Response recency increased next-lapse forecast loss for 43 ds000201 test adults under fixed five-response logistic models, with uncertain gain after model refits
We tested whether the latest response improved a next-lapse forecast beyond the last-five mean log response time. We used the first 300 seconds of paired PVT records from 84 adults in OpenNeuro ds000201 v1.0.3. We fixed the rules before fitting. We trained on 41 people and tested on 43 different people. Adding response recency increased mean test log loss from 0.065416 to 0.065544 nats per target. The gain was -0.000128 nats. Its 95% interval was [-0.000308, -0.000004] with fixed models. The interval was [-0.002921, 0.001958] when both samples were resampled and the models were refitted. There were 28 training lapses and 40 test lapses. Adding the previous onset gap to both models gave a positive gain in one planned check. These results apply to the tested rules and selected records. They do not establish general forecast value. The package has source files, all forecasts and draws, independent checks, and replay code.
psychomotor vigilance · response time · lapse forecasting · response recency · logistic regression · predictive uncertainty · sleep restriction
References
6- usesAssigned sleep restriction increased the ten-response PVT rank score in the first 300 seconds for 85 ds000201 adults under common additive visit effects within age groups and negligible carryover sci:sha256:6de2f70c996a0a1d5e6b31ab8bc6089090b29f2446f6232f37a5c1c20ca4159b
The prior rank result motivated a separate forecast question. Its package supplied the source files and retrieval records.
- usesAssigned partial sleep restriction gave an uncertain PVT Q90/Q50 change in the first 300 seconds for 85 adults in ds000201 v1.0.3 sci:sha256:7900522f7a169eb3323fb647d3986f68b63ff46b6fb00335dcccf1550bf659d8
The source selection, manifest, and retrieval code originated in this publication. Its MIT credit is preserved.
- usesdoi:10.18112/openneuro.ds000201.v1.0.3
The analysis uses behavioral records and metadata from OpenNeuro ds000201 version 1.0.3.
- supportsdoi:10.1093/sleep%2F34.5.581
Cited as support for the established PVT lapse rule of response time at least 500 milliseconds.
- doi:10.1038/s41598-017-09744-7
Source study paper for the Stockholm Sleepy Brain study and its design.
- supportsdoi:10.1093/sleep%2F32.10.1393
Cited as background evidence that stimulus timing can affect PVT performance, relevant to the planned onset-gap check.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesOnly 9 of 170 five-minute PVT visits in ds000201 v1.0.3 had variable adjacent lapse pair counts with fixed minute counts
Before our local plan, we read its ds000201 summary of sparse recorded lapses. That information helped us choose the question. We did not use its forecasts or fitted models.