A test of theta slope change gave 6.5% and 7.1% false positives in two Gaussian models of 12 SEEG patients

We tested a procedure for changes in hippocampal theta load slopes between two task sessions. We used public OpenNeuro ds004752, version 1.0.1. The sample had 12 patients with epilepsy, 44 matched SEEG contact pairs, and 983 retained trials. This was the same cohort as the cited prior analysis. The real-data test gave p = 0.30202. This result did not show stable slopes. We checked the test with fixed trial designs and Gaussian errors. Each of two null models generated 1,000 datasets with equal session slopes. The test used 999 sign draws per synthetic dataset. It gave 65 and 71 false positives, or 6.5% and 7.1%, against a nominal 5% rate. Errors were independent between trials in one model. They had correlation 0.5 within five-trial blocks in the other model. At a 0.5 dB slope change per two letters, rejection frequencies were 23.6% and 33.0%. These were nominal frequencies, not power at a corrected test size. Calibration failed in the tested models. The real session-change question remains unresolved. The archive contains source records, trial features, code, all planned results, null draws, simulations, and repeat commands. Fresh source checks and independent numerical checks passed. A fresh run reproduced all saved statistical results exactly in the same environment.

hippocampal theta · SEEG · working memory · slope change · wild score test · simulation calibration