[Eeglablist] Questions over robust reference (PREP pipeline) and pre-ICA eeglab cleaning tools

[ENRIQUE TABILO ROMERO]

Hello,
i've been using the robust reference of Nima's Prep to analyse my EEG data
on a batch (with kevin's pipeline parameters). I recently noticed that my
ROI channels has been partially deleted in several of my subjects by the
algorithm of robust reference, even with the ransac criteria disabled.
1.* Is there a way to protect my ROI from being deleted?* I know i can
exclude them from the parameter "reference channels" in the PREP, but i
don't know if that might introduce an artifactual difference to the values
of the ROI channels in difference to the non-ROI's (i'm making ERP and
time-frequency research).
 2. If i exclude the ROI from "reference channels" should i exclude it from
"evaluation channels" also?

Bonus Question:
I'm also using the epoch rejection tools from eeglab on ica components with
the objective of improving a second ICA, and while both the Makoto's and
Kevin's pipelines use algorithms based on the improbability of a channel's
signal (joint probabilty and kurtosis) and absolute threshold, when i
scroll through the data i've realized that none of the epochs that i would
visually reject are selected by those algorithms. In change i'm using
pop_rejtrend and pop_rejspec (the latter at 20-40 Htz) and rejecting about
13% of total trials.
*¿Is there a technical reason why to prefer threshold and improbability
measures for epoch rejecting with the objective of improving ICA?*