isn’t a footnote for statisticians. It means the entire premise of “counting SCRs per minute” as a proxy for stress load, which is how most lab studies and most apps still quantify arousal, is measuring a shadow of the real signal. The deconvolution methods exist and are published, but they require raw sampling rates and processing pipelines that almost no consumer device ships with. The research moved past simple peak-counting fifteen years ago. The products on your wrist mostly haven’t caught up.
Your Skin Reacts Before Your Brain Can Explain Why
The other complication is where the signal originates. Eccrine sweat glands, the ones packed densest into your palms and the soles of your feet, are driven by sympathetic nerve fibers that release acetylcholine. That’s unusual. Almost every other organ under sympathetic control, your heart, your blood vessels, your pupils, runs on norepinephrine. Eccrine sweat glands are the one major exception, wired more like a parasympathetic circuit even though they’re carried on sympathetic nerves. There’s no competing brake system pulling in the other direction the way there is for heart rate. Once the hypothalamus and the limbic structures feeding it decide something is worth a response, the signal goes out largely unopposed.
That wiring produces a strange property: the response can run ahead of conscious awareness. Antonio Bechara, Antoine Bechara’s colleague Antonio Damasio, and Daniel Tranel at the University of Iowa demonstrated this directly in their Iowa Gambling Task studies through the 1990s. Healthy subjects playing a card game with hidden risky and safe decks began generating skin conductance responses before risky choices well before they could articulate which decks were dangerous. Patients with damage to the ventromedial prefrontal cortex never developed that anticipatory signal, and they kept making the same bad bets over and over, unable to feel their way out of a pattern they couldn’t yet name. The finding became the empirical backbone of Damasio’s somatic marker hypothesis: the body sometimes knows before the mind does, and skin conductance is one of the few instruments sensitive enough to catch it happening.
The skin reacted to the risk before the subject could say why.
That’s a remarkable capability. It’s also exactly why treating EDA as a simple stress meter undersells and oversells it at the same time. It can pick up something before conscious cognition catches up, which is scientifically valuable and hard to fake. But it can’t tell you what that something is. Fear, excitement, a hard math problem, and a sudden loud noise all produce the same upward flick in conductance. The signal is fast and honest. It is not, on its own, specific.
Some People’s Skin Barely Registers Anything
There’s a further wrinkle that gets lost in every “wear this and see your stress score” pitch: not everybody’s skin talks at the same volume. David Lykken, working at the University of Minnesota in the late 1950s, found that individuals varied enormously in how strongly and how quickly their skin conductance recovered after a stimulus, and used that variability to study anticipatory fear in psychopathic individuals, who showed blunted responses before anticipated punishment. Later work by researchers including Michael Dawson, Anne Schell, and Diane Filion at the University of Southern California formalized the idea of electrodermal “responders” and “non-responders,” a real, stable individual difference in which some people show almost no measurable skin conductance response to standard laboratory stimuli, no matter how the stimulus is escalated.
This matters for the same reason a thermometer that reads differently on different bodies at the same actual temperature would matter. Skin conductance isn’t a fixed unit like millivolts of EEG or beats of a pulse. It rides on the number and sensitivity of sweat glands recruited, which varies by person, by hydration, by skin thickness, by ambient humidity, and by nothing at all related to emotional state. A flat trace on a wearable might mean a genuinely calm person. It might also mean a person whose electrodermal system has always run quiet, calm or not.
A flat line on the graph doesn’t always mean a calm person. Sometimes it just means quiet glands.
The Wearable on Your Wrist Is Guessing More Than It Lets On
None of this is purely academic. Empatica, the MIT Media Lab spinout founded by Rosalind Picard and Matteo Lai, built its Embrace device around EDA and motion sensing specifically to catch generalized tonic-clonic seizures, and the underlying research, published by Francesco Onorati, Giulia Regalia, and colleagues working with Picard’s group, showed the combination could flag seizures reliably enough that the device earned FDA clearance in December 2018 as a seizure alert system. That’s a genuine, hard-won clinical win, and it worked because the researchers constrained the problem tightly: nighttime, minimal confounding movement, one specific physiological event with a well-characterized signature.
Consumer stress tracking asks EDA to do something much harder with much less structure. Fitbit’s Sense line, launched in 2020, added an EDA Scan mode that asks you to sit still with your palm over the sensor for a couple of minutes, precisely because the signal is too fragile to interpret while you’re walking, typing, or adjusting the thermostat. Outside that narrow, deliberately staged window, motion artifact, skin temperature, and ambient humidity all move the needle as much as anything resembling emotion does. The honest limitation is this: EDA is genuinely informative about sympathetic arousal in controlled conditions, and genuinely ambiguous about anything specific, like whether that arousal is dread or anticipation, once you leave the lab.
| System | Context | What It Measures Well | Where It Breaks Down |
|---|---|---|---|
| Empatica Embrace2 (FDA cleared 2018) | Nocturnal seizure alert | One well-defined event, minimal background movement | Vigorous non-seizure motion still triggers false alarms |
| Fitbit Sense EDA Scan (2020) | Consumer stress check | Brief, deliberately still snapshot of arousal | Can’t run continuously; no protection from temperature or hydration drift |
| Empatica E4 research wristband | Lab and field research | Exports raw EDA for offline decomposition | Most users still just count peaks instead of deconvolving bursts |
| Iowa Gambling Task lab rig (Bechara & Damasio, 1990s) | Decision-making research | Captures anticipatory, pre-conscious arousal | Requires bulky lab-grade electrodes, not wrist wearables |
Set the two halves of this research next to each other and a specific picture forms. On one side, decomposition work out of Graz shows the raw signal contains far more structure than anyone reading a simple graph would guess, multiple nerve bursts hidden inside what looks like a single event. On the other, decades of individual-differences research shows the baseline volume of that signal varies by person in ways that have nothing to do with feeling. Put those together and you get a measurement that is simultaneously richer and less trustworthy than the marketing around it admits. The question worth asking isn’t whether your smartwatch can detect stress. It’s whether anyone selling you that promise has actually read the deconvolution papers, or whether they’re still just counting bumps on a line and calling it science.