What Sleep Trackers Can And Cannot Measure

Wearable devices report a nightly breakdown of sleep stages with considerable confidence. Understanding how those figures are produced explains which of them deserve attention.

What laboratory measurement actually requires

Sleep staging in a laboratory uses electrical recordings from the scalp, around the eyes and at the chin, scored in short epochs against agreed criteria.

The three signals matter because stages are defined by combinations of them. Dreaming sleep, for instance, is identified partly by muscle tone dropping away while the brain looks close to awake.

No wrist-worn device records any of these. Whatever a tracker reports about stages is inferred from other signals entirely.

What the devices do record

Accelerometers register movement, and optical sensors estimate pulse and its variability from blood flow at the skin. Some add temperature or breathing rate derived from the same source.

These signals do correlate with sleep. People move less when asleep, heart rate falls, and its variability changes in ways that differ between deeper and lighter sleep.

Algorithms map those correlations onto stage labels. The mapping is statistical, trained on populations, and applied to an individual whose physiology may not match the training set.

Which numbers hold up

Total sleep time and the timing of sleep onset and waking are the most dependable outputs, because movement is a reasonably good indicator of the difference between sleeping and being awake.

Night-to-night consistency in bedtime and rise time is tracked well and is arguably the most useful thing these devices provide, since regularity is genuinely important.

Stage breakdowns are the weakest output. Comparisons against laboratory scoring find agreement that is far from complete, and deep and dreaming sleep are the categories most often misassigned.

The systematic errors worth knowing

Lying still while awake looks like sleep to a movement-based system, so trackers tend to overestimate sleep in people who spend long periods motionless in bed.

That bias falls hardest on those with difficulty sleeping, who are precisely the users most likely to be scrutinising the output.

The reverse also occurs. Restless sleepers can have genuine sleep scored as wakefulness, producing a picture worse than the night actually was.

When the data becomes the problem

Sleep clinicians have described a pattern in which anxious attention to nightly scores becomes a cause of poor sleep in itself, with users pursuing a metric rather than rest.

A single night's figures carry little information in any case. Trends across weeks are where a consumer device has something to say, and individual scores are noise around them.

Where a tracker suggests something clinical, such as frequent breathing disruption, that is a prompt to see a clinician rather than a diagnosis. The device flags a possibility; a proper study settles it.