What This Means
Wearable accelerometers, like those in fitness trackers and research devices, record movement many times per second. Higher recording rates use more battery and storage, so researchers want to know how low the rate can go without distorting measures of physical activity. This study took data from 321 people across 11 public datasets, treated 50 samples per second as the reference, and artificially reduced the rate to see how several common activity measures changed.
The results suggest that 20 samples per second was good enough to keep one popular measure (MIMS) within a 5% error for many everyday activities such as walking, climbing stairs, cycling, jumping, household tasks, and riding in vehicles. Running was an exception, with errors exceeding 5% at 20 Hz for sensors on the legs and trunk. At 10 samples per second, errors grew for most walking-type activities, and only about 60-72% of the signal's spectral power was kept for walking and running, although the main rhythm of movement was still detectable. Errors in MIMS did not relate to age, sex/gender, or BMI, while small sex/gender differences appeared for two other measures (ENMO and AI).
Overall, this research suggests there is no one-size-fits-all sampling rate. The right setting depends on which activity measure is used, what activities are being tracked, and where the sensor is worn. This is relevant for designing studies and comparing results across devices.