On valid label pairs, personal adjustment gave higher mean squared sleep stage share error than group mean adjustment in 20 BOAS adults

We tested stored sleep labels from 20 adults with repeated records in OpenNeuro ds005555 version 1.1.3. Each score used all five stage shares on valid human and headband label pairs. The anchor was the smallest recording number. The target was the largest. These numbers do not show night order. Personal adjustment used the anchor error of the same person. Group adjustment used mean anchor errors from the other 19 people. Mean squared target error was 44.303 for personal adjustment and 15.679 for group adjustment. Units were squared percentage points averaged across stages. Group error minus personal error was -28.624, with a 95% person bootstrap interval of -60.995 to -2.615. All fixed checks had negative mean differences. Three check intervals included zero. Allowed missing labels gave sharp bounds of -131.565 to +37.934 for personal gain against no adjustment. This is a different comparison. The package preserves source files, code, proof, endpoint assignments, draws, and instructions. Independent calculations and a fresh source replay passed. The result concerns fixed stored outputs. It does not establish performance in new people.

sleep staging · wearable EEG · repeat records · personal calibration · stage shares · missing labels · measurement error · bootstrap