What does each method measure?
Polysomnography (PSG), the overnight lab sleep study, is the reference that wearables and actigraphs are scored against. Actigraphy estimates sleep and wake from wrist movement, and the American Academy of Sleep Medicine (AASM) guideline on it applies only to FDA-approved actigraphy devices. Consumer wrist and ring wearables also estimate sleep from sensor data, and a 2024 Sleep Research Society review found that the evidence suggests they "exceed the performance of traditional actigraphy in assessing sleep as defined by polysomnogram."
| Method | What it is | Sleep/wake vs. stages | Typical validation design |
|---|---|---|---|
| Polysomnography (PSG) | Lab sleep study | Reference standard for sleep/wake and stages | Used as the reference |
| Actigraphy | Wrist-worn movement sensor | Sleep/wake | Epoch-by-epoch comparison with PSG (for example, 30-second epochs) |
| Consumer wrist and ring wearables | Consumer devices with motion and other sensors | Sleep/wake, and stages on many devices | Epoch-by-epoch comparison with PSG in the lab, or with actigraphy in free-living studies |
Sources: Marino et al., Sleep 2013; Chinoy et al., Sleep 2021; de Zambotti et al., Sleep 2024; Smith et al., AASM guideline, J Clin Sleep Med 2018.
How validation studies report accuracy
Most studies split the night into short epochs and compare each one with PSG. Sensitivity is how often the device says sleep when PSG says sleep, and specificity is how often it says wake when PSG says wake. Because people are asleep for most of the night, sensitivity is usually high and specificity is the weak spot. In a study of 77 people and 232,849 30-second epochs, wrist actigraphy had sensitivity 0.965, accuracy 0.863 and specificity 0.329. In a 2021 lab study of seven consumer devices in 34 healthy young adults, sensitivity was at least 0.93 for every device and specificity ranged from 0.18 to 0.54.
Whole-night totals such as total sleep time (TST) and wake after sleep onset (WASO) are usually compared with Bland-Altman plots, which show the average bias and how far individual nights stray from it. Bias can depend on the night: actigraphy was unbiased for WASO under 30 minutes but underestimated larger amounts of wake (mean WASO 49.1 minutes per night by PSG vs. 36.8 by actigraphy). For sleep stages, results are less consistent; the 2021 consumer-device study called its stage comparisons "mixed." The Sleep Research Society has also argued for standardized "rigorous performance evaluation (previously, 'validation')" of sleep technologies.
Sources: Marino et al., Sleep 2013; Chinoy et al., Sleep 2021; de Zambotti et al., Sleep Health 2022.
What affects accuracy
- Wake during the night. The 2024 review lists wake misclassification during the sleep period as a limitation. It shows up in studies as low specificity and, for actigraphy, underestimated WASO when there is a lot of wake.
- Disrupted sleep. Consumer devices "tended to perform worse on nights with poorer/disrupted sleep."
- Sleep outside the main bout. Naps and other sleep outside the main sleep period are harder to track.
- Artifacts and wearer characteristics. The 2024 review lists artifacts and unclear performance in some groups, and notes that skin color and obesity "negatively impact sensor performance."
- What the device is not built for. The AASM review found actigraphy "is not a reliable measure of periodic limb movements."
For a study or product, that means testing in the people and conditions you will actually measure, as laid out in our sensor accuracy and validation guide. This page is not clinical advice.
Sources: Marino et al., Sleep 2013; Chinoy et al., Sleep 2021; de Zambotti et al., Sleep 2024; AASM systematic review, J Clin Sleep Med 2018.
From our research
Our principal, Mike Hittle, is a co-author of a 2019 study comparing Apple Watch sleep monitoring with a clinically validated actigraph (the Philips Actiwatch Spectrum Pro) over 27 nights from 14 healthy adults in free-living conditions. Using 15-second epochs and a movement measure (ENMO) from the watch's 50 Hz accelerometer, the abstract reports 97% accuracy, 99% sensitivity and 79% specificity for actigraphy-defined sleep and wake, and a total sleep time correlation of r = 0.85. The watch overestimated total sleep time by 6.31 minutes and underestimated WASO by 5.74 minutes, shown with Bland-Altman plots. The reference here was actigraphy, not PSG, and the authors note that "further study is needed" against polysomnography in clinical settings.
Source: Roomkham et al., F1000Research 2019 (abstract).
Related resources
Frequently asked questions
Is a sleep tracker as accurate as a sleep study?
Not fully. Lab studies comparing consumer devices with polysomnography found high sensitivity for detecting sleep (at least 0.93) but much lower specificity for detecting wake (0.18 to 0.54), and mixed results for sleep stages.
Are consumer wearables better than actigraphy for sleep?
A 2024 Sleep Research Society review found the evidence suggests consumer devices exceed traditional actigraphy in assessing sleep as defined by polysomnography, while still misclassifying some wake and performing less clearly in some groups of people.
Why is specificity so low in sleep tracking studies?
Specificity measures how often the device correctly detects wake. Validation studies find that devices often score wake during the night as sleep, so specificity stays low even when overall accuracy looks high; in one actigraphy study, specificity was 0.329 while accuracy was 0.863.
Scope it in 30 minutes
Book a scoping call and we'll name the smallest engagement that moves you forward. Most teams start with the fixed-scope Initial Audit.
Book a 30-min scoping callPrefer to write? Send a short note.