Sensor placement causes outcome-dependent bias in ambulatory light-exposure estimates

Population studies use wearable sensors to investigate light as an environmental determinant of health, yet anatomical placement may misclassify exposure across outcomes. We analysed concurrent free-living measurements from one dosimeter model worn near the eyes, chest, and wrist across seven countries. Generalised additive mixed models characterised momentary errors; mixed models compared 54 daily metrics; hierarchical bootstrap estimated validation precision. Context-specific eye-level underestimation ranged from −23.1% to −4.2% at the chest and −54.7% to −30.3% at the wrist. Across 54 metrics, median absolute bias was 1% at the chest and 2% at the wrist; timing outcomes were robust, whereas level and temporal-dynamics outcomes were more placement-sensitive, with maximum absolute biases of 50% and 84%, respectively. With seven days, most metric classes reached 5% precision with 2–16 participants; temporal dynamics required 33 at the chest and 59 at the wrist. Sensor placement should match the intended outcome, with validation subsamples supporting scalable exposure epidemiology.