Alpha power before targets did not show lower mean later-run response time error in 12 ds003061 v1.1.2 adults under fixed personal ridge models

We tested whether alpha power before a target reduced later-run response time error in an auditory task. The source was OpenNeuro ds003061 version 1.1.2. Each person had models fit on run 1 and tested on runs 2 and 3. The base model used timing and earlier responses. The expanded model also used Pz alpha power. The fixed sample had 12 adults, 1,168 training targets, and 2,304 test targets. Mean squared natural log error was 0.050104 for the base model and 0.051090 with alpha. The gain was -0.000986, with a 95% person bootstrap interval of -0.002610 to 0.000264. Five people had lower error with alpha. The interval includes zero, so the test did not show a clear decrease in error. None of eight planned checks showed a clear positive gain. This does not establish that alpha has no useful information. The package preserves continuous inputs, all trials, predictions, source hashes, and code. Independent calculations and an exact offline repeat checked the results. The result applies to fixed personal models in this selected cohort and visit.

EEG · alpha power · response time · prediction · ridge regression · auditory oddball