Read on Twitter
EEG/ERP researchers: How do you decide whether a channel is "bad" and should be interpolated? I don't think I've ever seen this discussed in a paper, and I suspect that it's usually an informal process. But I'd like to know if anyone uses a more algorithmic approach.
Thanks for the comments! We don't typically need to interpolate, but I know that interpolation is crucial with some systems / research areas. I'll add a summary of this conversation to my next book.
By the way, the impact of noise in a given channel depends on the nature of the noise and the algorithm used to quantify the data (e.g., high-frequency noise has a much bigger impact on peak amplitude than on mean amplitude). Do any of these algorithms take that into account?
Our new metric of ERP data quality, the standardized measurement error, quantifies the effect of noise on a specific measure, so it seems like it would be ideal for determining whether a channel should be interpolated. https://onlinelibrary.wiley.com/doi/10.1111/psyp.13793
It should also be useful for determining what artifact rejection parameters are ideal for a given study (with respect to the signal-to-noise ratio for the measure of interest in that study).
