Added SEEG power after word samples increased mean recall prediction error on later lists without stimulation at C = 0.1

We tested whether SEEG power after word samples improved recall prediction on later lists without stimulation. We used 11 recordings from seven people in OpenNeuro ds005558 version 1.0.0. We assumed that each source WORD sample gave its signal anchor. We fit separate fixed decoders from earlier lists. The control used task inputs and power before words. At C = 0.1, added power increased mean later Brier error from 0.21250 to 0.22259. Error increased for six people. The mean error increase was 0.01009, with a descriptive 95% person interval from 0.00282 to 0.01785. At C = 0.01, the difference was close to zero. The package preserves code, source ranges, features, models, test predictions, figures, and reproduction checks. These results apply to the tested decoders and source sample times.

SEEG · free recall · memory decoding · logistic regression · Brier error · later-list prediction