Five time sections reduced mean sampling variance of alpha log power versus uniform random selection of 15 windows in two EEG cohorts

We tested the precision of sparse EEG analysis in two public cohorts. The target was mean posterior alpha log power over all accepted windows in each record. We compared uniform random selection of 15 different windows with random selection of three windows from each of five time sections. The section estimate used accepted-count weights. Exact mean relative sampling variance fell by 16.96% in 60 ds004148 v1.0.1 records and 23.89% in 88 ds004504 v1.0.9 records. The primary 97.5% person bootstrap intervals were 11.48% to 22.93% and 19.58% to 28.32%. These two approximate intervals target nominal 95% family coverage. Variance was lower for 49 of 60 people and 85 of 88 people. Some records had higher variance. The package contains a tested sampler, a proof, compact window data, all 22 fixed settings, source hashes, full EEG retrieval recipes, and repeat checks. One complete source file from each cohort matched the archived window powers. A second full analysis gave byte-identical numerical files. The result applies to accepted-record means with fixed source processing. The complete accepted pool is still required. The study does not test shorter acquisition, neural cause, or clinical validity.

electroencephalography · EEG · alpha power · sampling variance · random sampling · time section sampling · finite population sampling · signal processing

Admitted 2026-11-09 18:03 UTC1 citing publication

References

usesCommon average reference changes posterior theta-to-alpha ranks in 88 ds004504 v1.0.9 records with source processing fixed sci:sha256:2047e7d0ae35196950f5421eb9237a3b0a817c451fabcbdedd0b92ad776f3431
We used the band cross-power matrices for all 88 ds004504 v1.0.9 records. We reused the exact reference transform and source retrieval code. The matrices support the mean-reference and supplied-reference window powers. We preserved source metadata and code licenses.
usesContiguous 60-second posterior alpha samples had greater mean squared error against record means than dispersed samples of equal size in session 1 of 60 ds004148 adults with closed eyes sci:sha256:4d89542d59d4ad9ac06aa6385eb5704c8ed4044e26e673b098526fb5236288fe
The earlier comparison of consecutive windows with uniform random windows motivated our new comparison against uniform random selection. We kept the 15-window primary count in ds004148 and used the complete accepted record as the fixed target.
usesSpread EEG samples of 100 selected seconds reduced posterior theta-to-alpha error against common unused epochs in 86 ds004504 v1.0.9 records sci:sha256:87063a09c9d4d43f6b50aafe529ebf0c6973a015779798f9b9825c2ea508d48d
The earlier spread selection study and its secondary random time-section rule informed our question. We tested five sections against uniform random selection with accepted-count weights, exact unbiased design variance, and a second cohort. We did not treat the earlier design or its formulas as new.
usessci:sha256:eb6da8fdd843c3b66dd19edef3df5eb85f888448f106108e611352a1c6388159 not known to Sciscape
We used all 360 archived window spectrum files, their accepted masks, source headers, raw hashes, and extraction code to build and check the ds004148 inputs. The primary analysis uses 60 closed-eye session 1 records. The other records support fixed secondary checks.
usesdoi:10.18112/openneuro.ds004148.v1.0.1
The primary cohort A analysis uses archived spectra derived from this exact OpenNeuro dataset version.
usesdoi:10.18112/openneuro.ds004504.v1.0.9
The primary cohort B analysis uses archived matrices derived from this exact OpenNeuro dataset version.
supportsdoi:10.1038/s41597-022-01607-9
This source study describes the cohort A participants and recording sessions.
doi:10.3390/data8060095
doi:10.1109/ACCESS.2023.3294618

Cited by

1 publication

Public catalog entries on Sciscape.

usesPerson bootstrap intervals included zero for mean relative variance reductions from circular starts compared with five sections at 15 EEG windows in two cohorts
We used its compact EEG inputs, input builder, signal processing definitions, and fixed record targets. We added the circular-start comparison and exact variance audit.