Individual Differences in Neural Decoding Are Stable but Instrument Dependent
Some participants are consistently easier to decode than others. We asked whether this difference belongs to the person or to the conditions under which the brain is measured. We compared visual decoding in EEG, MEG, intracranial EEG, and fMRI, testing classifiers on runs and stimulus identities excluded from training. Within-instrument decoding was reproducible. Attention, behavior, and peripheral signals affected recording quality but did not explain most of the stable differences between participants. Anatomy provided one explanation. Source-to-sensor distance strongly predicted modeled signal gain, and its effect was steepest in planar gradiometers. Intracranial recordings provided another: participants with sharper local category selectivity were easier to decode. This association chiefly reflected a smaller response to the nonpreferred category relative to overall visual responsivity. Cross-site prediction was uneven. EEG and MEG expressed the preferred and nonpreferred responses differently. MEG category separation and task-aligned variance described the information available to the decoder. Simultaneous EEG-MEG recordings contained a shared participant component, but external paired cohorts did not yield a common ordering across instruments. Correspondence depended on the measure, the instrument pair and, in one cohort, participant-linked stimulus sets. Repeated observations improved performance in participants with weak single-observation decoding. Stable decoding differences therefore reflect the relation between the individual, the represented distinction, and the instrument. They chiefly determine how much information a recording supplies per observation.