Hi all,

I have been advised to include an overview of how much pupil size data I had to reject per participant, that was labeled as outlier (biologically invalid samples, blinks, samples with high dilation speed, etc.).

Do you have an example in mind, of a paper where they have done such a thing? or do you have a measure for doing this?

If I want to go in details, I seem to have to repeat all my preprocessing steps and somehow make a note of how many samples are being rejected or interpolated or somehow compensated for..

Hi @user-98789c 👋 . Reporting the % rejection should be enough (in published papers, you will just find the mean rejection rate across all participants). As for how to detect/measure your outlier/blink data, here are some options (but I'm sure there are many more out there that I'm not aware of).

• For blinks, you can calculate rejection rate based on your system's blink detector (e.g., how many samples were blinks & were rejected out of the timeseries pupil data of X participant). There are more ways of detecting/removing blinks. For example, you can estimate pupil responses to blinks using deconvolution and then regress out the blink-related pupil responses from your pupil timeseries. See here for an example in Python: https://github.com/tknapen/FIRDeconvolution/blob/master/test/pupil_preprocess_python.ipynb - and a similar approach in MATLAB (specifically scripts blink_interpolate.m and blink_regressout.m): https://github.com/anne-urai/2017_Urai_pupil-uncertainty/tree/master/Analysis

• For detecting pupil dilation speed outliers, you can refer to this paper (specifically Eq. 1-3) and calculate rejection rates after having detected & removed these outliers. https://link.springer.com/article/10.3758/s13428-018-1075-y

Hope this helps!