EEG and fNIRS scores separated mental tasks from rest but had small mean adjusted trial agreement in two public studies

The fixed shared test placed mean adjusted trial agreement inside the planned small-agreement bound in both studies. The fixed shared test did not confirm positive mean agreement. Both measurements separated each mental task from rest in excluded recording runs. Separate public studies supplied simultaneous EEG and optical hemoglobin data. Five people from the arithmetic study supplied development data. Fixed tests used 24 other people from that study and 26 from the word study. Each device had a separate personal linear model. Each model fit two runs and scored the third. Arithmetic AUC used first trials of paired blocks. Word AUC used all randomized trials. We removed class, history, position and drift terms before trial-score correlation. A second fixed adjustment added eye power. Both adjustments had to pass. Mean person Fisher z was the test quantity. Arithmetic: EEG/fNIRS AUC 0.739/0.735; order/eye adjusted z +0.0585/+0.0802. Word generation: EEG/fNIRS AUC 0.752/0.698; order/eye adjusted z +0.0181/-0.0128. The shared task, positive-agreement and small-agreement Holm p values were 7.8811e-08, 0.63261 and 0.019361. The bound was +/-atanh(0.2). All negative results, timing controls and individual outcomes remain. Residual reliability is unknown, so measured agreement does not identify latent neural coupling. No causal claim is made. Unequal panels, time windows, source filters, physiology and possible participant overlap limit scope. Code, allowed word features, source hashes, all results and repeat instructions are supplied.

EEG · fNIRS · mental arithmetic · word generation · trial variation · cross-modal validation · conditional correlation