A widely used protocol for EEG classification experiments leads to a confound

Yanchen Liu, Ernest V. Pedapati, Ronnie Wilbur, Jeffrey Mark Siskind

Keywords: EEG

TL;DR: We demonstrate that the protocol used to collect about 18 datasets used in nearly 100 papers in irreparably flawed.

Abstract:

A temporal confound has been previously reported in a (now widely
used) dataset that others have tried to suggest is nonetheless
justified. Despite attempts to make the community aware of this
confound, a significant number of publications continue to use the
confounded dataset, thereby drawing unsupported conclusions. We
present a new experiment that conclusively demonstrates that the
identified confound in the dataset cannot be explained away by
recourse to factors such as block design, session duration, number of
subjects, or pooling multiple subjects. We advise caution when
designing, conducting, and interpreting the results of experiments
that use this problematic protocol.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 4802