Question regarding interpolating or removing invalid pupillometry data

dbquintela

Hi all,

I was reading Methods in cognitive pupillometry: Design, preprocessing, and statistical analysis which uses DataMatrix library to preprocess the pupillometry data, and one question I have is regarding the preprocessing step of detecting blinks and artifacts.

What is the best approach for applying those algorithms, such as blinkreconstruct()? Should these algorithms be applied to the entire pupillometry data of an experiment or to each block or trial separately, and are there any differences between these approaches?

Thank you all in advance!

sebastiaan

Hi @dbquintela ,

Thanks for your interest in the paper!

For most preprocessing steps it doesn't matter whether you apply them to the full time series that spans the entire experimental session, or to shorter 'epochs' that span individual trials. However, for the blink reconstruction/ artifact removal it indeed does, you're right about that.

If you apply blink reconstruction to individual trials (so after epoching) then you have a slightly higher risk of edge effects that can avoid blinks from being interpolated. Basically, what can happen is that the start or end of a trial coincides with a blink, in which case no interpolation is possible. That might be a reason to apply blink reconstruction to the full time series, before cutting it up into individual trials.

That being said, we always apply all processing steps, including blink reconstruction, to individual trials, simply because our eye tracker (EyeLink) doesn't record continuously, or at least not if you also want to perform drift corrections before each trial. I think that's fine too.

Hope this helps!

— Sebastiaan